Datasets:

mlcore

/

arxiv-classifier

like 1

Follow

Kilian's Group 9

1211.2046

2

1211

Height correlation of rippled graphene and Lundeberg-Folk formula for magnetoresistance

Application of an in-plane magnetic field to rippled graphene will make the system be a plane with randomly distributed vector potentials. Massless Dirac fermions carrying charges on graphene are scattered by the vector potentials and magnetoresistance is induced proportional to the square of amplitude of in-plane magnetic field $B_{\parallel}^2$. Recently, Lundeberg and Folk proposed a formula showing dependence of the magnetoresistance on carrier density, in which the coefficient of $B_{\parallel}^2$ is given by a functional of the height-correlation function $c(r)$ of ripples. In the present paper, we give exact and explicit expressions of the coefficient for the two cases such that $c(r)$ is (i) exponential and (ii) Gaussian. The results are given using well-known special functions. Numerical fitting of our solutions to experimental data were performed. It is shown that the experimental data are well-described by the formula for the Gaussian height-correlation of ripples in the whole region of carrier density. The standard deviation $Z$ of ripple height and the correlation length $R$ of ripples are evaluated, which can be compared with direct experimental measurements.

cond-mat.mes-hall

cond-mat

Height correlation of rippled graphene and Lundeberg-Folk formula for magnetoresistance Kazuyuki Gemma and Makoto Katori ∗ 2 December 2013 Abstract Application of an in-plane magnetic field to rippled graphene will make the sys- tem be a plane with randomly distributed vector potentials. Massless Dirac fermions carrying charges on graphene are scattered by the vector potentials and magnetoresis- tance is induced proportional to the square of amplitude of in-plane magnetic field B2 k. Recently, Lundeberg and Folk proposed a formula showing dependence of the magne- toresistance on carrier density, in which the coefficient of B2 k is given by a functional of the height-correlation function c(r) of ripples. In the present paper, we give exact and explicit expressions of the coefficient for the two cases such that c(r) is (i) exponential and (ii) Gaussian. The results are given using well-known special functions. Numerical fitting of our solutions to experimental data were performed. It is shown that the ex- perimental data are well-described by the formula for the Gaussian height-correlation of ripples in the whole region of carrier density. The standard deviation Z of ripple height and the correlation length R of ripples are evaluated, which can be compared with direct experimental measurements. 1 Introduction and main results Graphene, a one-atom-thick allotrope of carbon [1, 2], has attracted much attention of theoretical physicists, since its electronic excitations are described by two-dimensional (2D) massless Dirac fermion (see, for instance, [3, 4]). Let vF be the Fermi velocity. Then, for a free fermion, the 2D massless Dirac equation is given by −ivF(cid:18) 0 ∂x + i∂y ∂x − i∂y 0 (cid:19) ψ(r) = Eψ(r), (1.1) ∗Department of Physics, Faculty of Science and Engineering, Chuo University, Kasuga, Bunkyo-ku, Tokyo 112-8551, Japan; e-mail: katori@phys.chuo-u.ac.jp 1 where i = √−1, ∂x = ∂/∂x, ∂y = ∂/∂y, and ψ(r) is a two-component wave function with the particle position r = (x, y). For a wave-number vector k = (kx, ky), the eigenvalues of energy are given by E = E± = ±vFk, k = k, where the ± signs correspond to two values of the helicity of fermion. The wave functions are determined as [3] ψ±(r; k) = hrki = where V is the volume of the system and 1 √2V eik·r(cid:18) e−iθk/2 ±eiθk/2 (cid:19) , kx(cid:19) . θk = arctan(cid:18) ky (1.2) (1.3) Although the basic description of the system is very simple as shown above, physics of graphene is extremely rich, since the 2D Dirac equation is much modified by external electric and magnetic fields applied to the graphene plane as well as by changing geometric and topological features of the 2D surface [3, 4]. As a matter of fact, an ultrathin membrane is unstable in the 3D real space and in suspended graphene and in graphene placed on a substrate nanometer-scale corrugations, which are simply called ripples [5], are observed [6, 7, 8, 9, 10, 11, 12]. Since spatial distribution of such microscopic structures on the surface cannot be controlled, even when we apply uniform electric and/or magnetic fields, the rippled graphene will behave as a 2D system with randomly distributed scalar and vector potentials on it. Effect of ripples on electronic transport on graphene is an interesting subject both from theory and experiment [3, 4, 13, 14, 15, 16]. Recently, magnetotransport measurements on rippled graphene have been reported, where a magnetic field is applied parallel to the averaged 2D plane of the graphene [17, 18, 19, 20] (see also [21] for bilayer graphene). If the surface is a perfect plane, such an in-plane mag- netic field Bk does not affect any motion of 2D electron, since electronic motion couples only to the component of magnetic field normal to the 2D plane. But on real graphene, the magnetic field becomes to include a component normal to the surface depending on the local slope around ripples. Then the system has inhomogeneous out-of-plane magnetic fields. With respect to electromagnetic property, we will be able to regard the rippled graphene with Bk as a system of traveling 2D massless Dirac fermions, which are randomly scattered by ripples [4, 17]. In the present paper, we study a formula of Lundeberg and Folk for the magnetoresistance ∆ρ, which was derived from a model based on the Boltzmann transport theory [17, 18]. The relative height of ripple on the averaged 2D plane is denoted by h(r) for each 2D position r = (x, y). We characterize the height distribution by the two-point height correlation c(r; r0) = hh(r0)h(r0 + r)i, (1.4) which is defined as a statistical average of product of the height at r0 and that at the position separated from r0 by a displacement vector r. For simplicity, we assume that the distribution of h(r) is translation invariant and isotropic in this paper. By this assumption, 2 (1.4) becomes independent of r0 and it depends only on the distance of two points r = r; c(r; r0) = c(r). As a function of wave-number k = k of fermion, we define 0 g(k) =Z ∞ W (z) =Z 2π 0 rW (rk)c(r)dr, J0(cid:18)2z sin φ 2(cid:19) sin2 φ 2 (1.5) (1.6) dφ where with the Bessel function J0(z) (see Appendix A for special functions [22]). The signed carrier-density of graphene is denoted by n = n+ − n−, which is the difference of the density of electrons n+ and that of holes n−. We assume that the temperature is low and the state is degenerated so that the wave-number of fermion k is well-approximated by the Fermi wave-number kF. With counting contributions from the two Dirac points K and K ′ and spin, k ≃ kF is related with n as (see, for instance, Sect. II.A in [3]), k = k(n) =pπn, (1.7) where the absolute value of n is used by the particle-hole symmetry; E± = ±vFk. Let Bk = Bk and θ be the angle between Bk and the direction of the current J along which the resistance ρ is measured. Set ∆ρ = ρ(Bk) − ρ(0), the Bk-induced part of resistance, which we call magnetoresistance. Then the Lundeberg-Folk formula is given as ∆ρ(n, θ, Bk) = πB2 k 2 (sin2 θ + 3 cos2 θ)g(k(n)). (1.8) Remark that this formula is not given in the published paper [17] of Lundeberg and Folk, but the corresponding equations are found in Appendix of its arXiv version [18]. Lundeberg and Folk introduced a correlation length R > 0 for ripple-height correlation c(r) and argued the asymptotic behavior of g(k) in k ≪ 1/R and k ≫ 1/R, and from the latter estimate the behavior of (1.8) in high carrier-density region was predicted as ∆ρ ∼ n−3/2,n ≫ 1. This n−3/2-law was emphasized in their paper and has been used for analyzing experimental data [17, 19, 20], while it is an asymptotic law in n ≫ 1 for a special case with the Gaussian height-correlation of ripples as clarified below in the present paper. Here we call the general equation (1.8) the Lundeberg-Folk formula. The main purpose of the present paper is to show that, for the two cases (i) (ii) c(r) = Z 2e−r/R c(r) = Z 2e−(r/R)2 (exponential) and (Gaussian) with a variance of height Z 2 = c(0) = hh(r0)2i = hh2i, exact and explicit representations for the anisotropic magnetoresistance (1.8) are obtained. (For the Gaussian correlated disorder, see [23].) The results are the following. Let F (α, β, γ; z) and F (α, γ; z) be the Gauss hyper- geometric function and the confluent hypergeometric function, respectively. (See Appendix 3 A for definitions and basic properties of special functions used in this paper.) Then, for (i) the exponential height-correlation, (1.5) is calculated as g(k) = π(ZR)2F(cid:18) 3 k2hK(2ikR) − Z 2 3 2 = , 2 , 2;−(2kR)2(cid:19) 1 + (2kR)2 E(2ikR)i, 1 (1.9) where K(z) and E(z) are the complete elliptic integrals of the first kind and of the second kind, respectively (see Appendix A). For (ii) the Gaussian height-correlation, (1.5) becomes g(k) = πZ 2R2 2 = π(ZR)2 2 2 , 2;−(kR)2(cid:19) F(cid:18) 3 hI0((kR)2/2) − I1((kR)2/2)ie−(kR)2/2, (1.10) where Iν(z) is the modified Bessel function of the first kind (see Appendix A). From these exact expressions, we can readily obtain the asymptotics of the magnetoresistance (1.8) in high carrier-density region as follows. For n ≫ 1, ∆ρ(n, θ, Bk) ≃ 1 2 1 4 (sin2 θ + 3 cos2 θ) (ZBk)2 √πR n−3/2 log(Rn1/2), for exponential (i) (sin2 θ + 3 cos2 θ) (ZBk)2 R n−3/2, for Gaussian (ii). (1.11)   The asymptotic for the Gaussian case (ii) shown in the second line in (1.11) is exactly the same as Eq.(3) reported in [17]. On the other hand, there is a logarithmic correction to the n−3/2-law in the case (i) with the exponential height-correlation of ripples. toresistance; when n ≃ 0, For the low carrier-density region, we have found the following behavior for the magne- (sin2 θ + 3 cos2 θ) (sin2 θ + 3 cos2 θ) (πZRBk)2  (πZRBk)2  1 2 1 4 (cid:18)1 − (cid:18)1 − 9 2 3 4 πR2n(cid:19), πR2n(cid:19), for exponential (i) (1.12) for Gaussian (ii). ∆ρ(n, θ, Bk) ≃ That is, ρ(n, θ, Bk) shows a cusp at the charge neutrality point n = 0 and the dependence on n around this point is linear ∝ −n. In the present paper we apply our analytic results to the experimental data in the whole region of n. The result seems to be excellent if we assume the Gaussian height-correlation of ripples. See Figures 2 and 3 in Section 4. In the vicinity of charge neutrality point, electronic transport will be affected by strong density inhomogeneity in rippled graphene, which is called electron-hole puddle [4]. Moreover, in low carrier-density region, n ≪ 1, we 4 have kR ≪ 1 and the semi-classical Boltzmann transport theory is afraid to be broken-down. To the present analysis corrections by more elaborate approximations or full quantum theory shall be added [4]. The paper is organized as follows. In Section 2 we give a derivation of the Lundeberg-Folk formula (1.8) from a semi-classical model using the Boltzmann transport theory. Section 3 is devoted to the proofs of our exact expressions (1.9) and (1.10) and their asymptotics (1.11) and (1.12). In Section 4, we discuss comparison of the present analytic results with the experimental data measured by Wakabayashi and his coworkers. Appendix A is prepared for giving the mathematical formulas for special functions used in the text. In Appendix B an explanation is given for the linear relation between resistance and inverse of relaxation time based on the Boltzmann transport theory. 2 Derivation of Lundeberg-Folk formula As remarked just below Eq.(1.8), the general expression of ∆ρ as a functional of the height- correlation function c(r) of ripples, (1.8) with (1.5) and (1.6), is found only in [18]. Although the essence of its derivation is given there, the description is very brief. For convenience, here we would like to complete the derivation of the Lundeberg-Folk formula (1.8) following [18]. We take the averaged 2D plane of graphene as the x-y plane. Let h(r) be the height of ripple at r = (x, y). 2.1 Geometric vector potential Under the in-plane magnetic field in the x-direction, Bk = (Bk, 0), (2.1) the geometric vector potential of Berry [24], A(r) = (Ax(r), Ay(r)), is calculated as [25] Ax(r) = 0, Ay(r) = −Bkh(r). (2.2) It gives a potential expressed by a 2 × 2 complex matrix for Dirac fermions with electric charge −e and the Fermi velocity vF, U(r) = −evFA(r) · σ = evFBkh(r)σy 0 (cid:19) , = evFBkh(r)(cid:18) 0 −i i (2.3) where σ = (σx, σy) are Pauli matrices. 5 2.2 Scattering matrix The scattering matrix for an interaction operator U is calculated by the Fermi golden rule as Sripple(k, k′) = = 2πV  hk′Uki2δ(E(k′) − E(k)) 2πV 2vFhk′Uki2δ(k′ − k), where E(k) = vFk. We note that hk′Uki =Z d2r′Z d2r hk′r′ihr′Urihrki with (2.4) (2.5) hr′Uri = U(r)δ(r − r′), (2.6) where the potential U(r) is given by (2.3), δ(r−r′) = δ(x−x′)δ(y−y′), and hk′ri = hr′k′i†. By using the two-component plane wave solution (1.2) for E = E+ = vFk, (2.5) becomes √2V (cid:18) e−iθk/2 eiθk/2 (cid:19) hk′Uki = Z d2r (eiθk′ /2 e−iθk′ /2)(cid:26)evFBkh(r)(cid:18) 0 −i 0 (cid:19)(cid:27) eik·r ·r i e−ik′ √2V sin evFBk θk + θk′ 2 = V with q = k′ − k and bh∗(q) bh∗(q) =Z d2r e−iq·rh(r). The square of (2.8), bh(q)2 =bh(q)bh∗(q), is then calculated as bh(q)2 = Z d2rZ d2r′ eiq·(r ′−r)h(r)h(r′) = V Z d2r eiq·r 1 V Z d2r0 h(r0)h(r0 + r), (2.7) (2.8) (2.9) involving the second integral in the second line of (2.9), (1/V )R d2r0h(r0)h(r0 + r), is a where we have renamed the integration variable as r → r0 and r′ − r → r. The quantity volume average of a product h(r0)h(r0 + r) with respect to r0 over the system. If the distribution of the ripple height h(r) on graphene is translation invariant, this volume average will converge to the height correlation function (1.4) as V → ∞, by the self-averaging property of usual random systems. Moreover, it can be written as c(r), if the system is also isotropic. Therefore, provided that V is large enough, (2.9) is equal to the Fourier transform of c(r) bc(q) =Z d2r eiq·rc(r), 6 q = q (2.10) multiplied by the volume V of the system; bh(q)2 = Vbc(q). Inserting the above results into (2.4), the volume factor V is cancelled out as desired, and we obtain the following expression of scattering matrix, Sripple(k, k′) = 2πe2vFB2 k 2 sin2 θk + θk′ 2 which was given as Eq.(A.1) in [18]. bc(q)δ(k′ − k), q = k′ − k, q = q, (2.11) By definition of θk for k = (kx, ky) given by (1.3) and its analogue θk′ for k′ = (k′ x − kx = k(cos θk′ − cos θk), qy = k′ x, k′ y), y − ky = under the energy conservation k′ = k, qx = k′ k(sin θk′ − sin θk), and thus x + q2 y q = q =qq2 = 2kp1 − cos(θk − θk′) θk − θk′ = 2k sin . 2 (2.12) (2.13) Then, when q = k′ − k with the condition k′ = k, (2.10) becomes bc(q) = Z ∞ = Z ∞ 0 0 dr rZ 2π dr rc(r)Z 2π 0 0 dϕ c(r)eiqr cos ϕ dϕ exp(cid:20)2ikr sin θk − θk′ 2 cos ϕ(cid:21) . 2.3 Boltzmann equation and scattering rates Lundeberg and Folk introduced a semi-classical model based on the Boltzmann transport theory for the system, in which the 2D electronic state with a wave-number vector k is described by probability distribution f (k). The time evolution of f (k) follows the Boltzmann equation (see, for example, [26, 27]) (cid:18) ∂ ∂t + e  ∂t (cid:21)S E · ∇k(cid:19) f (k) =(cid:20)∂f (k) , (2.14) in the external electric field E, where we have assumed the spatial homogeneity of distribu- tion f and omitted the term v · ∇rf (k) in the LHS. The 'collision term' in the RHS, which describes the scattering of particles or holes by random vector potentials caused by ripples in the present system, is given by with the operator = D[S, f ](k) (cid:20) ∂f (k) ∂t (cid:21)S D[S, f ](k) =Z d2k′ (2π)2 S(k, k′)hf (k′) − f (k)i 7 (2.15) (2.16) where S = S0 + Sripple is the total scattering matrix. Here S0 is the zero field scattering matrix assumed to be in the isotropic form, S0(k, k′) = s0(k, q)δ(k − k′), and Sripple is given by (2.11). We set S(k, k′) = eS(k, k′)δ(k − k′) and rewrite (2.16) as dθk′ eS(k, k′)hef (θk′) − ef (θk)i, where ef (θk) shows the dependence of f (k) on the angular component θk of k, when k = k is given. In order to study θk-dependence of scattering induced by Bk, which is imposed in the x-direction, we introduce the following orthogonal bases, (2π)2Z π D[S, f ](k) = (2.17) −π k satisfying The Bk-induced parts of scattering rates, which give the inverses of transport times enhanced by Bk, are obtained as sin θk√π , (2.18) α, β = x, y. (2.19) Z π −π cos θk√π , efx(θk) = efy(θk) = dθk efα(θk)efβ(θk) = δα,β, α,β(k) = −Z π −π ∆τ −1 dθk efα(θk)D[eSripple,efβ](θk) (2.20) for α, β = x, y, by the orthogonality relation (2.19) of the two bases (2.18). Now combining (2.20) and (2.17) with (2.11) and (2.13), we have multiple-integral ex- α,β such that integrals are performed with respect to θk, θk′, r and ϕ. Change pressions for ∆τ −1 the integral variables as θk + θk′ → θ+, θk − θk′ → θ−, and let θ− 2 Lα,β(k, r, θ−) = Z 4π 0 0 dϕ exp(cid:20)2ikr sin (cid:19)(cid:20)efβ(cid:18) θ+ − θ− 2 2 cos ϕ(cid:21) sin2 θ+ (cid:19) − efβ(cid:18) θ+ + θ− 2 (cid:19)(cid:21) , (2.21) 2 dθ+Z 2π ×efα(cid:18)θ+ + θ− dr rc(r)Z 2π 0 α, β = x, y. Then ∆τ −1 α,β(k) = − e2vFB2 k 4π2 kZ ∞ 0 dθ− Lα,β(k, r, θ−), α, β = x, y. (2.22) By the Boltzmann transport theory (see Appendix B for an explanation), the resistances induced by Bk are related with τ −1 α,β as ∆ρα,β(k) = 2π e2vFk ∆τ −1 α,β(k), α, β = x, y. (2.23) 8 We performed the integrals (2.21) and obtained the results Lxx(k, r, θ−) = 3Lyy(k, r, θ−) = −3πJ0(cid:18)2kr sin Lyx(k, r, θ−) = −πJ0(cid:18)2kr sin Lxy(k, r, θ−) = − 1 3 2 (cid:19) sin2 θ− 2 (cid:19) sin θ−, θ− 2 θ− , where J0(z) is the Bessel function with index ν = 0 (see Appendix A). Then (2.23) with (2.22) gives ∆ρxx(k) = 3∆ρyy(k) = 3πB2 k 2 g(k), ∆ρxy(k) = ∆ρyx(k) = 0, (2.24) (2.25) (2.26) (2.27) where g(k) is given by (1.5) with (1.6). The 'threefold relation' between ∆ρxx and ∆ρyy is found in (2.26) [28, 17, 18]. 2.4 Lundeberg-Folk formula Let E = (Ex, Ey) be the in-plane electric field applied in order to measure resistance. If the current is written as J = (Jx, Jy), then the resistance coefficients ∆ρα,β, α, β = x, y, are defined as Ey (cid:19) =(cid:18) ∆ρxx ∆ρxy (cid:18) Ex Jy (cid:19) . ∆ρyx ∆ρyy (cid:19)(cid:18) Jx (2.28) In the isotropic system, ∆ρxy = ∆ρyx = 0 as in (2.27), then Ex = ∆ρxxJx, Ey = ∆ρyyJy. We have chosen the direction of Bk in the x-direction as (2.1). Then, if we denote the angle between Bk and J as θ, Jx = J cos θ, Jy = J sin θ with J = J and the Bk-induced resistance ∆ρ for the current J is given by E · J J 2 1 J 2 (∆ρxxJ 2 x + ∆ρyyJ 2 y ) ∆ρ = = = ∆ρxx cos2 θ + ∆ρyy sin2 θ. (2.29) Then by the threefold relation (2.26) between ∆ρxx and ∆ρyy, the Lundeberg-Folk formula (1.8) is obtained. 3 Proofs of results 3.1 Proof of Eq.(1.9) Let c(r) = Z 2e−r/R and use the formula (A.3) for ν = 0. Then (1.5) with (1.6) is written as g(k) = Z 2 ∞Xn=0 (−k2)n (n!)2 Mr(n)Mφ(n) 9 (3.1) with Mr(n) = Z ∞ Mφ(n) = Z 2π 0 0 dr r2n+1e−r/R = R2n+2Γ(2n + 2), 2√πΓ(n + 3/2) dφ sin2n+2 φ 2 = Γ(n + 2) . (3.2) We apply the duplication formula (A.2) of Gamma functions and use the Pochhammer symbol (A.5), then we have g(k) = (ZR)2 (−(2kR)2)n {2Γ(n + 3/2)}2 n!Γ(n + 2) {(3/2)n}2 (2)n ∞Xn=0 ∞Xn=0 (−(2kR)2)n n! = π(ZR)2 , (3.3) which prove the first equality of (1.9) by the definition of the Gauss hypergeometric function (A.6). From the recurrence relations (A.8) and (A.9), we obtain the following two equalities, , 1 1 2 2(kR)2(cid:20)F(cid:18)1 1 + (2kR)2 F(cid:18)− 1 2 , 1;−(2kR)2(cid:19) − F(cid:18) 1 , 1;−(2kR)2(cid:19) . 1 2 1 2 2 , , 3 2 , 1;−(2kR)2(cid:19)(cid:21) , (3.4) Combining them gives , , 2 2 3 2 3 2 F(cid:18) 3 F(cid:18) 1 F(cid:18)3 2 = , 2;−(2kR)2(cid:19) = , 1;−(2kR)2(cid:19) = , 2;−(2kR)2(cid:19) 2(kR)2(cid:20)F(cid:18) 1 3 2 1 1 2 2 , , , 1;−(2kR)2(cid:19) − 1 1 + (2kR)2 F(cid:18)− 1 2 , 1 2 , 1;−(2kR)2(cid:19)(cid:21) . (3.5) By the formulas (A.14), which express the complete elliptic integrals K(z) and E(z) by using the Gauss hypergeometric functions, the second equality of (1.9) is proved. 3.2 Proof of Eq.(1.10) Let c(r) = Z 2e−(r/R)2 as where and use the formula (A.3) for ν = 0. Then (1.5) with (1.6) is written g(k) = Z 2 fMr(n) =Z ∞ 0 ∞Xn=0 (−k2)n (n!)2 fMr(n)Mφ(n), 1 2 R2n+2n! dr r2n+1e−r2/R2 = 10 (3.6) (3.7) and Mφ(n) is given as in (3.2). Following the similar procedure mentioned above, we have g(k) = π(ZR)2 2 ∞Xn=0 (3/2)n (2)n (−(kR)2)n n! , (3.8) which proves the first equality of (1.10) by the definition of the confluent hypergeometric function (A.7). From the recurrence relations (A.10) and (A.11), we obtain the following equality, F (α + 1, γ; z) = F (α, γ − 1; z) + z(α − γ + 1) γ(1 − γ) F (α + 1, γ + 1; z). (3.9) By using this equality for α = 1/2 and γ = 2, we can rewrite the first line of (1.10) as g(k) = π(ZR)2 2 (cid:20)F(cid:18)1 2 , 1;−(kR)2(cid:19) − (kR)2 4 F(cid:18)3 2 , 3;−(kR)2(cid:19)(cid:21) . (3.10) By the formula (A.15), which expresses the modified Bessel function Iν(z) by using the confluent hypergeometric functions, the second equality of (1.10) is proved. 3.3 Asymptotics in high carrier-density region Case (i): the exponential height-correlation In Gauss' integral representation (A.12), if we change the variable as u = (w−1)/(4(kR)2), we have F(cid:18) 3 2 , 3 2 1 4π(kR)3Z 1+4(kR)2 1 dw (w − 1)1/2(cid:18)1 − w − 1 4(kR)2(cid:19)−1/2 w−3/2. (3.11) Then we obtain the equality , 2;−(2kR)2(cid:19) = F(cid:18)3 π(kR)3(cid:18)4 + 3 2 = 1 2 , , 2;−(2kR)2(cid:19) By the asymptotics of the complete elliptic integrals [29] 1 + (2kR)2! − E s (2kR)2 1 + (2kR)2!# . (3.12) 1 (kR)2(cid:19)−1/2"K s (2kR)2 √1 − z2(cid:19) , K(z) ≃ log(cid:18) 4 E(z) ≃ 1, as z → 1, we have the behavior F(cid:18) 3 2 , 3 2 , 2;−(2kR)2(cid:19) ≃ 1 π(kR)2h log(4p1 + (2kR)2) − 1i, 11 as k → ∞. (3.14) (3.13) Through the relation (1.7) between k and n, the Lundeberg-Folk formula (1.8) with the present result (1.9) gives the first line of the RHS of (1.11). Case (ii): the Gaussian height-correlation In this case, we use the asymptotic expansion formula of the confluent hypergeometric function, It gives F (α, γ; z) ≃ (−z)−α Γ(γ) Γ(γ − α) Γ(γ) + Γ(α) ezzα−γ (−1)n (α)n(γ − α)n (1 − α)n(γ − α)n z−n. n! n! ∞Xn=0 ∞Xn=0 F(cid:18)3 2 , 2;−(kR)2(cid:19) ≃ 1 √π(kR)3 + O((kR)−5), z−n (3.15) (3.16) and we obtain the second line of the RHS of (1.11). This result is exactly the same as Eq.(3) reported in [17]. 3.4 Behavior in low carrier-density region Case (i): the exponential height-correlation By using the first two terms of the series (A.6) defining the Gauss hypergeometric func- tion, we obtain from (1.9) g(k) = π(ZR)2(cid:20)1 − 9 2 (kR)2 + O((kR)4)(cid:21) . (3.17) The first line of the RHS of (1.12) is concluded from (3.17). Case (ii): the Gaussian height-correlation By using the first two terms of the series (A.7) defining the confluent hypergeometric function, we obtain from (1.10) g(k) = π(ZR)2 2 (cid:20)1 − 3 4 (kR)2 + O((kR)4)(cid:21) . (3.18) It gives the second line of the RHS of (1.12). 4 Concluding remarks The Lundeberg-Folk formula (1.8) shows that the magnetoresistance induced by Bk is pro- portional to B2 k, ∆ρ(n, θ, Bk) = a(n, θ)B2 k. (4.1) 12 The coefficient a(n, θ) is given by a(n, θ) = π 2 (sin2 θ + 3 cos2 θ)g(pπn), (4.2) in which the function g depends on the height-correlation function c(r) of ripples on graphene. In the present paper, we have clarified the dependence of ∆ρ on c(r) for the two cases, (i) c(r) = Z 2e−r/R (exponential) and (ii) c(r) = Z 2e−(r/R)2 (Gaussian), where Z 2 = c(0) = hh2i is the variance of height of ripples and R indicates the correlation length of ripples. In the low carrier-density region, the so-called electron-hole puddles appear on the rippled graphene and the homogeneity assumption of system will be invalid. In order to handle the inhomogeneity of the carrier density around the Dirac point, the present Boltzmann approach should be corrected including higher-level approximations [4, 30, 31]. Away from the charge neutrality point, however, the semi-classical modeling based on the Boltzmann transport theory works well to describe the electronic transport properties of graphene. Here we consider our analytic expressions for ∆ρ as 'trial interpolation formulas' connecting two high-carrier-density regimes, n ≫ 1; the positive-charge regime (n > 0) and the negative one (n < 0). If experimental data are provided, we can evaluate the standard deviation of ripple height Our expressions for ∆ρ reported in this paper include two parameters Z and R. Z =phh2i and the correlation length of ripples R by numerical fitting of the data to our Figures 1 and 2 show results of numerical fitting of the Lundeberg-Folk formula (1.8) with our exact solutions (1.9) and (1.10) for the exponential and the Gaussian height-correlations to the experimental data, respectively, when θ = 0 in the both cases. The experimental data are given by Wakabayashi and his coworkers [19, 20]. As we can see in these figures, the fitting in the high carrier-density region is more excellent in the Gaussian case (Fig.2) than in the exponential case (Fig.1). analytic expressions (1.9) or (1.10). For the vicinity of the charge neutrality point, n ≃ 0, the fitting result for the Gaussian height-correlation is shown by Fig. 3, in which the n−3/2-law for the high carrier-density region n ≫ 1 (the second line of (1.11)) and the expression for n ≃ 0 (the second line of (1.12)) are also plotted for comparison. The obtained values of parameters by the present numerical fitting are Z = 0.376 [nm] and R = 17.0 [nm], respectively. These values should be compared with the direct experimental measurements [7, 12, 17, 19, 20]. In particular, it was reported in [20] the AFM measurements gave Z = 0.118 [nm] and R = 15.2 [nm]. See also experimental values cited in [17, 18] from [7, 12] and others. (We note that recent high- resolution measurements and analysis have revealed smaller-sized roughness of the substrate and graphene surfaces [32].) We hope that the our expressions will be used for systematic analysis of experimental data also for different values of θ. In the present paper, we have assumed translation invariance and isotropy for distribution of ripple height. Even in the high carrier-density regions, apart from the effect of electron-hole puddles, further calculation for general setting of ripple- height-correlations will be an interesting future problem. 13 
2 T
W
a 2.5 2.0 1.5 1.0 0.5 0.0 ae ae ae aeae ae ae ae ae ae ae ae aeae ae ae ae ae ae ae ae ae ae 0 1 2 3 ae ae ae ae -3 -2 -1 n
1012cm-2
Figure 1: The coefficient a(n, θ) of the Lundeberg-Folk formula (4.2) for θ = 0 with the present exact solution (1.9) for the exponential height-correlation of ripples is drawn. The parameters are chosen as Z = 0.531 [nm] and R = 8.51 [nm] by the best fitting to the experimental data given by Wakabayashi and his coworkers (indicated by dots). 14 
2 T
W
a 2.5 2.0 1.5 1.0 0.5 0.0 ae ae ae aeae ae ae ae ae ae ae ae aeae ae ae ae ae ae ae ae ae ae 0 1 2 3 ae ae ae ae -3 -2 -1 n
1012cm-2
Figure 2: The coefficient a(n, θ) of the Lundeberg-Folk formula (4.2) for θ = 0 with the present exact solution (1.10) for the Gaussian height-correlation of ripples is drawn. The parameters are chosen as Z = 0.376 [nm] and R = 17.0 [nm] by the best fitting to the experimental data given by Wakabayashi and his coworkers (indicated by dots). The curve fits the experimental data very well in the whole region of carrier-density n. 15 
2 T
W
a 2.5 2.0 1.5 1.0 0.5 0.0 ae ae aeae ae ae ae ae ae ae ae ae ae ae ae -0.3 -0.2 -0.1 0.0 0.1 n
1012cm-2
ae ae 0.2 0.3 Figure 3: Enlarged figure of Fig.2 in the vicinity of the charge neutrality point (n = 0). Dots are the experimental data given by Wakabayashi and his coworkers. The bold curve showing a cusp at n = 0 is the present analytic solution (1.10) for the Gaussian height- correlation with Z = 0.376 [nm] and R = 17.0 [nm]. In this low carrier-density region, the n−3/2-law shown by dotted curves is invalid, since it diverges as n → 0. In the very vicinity of the point n = 0, the simple expression given in the second line of the RHS of (1.12) will work well, which is shown by thin black lines. 16 Acknowledgements The present authors would like to thank Junichi Wakabayashi for providing them experimental data of magnetoresistance measurements and for very useful discussion on the present work. They also thank Tohru Kawarabayashi and Mark Lundeberg for helpful comments on the present study. This work is supported in part by the Grant-in- Aid for Scientific Research (C) (No.21540397) of Japan Society for the Promotion of Science. A Definitions and formulas of special functions Here we list the definitions and formulas of special functions used in the text (see, for instance, [22]). The gamma function Γ(z) is defined by Γ(z) =Z ∞ 0 e−uuz−1du, ℜz > 0. (A.1) When n ∈ N0 ≡ {0, 1, 2, . . .}, Γ(n + 1) = n! (we set 0! ≡ 1). The following equality is called the duplication formula, The Bessel function Jν(z) with index ν ∈ R is given by Γ(2z) = Γ(z)Γ(z + 1/2). 22z 2√π 2(cid:17)ν Jν(z) =(cid:16)z ∞Xn=0 (−1)n(z/2)2n n!Γ(ν + n + 1) for a complex number z which is not negative real. Then the modified Bessel function of the first kind is defined as (A.2) (A.3) (A.4) (A.5) (A.6) (A.7) Iν(z) = (cid:26) e−νπi/2Jν(zeπi/2) e3νπi/2Jν(ze−3πi/2) (z/2)2n = (cid:16) z 2(cid:17)ν ∞Xn=0 n!Γ(ν + n + 1) (−π < argz ≤ π/2) (π/2 < argz ≤ π) . For n ∈ N0, the Pochhammer symbol is defined as (α)n = α(α + 1)(α + 2) · · · (α + n − 1), n ∈ N ≡ {1, 2, 3, . . .}, (α)0 = 1. The Gauss hypergeometric function is defined by F (α, β, γ; z) = (α)n(β)n (γ)n zn n! , ∞Xn=0 and the confluent hypergeometric function is defined as F (α, γ; z) = lim β→∞ F (α, β, γ; z/β) (α)n (γ)n zn n! . = ∞Xn=0 17 The following recurrence relations for F (α, β, γ; z) are known, γ[F (α, β + 1, γ; z) − F (α, β, γ; z)] = αzF (α + 1, β + 1, γ + 1; z), α(1 − z)F (α + 1, β, γ; z) + [γ − 2α + (α − β)z]F (α, β, γ; z) +(α − γ)F (α − 1, β, γ; z) = 0. For F (α, β; z), zF (α + 1, γ + 1; z) = γ[F (α + 1, γ; z) − F (α, γ; z)], αF (α + 1, γ + 1; z) = (α − γ)F (α, γ + 1; z) + γF (α, γ; z) (A.8) (A.9) (A.10) (A.11) are satisfied. The Gauss hypergeometric function has the following integral expression, F (α, β, γ; z) = Γ(γ) Γ(β)Γ(γ − β)Z 1 0 du uβ−1(1 − u)γ−β−1(1 − uz)−α. (A.12) The complete elliptic integrals of the first kind, K(z), and of the second kind, E(z), are defined as respectively. They are expressed by using the Gauss hypergeometric functions as 0 dθ K(z) = Z π/2 =Z 1 p1 − z2 sin2 θ E(z) = Z π/2 0 p1 − z2 sin2 θdθ, , 1; z2(cid:19) , E(z) = F(cid:18) 1 ezF(cid:18)ν + (z/2)ν Γ(ν + 1) Iν(z) = K(z) = π 2 1 2 1 2 , 2 0 du p(1 − u2)(1 − z2u2) , , 1; z2(cid:19) . , 1 2 1 2 π 2 F(cid:18)− , 2ν + 1;−2z(cid:19) . (A.13) (A.14) (A.15) The modified Bessel function (A.4) is expressed by using the confluent hypergeometric func- tions as B Linear relation between ρ and τ−1 Let f0(k) = [e(E(k)−µ)/(kBT ) + 1]−1 be the equilibrium Fermi distribution function at tempera- ture T and chemical potential µ. The relaxation time τ (k) is phenomenologically introduced as ∂t (cid:21)S (cid:20)∂f (k) δf (k) τ (k) = − with δf (k) = f (k) − f0(k). (B.1) When we consider the long-term limit, ∂f (k)/∂t-term in the LHS is zero, and we have δf (k) = − ≃ − e  e  τ (k)E · ∇kf (k) τ (k)E · ∇kf0(k). 18 (B.2) In low temperature limit, E = vFk and E · ∇kf0(k) ≃ vFE∂f0(k)/∂E ≃ Eδ(k − kF), where kF denotes the Fermi wave-number. Then in the 2D plane the current is calculated as follows, provided J and E are parallel; J = J = Z d2k ≃ Z ∞ (2π)2 (−evF)δf (k) e dk 2π  k(−evF)(cid:16)− 0 e2vFkFτ (kF) τ (k)Eδ(k − kF)(cid:17) = 2π E. (B.3) Since resistance ρ is defined as J = ρ−1E, we have the linear relation between ρ and τ −1 as ρ = cτ −1 with the coefficient c = 2π/(e2vFkF). In Eq.(2.23) kF is written as k. References [1] Novoselov, K. S., Geim, A. K., Morozov, S. V., Jiang, D., Zhang, Y., Dubonos, S. V., Grigorieva, I. V., and Firsov, A. A. : Electric field effect in atomically thin carbon films. Science 306, 666-669 (2004) [2] Zhang, Y., Tan, Y.-W., Stormer, H. L., and Kim, P. : Experimental observation of the quantum Hall effect and Berry's phase in graphene. Nature 438, 201-204 (2005) [3] Castro Neto, A. H., Guinea, F., Peres, N. M. R., Novoselov, K. S., and Geim, A. K.: The electronic properties of graphene. Rev. Mod. Phys. 81, 109-162 (2009) [4] Das Sarma, S., Adam, S., Hwang, E. H., and Rossi, E. : Electronic transport in two- dimensional graphene. Rev. Mod. Phys. 83, 407-470 (2011) [5] Fasolino, A., Los, J. H., and Katsnelson, M. I. : Intrinsic ripples in graphene. Nature Mater. 6, 858-861 (2007) [6] Meyer, J. C., Geim, A. K., Katsnelson, M. I., Novoselov, K. S., Booth, T., and Roth, S. : The structure of suspended graphene sheets. Nature 446, 60-63 (2007) [7] Ishigami, M., Chen, J. H., Cullen, W. G., Fuhrer, M. S., and Williams, E. D. : Atomic structure of graphene on SiO2. Nano Lett. 7, 1643-1648 (2007) [8] Stolyarova, E., Rim, K. T., Ryu, S., Maultzsch, J., Kim, P., Brus, L. E., Heinz, T. F., Hybertsen, M. S., and Flynn, G. W. : High-resolution scanning tunneling microscopy imaging of mesoscopic graphene sheets on an insulating surface. Proc. Natl. Acad. Sci. U.S.A. 104, 9209-9212 (2007) [9] Stoberl, U., Wurstbauer, U., Wegscheider, W., Weiss, D., and Eroms, J. : Morphology and flexibility of graphene and few-layer graphene on various substrates. Appl. Phys. Lett. 93, 051906 (2008) 19 [10] Tikhonenko, F. V., Horsell, D. W., Gorbachev, R. V., and Savchenko, A. K. : Weak localization in graphene flakes. Phys. Rev. Lett. 100, 056802 (2008) [11] Lui, C. H., Liu, L., Mak, K. F., Flynn, G. W., and Heinz, T. F. : Ultraflat graphene. Nature 462, 339-341 (2009) [12] Geringer, V., Liebmann, M., Echtermeyer, T., Runte, S., Schmidt, M., Ruckamp, R., Lemme, M. C., and Morgenstern, M.: Intrinsic and extrinsic corrugation of monolayer graphene deposited on SiO2. Phys. Rev. Lett. 102, 076102 (2009) [13] Morozov, S. V., Novoselov, K. S., Katsnelson, M. I., Schedin, F., Ponomarenko, L. A., Jiang, D., and Geim, A. K. : Strong suppression of weak localization in graphene. Phys. Rev. Lett. 97, 016801 (2006) [14] Katsnelson, M. I., and Geim, A. K. : Electron scattering on microscopic corrugations in graphene. Phil. Trans. R. Soc. A 366, 195-204 (2008) [15] Morozov, S. V., Novoselov, K. S., Katsnelson, M. I., Schedin, F., Elias, D. C., Jaszczak, J. A., and Geim, A. K. : Giant intrinsic carrier mobilities in graphene and its bilayer. Phys. Rev. Lett. 100, 016602 (2008) [16] Deshpande, A., Bao, W., Miao, F., Lau, C. N., and LeRoy, B. J. : Spatially resolved spectroscopy of monolayer graphene on SiO2. Phys. Rev. B 79, 205411 (2009) [17] Lundeberg, M. B., and Folk, J. A. : Rippled graphene in an in-plane magnetic field: effects of a random vector potential. Phys. Rev. Lett. 105, 146804 (2010) [18] Lundeberg, M. B., and Folk, J. A. : e-print arXiv:0910.4413. [19] Wakabayashi, J., and Sano, T. : Magnetoresistance of rippled graphene in a parallel magnetic field. J. Phys: Conf. Ser.334, 012039 (2011) [20] Shiraishi, S. : Magnetoresistance of graphene in parallel magnetic fields, Master thesis, Chuo University (2012) [21] Wakabayashi, J., and Sano, K. : Magnetoresistance of bilayer graphene in parallel mag- netic fields. J. Phys, Soc, Jpn.81, 013702 (2012) [22] Andrews, G. E., Askey, R., and Roy, R. : "Special Functions" Cambridge University Press (1999) [23] Bardarson, J. H., Tworzydlo, J., Brouwer, P. W., and Beenakker, C. W. J. : One- parameter scaling at the Dirac point in graphene. Phys. Rev. Lett. 99, 106801 (2007) [24] Berry, M. V. : Quantum phase factors accompanying adiabatic changes. Proc. R. Soc. London, Ser. A 392, 45-57 (1984) 20 [25] Mathur H., and Baranger, H. U. : Random Berry phase magnetoresistance as a probe of interface roughness in Si MOSFET's. Phys. Rev. B 64, 235325 (2001) [26] Kubo, R. Toda, M., and Hashizume, N. : "Statistical Physics II, Nonequilibrium Sta- tistical Mechanics" 2nd ed. Springer (1985) [27] Le Bellac, M., Mortessagne, F., and Batrouni, G. G. : "Equilibrium and Non- Equilibrium Statistical Thermodynamics" Cambridge University Press (2004) [28] Rushforth, A. W., Gallagher, B. L., Main, P. C., Neumann, A. C., Henini, M., Marrows, C. H., and Hickey, B. J. : Anisotropic magnetoresistance in a two-dimensional electron gas in a quasirandom magnetic field. Phys. Rev. B 70, 193313 (2004) [29] Abramowitz, M., and Stegun, I. A. : "Handbook of Mathematical Functions with For- mulas, Graphs, and Mathematical Tables" Dover (1972) [30] Rossi E., and Das Sarma, S. : Ground state of graphene in the presence of random charged impurities. Phys. Rev. Lett. 101, 166803 (2008) [31] Rossi, E., Adam, S., and Das Sarma, S. : Effective medium theory for disordered two- dimensional graphene. Phys. Rev. B 79, 245423 (2009) [32] Cullen, W. G., Yamamoto, M., Burson, K. M., Chen, J. H., Jang, C., Li, L., Fuhrer, M. S., and Williams, E. D. : High-fidelity conformation of graphene to SiO2 topographic features. Phys. Rev. Lett. 105, 215504 (2010) 21 

1302.1087

1

1302

THz Generation and Detection on Dirac Fermions in Topological Insulators

This study shows that a terahertz (THz) wave can be generated from the (001) surface of cleaved Bi$_{\textrm{2}}$Se$_{\textrm{3}}$ and Cu-doped Bi$_{\textrm{2}}$Se$_{\textrm{3}}$ single crystals using 800 nm femtosecond pulses. The generated THz power is strongly dependent on the carrier concentration of the crystals. An examination of the dependence reveals the two-channel free carrier absorption to which Dirac fermions are indispensable. Dirac fermions in Bi$_{\textrm{2}}$Se$_{\textrm{3}}$ are significantly better absorbers of THz radiation than bulk carriers at room temperature. Moreover, the characteristics of THz emission confirm the existence of a recently proposed surface phonon branch that is normalized by Dirac fermions.

cond-mat.mes-hall

cond-mat

THz Generation and Detection on Dirac Fermions in Topological Insulators C. W. Luo1,∗ C. C. Lee1, H.-J. Chen1, C. M. Tu1, S. A. Ku1, W. Y. Tzeng1, T. T. Yeh1, M. C. Chiang1, H. J. Wang1, W. C. Chu1, J.-Y. Lin2,† K. H. Wu1, J. Y. Juang1, T. Kobayashi1,3, C.-M. Cheng4, C.-H. Chen4, K.-D. Tsuei4, H. Berger5, R. Sankar6, F. C. Chou6, and H. D. Yang7 1Department of Electrophysics, National Chiao Tung University, Hsinchu 300, Taiwan, R.O.C. 2Institute of Physics, National Chiao Tung University, Hsinchu 300, Taiwan, R.O.C. 3Advanced Ultrafast Laser Research Center, and Department of Engineering Science, Faculty of Informatics and Engineering, University of Electro-Communications, 1-5-1 Chofugaoka, Chofu, Tokyo 182-8585, Japan 4National Synchrotron Radiation Research Center, Hsinchu 30076, Taiwan, R.O.C. 5Institute of Physics of Complex Matter, EPFL, 1015 Lausanne, Switzerland 6Center for Condensed Matter Sciences, National Taiwan University, Taipei 106, Taiwan, R.O.C. and 7Department of Physics, National Sun Yat-Sen University, Kaohsiung 804, Taiwan, R.O.C. (Dated: July 28, 2021) This study shows that a terahertz (THz) wave can be generated from the (001) surface of cleaved Bi2Se3 and Cu-doped Bi2Se3 single crystals using 800 nm femtosecond pulses. The generated THz power is strongly dependent on the carrier concentration of the crystals. An examination of the dependence reveals the two-channel free carrier absorption to which Dirac fermions are indispensable. Dirac fermions in Bi2Se3 are significantly better absorbers of THz radiation than bulk carriers at room temperature. Moreover, the characteristics of THz emission confirm the existence of a recently proposed surface phonon branch that is normalized by Dirac fermions. PACS numbers: 71.27.+a, 78.47.D-, 78.68.+m Three-dimensional topological insulators (TIs) are characterized by a narrow band gap in the bulk and a Dirac cone-like conducting surface state [1 -- 3]. The sur- face state is a new state of quantum matter caused by the strong spin-orbit interaction and protected by time- reversal symmetry. The special properties of TIs have applications in spintronics and quantum computations. Certain TIs with a small band gap are especially use- ful for terahertz (THz) optoelectronics. One of the key issues about TIs has been the identification of the gap- less surface electronic states (Dirac fermions) and the characterizations of their fundamental properties. Angle- resolved photoemission spectroscopy (ARPES) [1, 4 -- 6] and scanning tunneling microscopy (STM) [7 -- 10] have successfully confirmed the existence of Dirac fermions in Bi1-xSex, Bi2Se3, and Bi2Te3. Regarding the trans- port measurements, a metallic channel associated with the protected surface state has been detected by either controlling the gate voltage in TI devices with a suffi- ciently low bulk carrier density or by using very thin TI films [11 -- 14]. However, these experiments used partic- ularly specialized instruments, and their procedures are excessively complex for quick and routine characteriza- tions of Dirac fermions in TIs. Hsieh et al. [15] showed an alternative approach using second harmonic generation (SHG) in arsenic- doped Bi2Se3 single crystals associated with Dirac fermions, which showed a new venue for exam- ining Dirac fermions by contact-free optical techniques. However, SHG is highly sensitive to the surface quality of samples and doping. THz waves may in principle be an ideal tool for distinguishing Dirac fermions from bulk carriers because they are not sensitive to the surface qual- ity of samples with long wavelengths. Furthermore, THz wave have a photon energy (approximately 4 meV) that is significantly lower than the bulk gap (approximately 300 meV) of TIs; thus, THz radiation would allow specific characterizations within the Dirac cone. Aguilar et al. [16] recently showed the THz responses of Dirac fermions in Bi2Se3 thin films. However, this type of THz exper- iments can only be applied to thin TIs of several tens of quintuple layers. This study shows a THz generation from pure Bi2Se3 and Cu-doped Bi2Se3 single crystals by pumping with femtosecond laser pulses. Dirac fermions were identified to have an indispensable role on the inten- sity of THz emission. Free carrier absorption is a crucial mechanism to the optoelectronic devices of TIs and was revealed from the dependence of generated THz power on the carrier concentration [17]. Moreover, the detailed characteristics of THz generation verified a newly pro- posed phonon branch normalized by Dirac fermions [18]. Single crystals of pure Bi2Se3 and Cu-doped Bi2Se3 were grown using either the Bridgeman, Melt growth, or CVT methods [19]. Single crystals of CuxBi2Se3 were ob- tained using a slow-cooling method from 850 to 650 ◦C at a rate of 2 ◦C/h and quenching in cold water. Scotch tape was used to cleave the (001) surface of the Bi2Se3 crystals to ensure a flat and bright surface for optical measurements. The carrier concentrations of the samples listed on Table I were obtained using the Hall measure- ments. The mobility was measured using the four-probe method. A reflection-type THz generation scheme was used to generate a THz wave on TIs, as shown in Fig. 1(a) and the inset of Fig. 1(b). An 800 nm Ti:sapphire laser (FemtoLasers, Inc.) beam with a repetition rate of TABLE I. Carrier concentration and the THz peak amplitude for samples grown by different methods (a Bridgman, b CVT, c Melt growth). All samples are n-type. Carrier Code Compounds Bi2Se3 Bi2Se3 Bi2Se3 Bi2Se3 a#1 b#2 a#3 c#4 a#5 Cu0.02Bi2Se3 a#6 Cu0.08Bi2Se3 c#7 Cu0.1Bi2Se3 c#8 Cu0.125Bi2Se3 (arb. units.) concentration THz peak amplitude (-1018 cm-3) 75.5±13.6 34.6±4.37 31.0±1.61 15.6±10.3 3.66±0.16 4.23±1.17 1.96±0.86 1.17±0.56 0.72±0.62 4.49±0.67 3.05±0.51 6.43±0.43 31.5±0.44 32.3±0.51 22.8±0.17 30.3±0.22 5.2 MHz and a pulse duration of 50 fs was incident at θ = 45◦ (to the surface normal) and focused on the surface of the samples with a diameter of 43 µm. The pumping fluence was tuned by varying the laser output power (the typical value for this study was 0.37 mJ/cm2). Follow- ing femtosecond pulse pumping, the generated THz wave was collected using a pair of off-axis parabolic mirrors and focused on a 1-mm-thick ZnTe crystal to allow its detection with electro-optical (EO) sampling [20]. The entire generation and detection systems were sealed in a nitrogen-filled plastic box to reduce the humidity to < 6.0%. All optical measurements were performed at room temperature. Fig. 1(b) shows the typical THz waveform generated from pure Bi2Se3 and Cu-doped Bi2Se3 single crystals with a reflection-type setup (THz radiation cannot be de- tected after TIs). The amplitude of the THz wave gen- erated from pure Bi2Se3 single crystals is significantly smaller than that from a Cu0.02Bi2Se3 single crystal. Certain Bi2Se3 crystals such as sample #1 produce nearly zero amplitude (below the S/N ratio in the detection sys- tem). The THz generation intensity is strongly depen- dent on carrier concentration and doping. In general, the THz waveform is composed of a large single pulse and a damped oscillation, which is due to the interference be- tween THz electric fields from different positions of the crystal. These electric fields are caused by the mismatch between the THz phase velocity and the group velocity of the optical pumping pulse [21]. Time-domain THz wave- forms (Fig. 1(b)) can be converted to frequency domain spectra (Fig. 2) by using fast Fourier transform (FFT). The central frequency for Cu0.02Bi2Se3 and bandwidth is approximately 1.2 THz and 1.6 THz, respectively. The intensity of pure Bi2Se3 THz spectra is relatively small, corresponding to small THz signals in the time domain. THz signals were measured at various azimuth angles ϕ along the surface normal to understand the THz genera- tion mechanism in TIs (inset of Fig. 1(b)). The data from inset of Fig. 3 clearly show that the THz peak amplitude (a) TI (b) 2 fs laser pulse Dirac-cone surface state THz pulse #1 Bi2Se3 #2 Bi2Se3 #3 Bi2Se3 #5 Cu0.02Bi2Se3 InAs x 0.01 ) . u . a ( d l e i f z H T 0 1 2 3 4 5 Delay Time (ps) FIG. 1. (color online) (a) Schematic illustration of THz gen- eration and FCA inside topological insulators. The dots in- dicate the free carriers. (b) THz waveform generated from various Bi2Se3 and Cu0.02Bi2Se3 single crystals, and an InAs wafer. Inset: schematic illustration of THz generation and the detection scheme. is virtually independent of ϕ and at odds with the opti- cal rectification simulation curve with six-fold symmetry [22]. Therefore, optical rectification is not the dominant mechanism for THz generation in TIs, and the nonlinear effect does not mainly contribute to THz generation in TIs. The band gap of 0.3 eV in Bi2Se3 is significantly smaller than the pumping photon energy of 1.55 eV. Free carriers are generated when the femtosecond laser illu- minates Bi2Se3 or Cu-doped Bi2Se3 crystals. These ex- cited carriers are located inside the bulk within 100 nm [23]. When the electric field is built inside the crystals, the excited carriers in the bulk are driven and form the currents. Two types of built-in electric fields are nor- mally present in semiconductors. The surface depletion field results from the bending of the conduction band on the semiconductor surface as in GaAs [24]. The photo- Dember field is caused by the inhomogeneous distribution of holes and electrons [25], which is usually present in narrow bandgap semiconductors, such as InAs and InSb ) . u . a ( r e w o p r e i r u o F 3.0 2.5 2.0 1.5 1.0 0.5 ) . u . a ( e d u t i l p m a k a e p z H T #5 Cu Bi Se 3 2 0.02 0.0 0.1 0.2 0.3 0.4 Fluence (mJ/cm2 ) Se 3 2 InAs x 10-4 #5 Cu Bi 0.02 #2 Bi Se 2 3 #3 Bi Se 3 2 #6 Cu Bi 0.08 Se 3 2 3 0.6 0.4 0.2 ) 1 - Å X q ( ) . u . a ( e d u t i l p m a k a e p z H T 15 10 5 0 0 Cu0.08Bi2Se3 Similation 50 100 150 200 250 300 350 (cid:73)(cid:3)(degree) PTHz STHz PTHz-STHz 1 2 3 4 Frequency (THz) 0.0 5 r e w o p r e i r u o F d e z i l a m r o N 1.2 1.0 0.8 0.6 0.4 0.2 0.0 0 0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0 4.5 5.0 Frequency (THz) FIG. 2. (color online) Fourier power spectra in the frequency domain converted from the time-domain THz signals in Fig. 1(b) using fast Fourier transform for Bi2Se3 and Cu0.02Bi2Se3 single crystals, and an InAs wafer. Inset: the THz peak am- plitude of a Cu0.02Bi2Se3 single crystal as a function of the pumping fluence. FIG. 3. (color online) Polarization-dependent Fourier power spectra of a Cu0.08Bi2Se3 single crystal. PTHz (STHz): The electric field of the p-(s-)polarized THz radiation is parallel (perpendicular) to the plane of incidence (see the inset of Fig. 1(b)). The solid squares are the surface topological phonon dispersion data taken from Ref. [18]. Inset: THz peak am- plitude for a Cu0.02Bi2Se3 single crystal as a function of the azimuth angles φ along the [001] direction, as shown in the inset of Fig. 1(b). with a bandgaps of 0.35 eV and 0.17 eV, respectively. The bandgap is approximately 0.3 eV for Bi2Se3 single crystals, which is close to that of InAs (reference sample in Fig. 1). Additionally, the intrinsic charge inhomogene- ity in the vicinity of the surface and, as a result, band- bending effects were reported [26]. Consequently, both photo-Dember and surface depletion effects are possible mechanism for THz generation in Bi2Se3 and Cu-doped Bi2Se3 single crystals. The currents formed by the ex- cited carriers are suppressed within several picoseconds, due to carrier scattering with impurities (e.g., Se vacan- cies) or with the layer boundary. The transient current further generates THz radiation by ETHz (t) ∝ ∂J(t)/∂t. Higher pumping fluences generate more free carriers, leading to larger changes of the transient current; thus , a stronger THz radiation should be generated. The THz peak amplitude increases linearly with the pumping fluences (inset of Fig. 2) [27]. The excited carriers in the bulk can diffuse either along the [001] direction or on the (001) plane to form two types of currents. In principle, the diffusion along the [001] direction is suppressed by the layer boundary or im- purity scattering to generate p-polarized THz radiation, and the diffusion on the (001) plane is suppressed mainly by impurity scattering to generate s-polarized THz ra- diation. The addition of a single wire-grid polarizer be- tween the sample and an EO detection system allows the p- and s-polarized THz radiation from TIs (e.g., from the Cu0.08Bi2Se3 single crystal) to be distinguished (Fig. 3). Intriguingly, the difference between p-polarized and s-polarized THz radiation is not only in the central fre- quency, but also in the shape of the spectra. Subtraction of the s-polarized THz spectrum from the p-polarized THz spectrum results in the gray area between 0.53 THz indicating additional absorp- and 1.72 THz (Fig. 3), tion for the s-polarized THz wave. Zhu et al. [18] re- cently measured the surface phonon dispersion in Bi2Se3 crystals using the coherent helium beam surface scatter- ing technique. They discovered a low-energy isotropic convex-dispersive surface phonon branch normalized by Dirac fermions and with a frequency range from 1.8 to 0.74 THz. This frequency range is consistent with that of the gray area in Fig. 3. Therefore, the missing power of an electromagnetic wave with an electrical field parallel to the surface manifests the additional effect due to the Dirac-fermion-normalized surface phonons. Table I shows that the carrier concentration decreases by more than one order of magnitude when Cu is doped into Bi2Se3 crystals (n-type, caused by Se va- cancies) because Cu replaces Bi in the lattice [28 -- 30]. The carrier concentrations from 15.6±10.3×1018 cm-3 to 75.5±13.6×1018 cm-3 were observed for the pure Bi2Se3 crystals, potentially because of the different growing con- ditions and methods. The THz signals from pure Bi2Se3 crystals are generally relatively small. Conversely, the Cu0.02Bi2Se3 crystal with a lower carrier concentration produces stronger THz emission than pure Bi2Se3 crys- tals. The spectral weight of the FFT spectra in Fig. 2 was plotted as a function of the carrier concentration to further quantify these results. Fig. 4 clearly shows that the THz output power (i.e., the spectral weight of the FFT spectrum) increases with the decreasing carrier concentration. The THz wave generated inside the bulk likely suffers free carrier absorption (FCA) during propagation. This effect can generally be reduced by suppressing the carrier concentration to increase the output intensity of the THz, as demonstrated in III-V semiconductors such as InAs [31]. The dependence of the THz intensity on the carrier concentration can be phenomenologically described using the Beer-Lambert Law: ITHz = I0,THze−σln (1) where I THz and I 0,THz are the transmitted THz intensi- ties outside and inside the samples, respectively, as shown in Fig. 1(a), l is the path length, n is the carrier concen- tration, and σ is the absorption cross-section. The exper- imental data in Fig. 4 can be fitted as the black dashed line by Eq. (1), and is qualitatively similar to the case of InAs [31]. However, a closer inspection reveals that the fit by Eq. (1) leads to a larger deviation from the data in the low concentration regime. Consequently, the experimental data in Fig. 4 cannot be explained solely by FCA due to the bulk carriers. It is known that Dirac fermions exist in Bi2Se3 crys- tals [1]. The Dirac band contribution to FCA on the THz emission must be considered. Fig. 1(a) shows that the THz wave can be absorbed by the electrons on the Dirac cone. The effective two-channel FCA including the con- tribution from both the bulk carriers and Dirac fermions is written as a modified Beer-Lambert equation: ITHz = I0,THze−(σb lb nb+σs ls ns ls ) (2) where σb is the absorption cross-section of bulk, l b is the path length of the bulk, n b is the carrier concentration in bulk, σs is the absorption cross-section of the surface state, l s is the path length of the surface state, and n s is the carrier concentration in the surface state. The empirical relation between n b and n s can be described by n s×10-13=0.51+2.10[1-exp(-n b×l b×10-13/20.48)] as shown in the inset of Fig. 4 [32]. The observed spec- tral weight of THz in Fig. 4 is significantly better fit by Eq. (2) (the red solid line) with l s = 2 nm (thickness of the surface state) [33] and l b = 23.5 nm (THz emit- ted from the bulk within 23.5 nm; i.e., the penetration depth of 800 nm pumping light) [34]. The acquired fitting parameters are σb =2.59×10-14 cm2 and σs =1.46×10-13 cm2, and the ratio of σs/σb = 5.64. Therefore, this study concludes that FCA caused by Dirac fermions is more ef- ficient than that caused by bulk carriers (at least at room temperature). FCA can be explained in the context of the Drude model. The FCA cross-section eµ σ = ncε0(1 + ω2τ 2) (3) where e is the electron charge, µ is the mobility, ω is the angular frequency of the THz wave, τ is the scatter- ing time of the carriers, n is the refractive index of the material, c is the seed of light, and ε0 is the permittiv- ity constant. Equation (3) indicates that the observed 4 6 8 10 12 14 (x1013 cm-2) 2 4 bxl n b ) . u a ( . m u r t c e p s T F F f o t h g e w i l a r t c e p S #8 #5 #6 #7 0.0 1.6 1.4 1.2 1.0 0.8 ) 2 - m c 3 1 0 1 x ( s n 0.6 0.4 -2 0 #2 #3 #4 #1 8.0x1019 2.0x1019 4.0x1019 6.0x1019 Carrier concentration (cm-3) FIG. 4. (color online) Spectral weight of the fast Fourier transform (FFT) spectra vs. the carrier concentration. The (1). The red black dashed line represents the fit by Eq. solid line represents the fit by Eq. the car- rier concentration n s of the Dirac fermions vs. n b of the bulk carriers. The dotted line represents the empirical fit of n s×10-13=0.51+2.10[1-exp(-n b×l b×10-13/20.48)]. Inset: (2). cross-sectional ratio is as follows: σs σb = µs µb (1 + ω2τ 2 b ) (1 + ω2τ 2 s ) (4) where s denotes the physical properties of the Dirac fermions of the surface state and b denotes the physi- cal properties of the bulk carriers. In this study, µb ≈ 1500 cm2/V s at room temperature for the typical Bi2Se3 single crystals. The mobility can be expressed as µb = eτ /m, where m is the effective mass of the carriers. With the effective mass of the bulk carrier mb=0.14m0 [35, 36], where m0 is the electron bare mass, τb ≈0.12 ps at room temperature. According to Ref. 36 and the references therein, τs ≪ τb at low temperatures (low T ); so is it as- sumed at room temperature. The above analysis results in ωτb ≈ 1.2 and ωτs ≪ 1. Usually, µs is smaller than µb at low T in Bi2Se3 single crystals of approximately 100 µm in thickness [11, 35, 36]. However, the experimen- tal data for µs at room temperature have been relatively rare. Equation (4) indicates that the ratio σs/σb = 5.64 may lead to comparable values for µs and µb at room temperature. Generally, the effective mass varies weakly with T. Conversely, τb decreases rapidly with increasing T because of electron-phonon scattering. The scattering of Dirac fermions may be more dominated by the im- purity scattering than that of the bulk carriers; thus, τs is less temperature dependent and does not decrease as rapidly with increasing T compared to τb. Therefore, the values of µs and µb may be comparable at room temper- ature. These results strongly suggest that a THz wave generated inside the Bi2Se3 crystals can be used not only to identify but also to effectively examine fundamental properties of Dirac fermions. In summary, THz radiation can be generated from Bi2Se3 and Cu-doped Bi2Se3 single crystals. Dirac fermions of the surface state are indispensable to explain- ing the strong dependence of the THz emission power on the carrier concentration. Furthermore, the detailed characteristics of the THz emission confirmed the pres- ence of a Dirac-fermion-normalized phonon branch and valuable information regarding the fundamental proper- ties of Dirac fermions. This work was support by the National Science Coun- cil of Taiwan, under grant: Nos. NSC101-2112-M-009- 016-MY2, NSC101-2112-M-009-017-MY2 and NSC 100- 2112-M110-004-MY3, and by the MOEATU program at NCTU of Taiwan, R.O.C. Technical help from C. K. Wen and discussions with H. T. Jeng and T. M. Uen are ap- preciated. ∗ cwluo@mail.nctu.edu.tw † ago@nctu.edu.tw [1] D. Hsieh, D. Qian, L. Wray, Y. Xia, Y. S. Hor, R. J. Cava, and M. Z. Hasan, Nature (London) 452, 970 (2008). [2] M. Z. Hasan and C. L. Kane, Rev. Mod. Phys. 82, 3045 (2010). [3] X.-L. Qi and S.-C. Zhang, Rev. Mod. Phys. 83, 1057 5 [13] D. H. Kim, S. J. Cho, N. P. Butch, P. Syers, K. Kir- shenbaum, S. Adam, J. Paglione, and M. S. Fuhrer, Nat. Phys. 8, 459 (2012). [14] A. A. Taskin, S. Sasaki, K. Segawa, and Y. Ando, Phys. Rev. Lett. 109, 066803 (2012). [15] D. Hsieh, J.W. McIver, D. H. Torchinsky, D. R. Gardner, Y. S. Lee, and N. Gedik, Phys. Rev. Lett. 106, 057401 (2011). [16] R. Vald´es Aguilar, A. V. Stier, W. Liu, L. S. Bilbro, D. K. George, N. Bansal, L. Wu, J. Cerne, A. G. Markelz, S. Oh, and N. P. Armitage, Phys. Rev. Lett. 108, 087403 (2012). [17] R. D. Kekatpure and M. L. Brongersma, Nano Lett. 8, 3787 (2008). [18] Xuetao Zhu, L. Santos, R. Sankar, S. Chikara, C. Howard, F. C. Chou, C. Chamon, and M. El-Batanouny, Phys. Rev. Lett. 107, 186102 (2011). [19] F.-T. Huang, M.-W. Chu, H. H. Kung, W. L. Lee, R. Sankar, S.-C. Liou, K. K. Wu, Y. K. Kuo, and F. C. Chou, Phys. Rev. B 86, 081104(R) (2012). [20] Q. Wu and X.-C. Zhang, Appl. Phys. Lett. 71, 1285 (1997). [21] Nick C. J. van der Valk, Paul C. M. Planken, Anton N. Buijserd and Huib J. Bakker, J. Opt. Soc. Am. B 22, 1714 (2005). [22] W.-C. Chu, S. A. Ku, H. J. Wang, C. W. Luo, Yu. M. Andreev, G. Lanskii, and T. Kobayashi, Opt. Lett. 37, 945 (2012). [23] N. Kumar, B. A. Ruzicka, N. P. Butch, P. Syers, K. Kir- shenbaum, J. Paglione, and H. Zhao, Phys. Rev. B 83, 235306 (2011). [24] X.-C. Zhang and D. H. Auston, J. Appl. Phys. 71, 326 (2011). (1992). [4] Y. Xia, D. Qian, D. Hsieh, L.Wray, A. Pal, H. Lin, A. Bansil, D. Grauer, Y. S. Hor, R. J. Cava and M. Z. Hasan, Nat. Phys. 5, 398 (2009). [5] D. Hsieh, Y. Xia, D. Qian, L. Wray, J. H. Dil, F. Meier, J. Osterwalder, L. Patthey, J. G. Checkelsky, N. P. Ong, A. V. Fedorov, H. Lin, A. Bansil, D. Grauer, Y. S. Hor, R. J. Cava and M. Z. Hasan, Nature (London) 460, 1101 (2009). [6] Y. L. Chen, J. G. Analytis, J.-H. Chu, Z. K. Liu, S.-K. Mo, X. L. Qi, H. J. Zhang, D. H. Lu, X. Dai, Z. Fang, S. C. Zhang, I. R. Fisher, Z. Hussain, and Z.-X. Shen, Science 325, 178 (2009). [7] Tong Zhang, Peng Cheng, Xi Chen, Jin-Feng Jia, Xucun Ma, Ke He, Lili Wang, Haijun Zhang, Xi Dai, Zhong Fang, Xincheng Xie, and Qi-Kun Xue, Phys. Rev. Lett. 103, 266803 (2009). [8] P. Roushan, J. Seo, C. V. Parker, Y. S. Hor, D. Hsieh, D. Qian, A. Richardella, M. Z. Hasan, R. J. Cava, and A. Yazdani, Nature (London) 460, 1106 (2009). [9] Zhanybek Alpichshev, J. G. Analytis, J.-H. Chu, I. R. Fisher, Y. L. Chen, Z. X. Shen, A. Fang, and A. Kapit- ulnik, Phys. Rev. Lett. 104, 016401 (2010). [10] Sunghun Kim, M. Ye, K. Kuroda, Y. Yamada, E. E. Krasovskii, E. V. Chulkov, K. Miyamoto, M. Nakatake, T. Okuda, Y. Ueda, K. Shimada, H. Namatame, M. Taniguchi, and A. Kimura, Phys. Rev. Lett. 107, 056803 (2011). [11] J. G. Checkelsky, Y. S. Hor, R. J. Cava, and N. P. Ong, Phys. Rev. Lett. 106, 196801 (2011). [12] H. Steinburg, D. R. Gardner, Y. S. Lee, and P. Jarillo- Herrero, Nano Lett. 10, 5032 (2010). [25] P. Gu, M. Tani, S. Kono, K. Sakai, and X.-C. Zhang, J. Appl. Phys. 91, 5533 (2002). [26] Dimitrios Galanakis and Tudor D. Stanescu, Phys. Rev. B 86, 195311 (2012). [27] R. Mendis, M. L. Smith, L. J. Bignell, R. E. Vickers, and R. A. Lewis, J. Appl. Phys. 98, 126104 (2005). [28] A. Vasko, L. Tich´y, J. Hor´ak, and J. Weissenstein, Appl. Phys. 5, 217 (1974). [29] Y. S. Hor, A. J. Williams, J. G. Checkelsky, P. Roushan, J. Seo, Q. Xu, H.W. Zandbergen, A. Yazdani, N. P. Ong, and R. J. Cava, Phys. Rev. Lett. 104, 057001 (2010). [30] L. A. Wray, S.-Y. Xu, Y. Xia, Y. S. Hor, D. Qian, A. V. Fedorov, H. Lin, A. Bansil, R. J. Cava and M. Z. Hasan, Nat. Phys. 6, 855 (2010). [31] K. Liu, Jingzhou Xu, Tao Yuan, and X.-C. Zhang, Phys. Rev. B 73, 155330 (2006). [32] The carrier concentration ns of Dirac fermions was deter- mined by the angle-resolved photoemission spectroscopy (ARPES) in NSRRC, Taiwan. [33] W. Zhang, R. Yu, H.-J. Zhang, X. Dai, and Z. Fang, New J. Phys. 12, 065013 (2010). [34] The 800 nm penetration depth was determined by the incident angle-dependent reflectivity measurements (with s- and p-polarizations) and the fitting of complex Fresnel equation. [35] J. G. Analytis, R. D. McDonald, S. C. Riggs, J.-H. Chu, G. S. Boebinger, and I. R. Fisher, Nat. Phys. 6, 960 (2010). [36] N. P. Butch, K. Kirshenbaum, P. Syers, A. B. Sushkov, G. S. Jenkins, H. D. Drew, and J. Paglione, Phys. Rev. B 81, 241301(R) (2010). 

1503.04076

2

1503

Electron waiting times in coherent conductors are correlated

We evaluate the joint distributions of electron waiting times in coherent conductors described by scattering theory. Successive electron waiting times in a single-channel conductor are found to be correlated due to the fermionic statistics encoded in the many-body state. Our formalism allows us also to investigate the waiting times between charge transfer events in different outgoing channels. As an application we consider a quantum point contact in a chiral setup with one or both input channels biased by either a static or a time-dependent periodic voltage described by Floquet theory. The theoretical framework developed here can be applied to a variety of scattering problems and can in a straightforward manner be extended to joint distributions of several electron waiting times.

cond-mat.mes-hall

cond-mat

a Electron waiting times in coherent conductors are correlated David Dasenbrook,1, ∗ Patrick P. Hofer,1, ∗ and Christian Flindt1, 2 1D´epartement de Physique Th´eorique, Universit´e de Gen`eve, 1211 Gen`eve, Switzerland 2Department of Applied Physics, Aalto University, 00076 Aalto, Finland (Dated: September 3, 2018) We evaluate the joint distributions of electron waiting times in coherent conductors described by scattering theory. Successive electron waiting times in a single-channel conductor are found to be correlated due to the fermionic statistics encoded in the many-body state. Our formalism allows us also to investigate the waiting times between charge transfer events in different outgoing channels. As an application we consider a quantum point contact in a chiral setup with one or both input channels biased by either a static or a time-dependent periodic voltage described by Floquet theory. The theoretical framework developed here can be applied to a variety of scattering problems and can in a straightforward manner be extended to joint distributions of several electron waiting times. PACS numbers: 72.70.+m, 73.23.-b, 73.63.-b I. INTRODUCTION Investigations of current fluctuations are useful to understand the quantum transport in small electronic conductors.1 In the mesoscopic regime, the transport is coherent and the measurement of shot noise provides in- formation about the effective charge and the quantum statistics of the involved quasiparticles.2 -- 5 A more gen- eral characterization of the transport is given by the full counting statistics (FCS) of transferred charge.6 -- 9 At long times, not only the noise, but all zero-frequency cur- rent correlators can be obtained from the FCS. Besides characterizing the elementary charge transfer events,10 -- 15 it has been realized that FCS is also intimately linked to entanglement in quantum many-body systems16 -- 20 and to fluctuation theorems at the nano-scale.21 -- 23 A complementary picture of the charge transport is provided by the electronic waiting time distribution (WTD). The WTD is the probability density for a wait- ing time of duration τ to occur between two succes- sive charges transmitted through a conductor. In recent years, WTDs have been considered for various electronic systems, including transport through quantum dots gov- erned by either Markovian24 -- 28 or non-Markovian29 mas- ter equations, and coherent conductors described by scat- tering theory30 -- 32 or tight-binding models.33 For period- ically driven single-electron sources,34 -- 36 the WTD pro- vides a useful characterization of the regularity of the emitter and it encodes information about the shape of the emitted wave functions which is not readily accessi- ble in the FCS.31,37 Experimentally, progress has been made towards the detection of individual electrons emit- ted above the Fermi sea,38,39 possibly paving the way for measurements of quantum transport on short time scales. The interest in WTDs has so far been focused on the distribution of individual electron waiting times. Such distributions, however, do not address the question of correlations between subsequent waiting times. One may ask if the observation of one waiting time will affect the following waiting time. Some results indicate that sub- sequent electron waiting times indeed are correlated.30,31 FIG. 1. (Color online) Joint WTD for a fully open conduction channel. The two side panels show the joint WTD along the cuts indicated with colored dashed lines in the main panel. The joint WTD is well captured by the generalized Wigner- Dyson distribution (dashed lines) in Eq. (3). The dotted lines in the side panels show the waiting time distributions if there were no correlations between subsequent waiting times. Exact results have been obtained using Eq. (51). However, to fully answer this question, a theory of joint WTDs is needed. Figure 1 shows an example of a joint WTD. Using a method that we develop in this paper, we have calculated the joint distribution of waiting times W(τ1, τ2) between electrons transmitted through a fully open conduction channel. In Refs. 30 and 32 it was shown that the distri- bution of individual electron waiting times in this case is W(τ1,τ2)0.00.51.01.52.0τ1/¯τ0.00.51.01.52.0τ2/¯τwell-approximated by the Wigner-Dyson distribution WWD(τ ) = 32τ 2 π2 ¯τ 3 e−4τ 2/π ¯τ 2 , with the mean waiting time ¯τ = h eV , (1) (2) determined by the applied voltage V . If subsequent wait- ing times are uncorrelated, the joint WTD in Fig. 1 should factorize as W(τ1, τ2) = WWD(τ1)WWD(τ2). Such a factorization is often referred to as a renewal property.40 However, as we find, the joint WTD in Fig. 1 cannot be written in this simple form. This demonstrates that subsequent electron waiting times are correlated. In fact, based on the analogy between WTDs and level spac- ing statistics exploited in Refs. 30 and 32, the joint WTD is expected to take the form41 WWD(τ1, τ2) = 4b4 √ π 3¯τ 6 τ 2 1 τ 2 2 (τ1 + τ2)2e − 2b 3¯τ 2 (τ 2 1 +τ 2 2 +τ1τ2), (3) with b = 729/(128π). Figure 1 shows that our results are well-approximated by this generalized Wigner-Dyson distribution, which cannot be factorized into products of the WTD in Eq. (1). The purpose of this work is to present our method for calculating joint WTDs. The method uses a relation be- tween WTDs and the idle time probability (ITP).29 -- 31 The ITP is the probability of detecting no electrons in a time interval of length τ . For stationary processes, the WTD can be expressed as the second derivative of the ITP with respect to τ . As a second-quantized formu- lation shows, each derivative corresponds to the detec- tion of an electron. Building on this principle, we obtain WTDs for detections in different channels and joint dis- tributions of successive electron waiting times. To illustrate our formalism, we consider a chiral setup where two incoming conduction channels are partitioned on a quantum point contact (QPC); see Fig. 2. One or both inputs may be voltage biased, either by a static voltage or with a series of Lorentzian-shaped voltage pulses7,42,43 as recently realized experimentally.35,36 We calculate the joint WTD and discuss correlations between electron waiting times together with WTDs for detec- tions in different channels. Our findings generalize ear- lier results for single30,31 and multi-channel32 conductors and provide insights into the correlations between subse- quent electron waiting times in phase-coherent conduc- tors. We focus here on electronic conductors, but our concepts may as well be realized in cold fermionic quan- tum gases.44 The article is organized as follows. In Sec. II we discuss the theory of WTDs in electronic conductors and explain the concept of a renewal process. We calculate the elec- tronic WTD for the setup in Fig. 2. In Sec. III we in- troduce a generalized ITP which enables us to formulate 2 FIG. 2. Chiral setup with electrons in two incoming channels being partitioned on a QPC. A (time-dependent) voltage can be applied to the contacts of the incoming channels. The QPC has transmission T and reflection R. The transmitted and reflected electrons are detected in the outgoing channels at different positions x = ts/e A/B with vF = 1. a theory of joint WTDs. In Sec. IV we discuss our re- sults for the joint WTDs with either individual electrons being partitioned on the QPC or electrons from different inputs interfering on the QPC in an electronic Hong-Ou- Mandel experiment. Finally, in Sec. V we present our conclusions. Technical details of our calculations are de- ferred to two appendixes at the end. II. ELECTRON WAITING TIMES In this section, we discuss the distribution of waiting times between successive electrons emitted into a single conduction channel. The WTD is the conditional prob- ability density of detecting a particle at a time te given that the last detection occurred at the earlier time ts. In general, the WTD is a two-time quantity depending on both ts and te. We call a system stationary if the WTD depends only on the difference of these times τ ≡ te − ts, and we write it as W(τ ). We denote time intervals by τ 's, and absolute times by t's. We first define the ITP. As we will see, the ITP plays a somewhat similar role to that of a generating func- tion: By differentiating the ITP with respect to its time arguments, we obtain the distributions of interest, for instance the WTD. This principle becomes particularly transparent when using a second-quantized formulation of the ITP. We start by deriving the first passage time distribution from the ITP before evaluating the WTD. We illustrate our formalism by calculating the WTD for the setup depicted in Fig. 2 using either a static or a periodically modulated voltage. Finally, to gain a better understanding of correlated waiting times, we discuss the concept of a renewal process. The approach developed in this section is important as it allows us to generalize our theory to multiple conduction channels and joint WTDs later in this work. A. Idle time probability The ITP Π(ts, te) is the probability of observing no transmitted electrons in the time interval [ts, te] at a point x0 after the scatterer. To evaluate the ITP we employ second-quantization. Working close to the Fermi level, we may linearize the dispersion relation as 3 expectation value at t = 0 and treat time-dependent sys- tems by shifting the integration interval in the definition of (cid:98)Q in Eq. (5). In general, the ITP is a two-time quantity depending both on ts and te. However, for stationary processes it depends only on the time difference τ = te − ts such that Π(ts, te) = Π(te − ts) = Π(τ ), (8) just as for the WTD.
k = vF k. (4) B. First passage time distribution All electrons above the Fermi level thus propagate with the Fermi velocity vF along the chiral conduction chan- nels. We may then evaluate the ITP by considering the operator that counts the number of transmitted parti- cles in the spatial interval x ∈ [vF ts, vF te]. Here x is the distance to the point x0 and the x-axis is oriented oppositely to the direction of propagation, see Fig. 2. In second-quantization this operator reads31 b†(x)b(x)dx, (5) (cid:98)Q = (cid:90) vF te vF ts where b†(x) and b(x) are the creation and annihilation operators of electrons at position x. To keep the notation simple, we have omitted the explicit time arguments of (cid:98)Q. We will later on make use of the relations To appreciate the usefulness of the ITP, we first show how one can obtain the first passage time distribution F(ts, te) from the ITP. The first passage time distribution is the probability density for the first detection of a parti- cle to occur at the time te, given that observations were started at the earlier time ts. Such a distribution has recently been evaluated for electron transport through a quantum dot, with ts being the time at which the exter- nal electronic reservoirs are connected to the quantum dot.49 To obtain the first passage time distribution from the ITP, we notice that the ITP may be expressed as Π(ts, te) = 1 − F(ts, t)dt. (9) (cid:90) te ts ∂ts(cid:98)Q = −b†(ts)b(ts), ∂te(cid:98)Q = b†(te)b(te), Here, the integral equals the probability that at least one electron is observed in the interval [ts, te]. Now, by differentiating the ITP with respect to te, we obtain F(ts, te) = −∂te Π(ts, te). (10) (6) setting vF = 1 throughout the rest of the paper. the normal-ordered exponential of −(cid:98)Q,7,31,45,46 The ITP can be expressed as the expectation value of Π(ts, te) = , (7) (cid:68) (cid:69) : e−(cid:98)Q : where the expectation value is taken with respect to the many-body state at t = 0. This state is a non-equilibrium state consisting of a Slater determinant of single-particle scattering states in the reservoirs.47 To obtain a unidi- rectional process, we limit our analysis to particles in the transport window above the Fermi energy. Depending on the specific measurement scheme, the presence of the Fermi sea may influence the ITP. This is a subject of ongoing investigations. In this work, the notation : ··· : denotes the normal-ordering of operators with respect to the Fermi sea,48 i. e., all operators that create elec- trons above or holes below the Fermi level are moved to the left side of the operators that annihilate such exci- tations, with every individual permutation contributing a minus sign. We note that any time evolution of the many-body state from t = 0 to t = to can be absorbed into a shift of the spatial interval (vF = 1) from [ts, te] to [ts +to, te +to]. With this in mind, we always evaluate the Furthermore, using Eqs. (6) and (7) we find (cid:68)b†(te) : e−(cid:98)Q : b(te) (cid:69) F(ts, te) = . (11) Equation (11) is very similar to Eq. (7), however, the expectation value of the normal-ordered exponential is taken with respect to the many-body state with a par- ticle removed at time te. This removal constitutes the detection event in the definition of the first passage time distribution. A related quantity is the probability density F(ts, te) for detecting a particle at ts with no subsequent detec- tions until the time te. Following the line of arguments above, we readily find F(ts, te) = ∂tsΠ(ts, te) (cid:68)b†(ts) : e−(cid:98)Q : b(ts) (cid:69) (12) . (13) and F(ts, te) = We see that the expectation value of the normal-ordered exponential is now taken with respect to the many-body state with a particle removed at the initial time ts. For stationary problems, the first passage time only depends on the time difference τ ≡ te − ts and we have F(τ ) = F(τ ) = −∂τ Π(τ ), (14) and for the WTD W(τ ) = (cid:90) T 0 (cid:104)τ(cid:105) T I(ts)W(ts, ts + τ )dts. (20) 4 as a direct consequence of Eq. (8). The derivations of Eqs. (10,11,12,13) illustrate an im- portant principle that we will use in the following: By dif- ferentiating the ITP with respect to its time arguments, pairs of operators are pulled down from the exponential function corresponding to detection events at the begin- ning or at the end of the interval [ts, te]. C. Waiting time distribution We are now in a position to derive the WTD from the ITP. We recall that the WTD is the conditional proba- bility density of detecting a particle at a time te given that the last detection occurred at the earlier time ts. The joint probability density of detecting a particle both at ts and te with no detection events in between is equal to the WTD multiplied by the probability density of a detection at ts. The joint probability density can be ob- tained by differentiating the ITP with respect to both the initial time ts and the final time te. Moreover, for uni-directional charge transport, the probability density of a detection at ts is simply the average particle current I(ts) at time ts. We then find I(ts)W(ts, te) = −∂te ∂tsΠ(ts, te), (15) where the minus sign comes together with the partial derivative with respect to te, c.f. Eq. (10). Using Eq. (7) we now arrive at the second quantized expression I(ts)W(ts, te) = (cid:68)b†(te)b†(ts) : e−(cid:98)Q : b(ts)b(te) (cid:69) (16) . For stationary processes, the average particle current equals the inverse mean waiting time, I(ts) = 1/(cid:104)τ(cid:105). Combining Eqs. (8) and (15), we then have W(τ ) = (cid:104)τ(cid:105)∂2 τ Π(τ ), (17) as previously found in Refs. 30 and 32. For conductors driven with period T , we define a one-time ITP by aver- aging over all possible starting times ts,31,37 keeping the interval τ ≡ te − ts fixed, Π(τ ) = 1 T Π(ts, ts + τ )dts. (18) This yields the relevant ITP if the observation starts at a random time. Employing Eqs. (14, 17), we obtain one- time distributions which are independent of the detection time of the first electron. For the first passage time dis- tribution, we find F(τ ) = = 1 T 1 T F(ts, ts + τ )dts F(ts, ts + τ )dts, (cid:90) T 0 (cid:90) T (cid:90) T 0 0 D. Applications To illustrate our formalism, we consider the setup in Fig. 2. Two incoming channels are partitioned on a QPC into outgoing channels labeled as α = A, B. The elec- trons are non-interacting and the temperature is zero. The QPC is described by the scattering matrix (cid:18) r (cid:19) S = d −d∗ r∗ , (21) which relates the operators bα(E) for electrons in the outgoing channels to the operators ai(E) for incoming electrons originating from contact i = 1, 2. The trans- mission and reflection amplitudes fulfill d2 + r2 = T + R = 1 (22) as a consequence of current conservation. The incoming channels are independently biased with a voltage Vi(t) which is either constant or periodic in time. For now, we keep contact 2 grounded and we focus on the WTDs in one of the outgoing conduction channels. The general- ization to multiple channels and joint WTDs follows in the next section. The ITP in the outgoing channel α is given by Eq. (7) upon attaching an index α to all operators Πα(ts, te) = Using the scattering matrix, we may express the (cid:98)Qα's in (23) terms of the operators for the incoming channels. Specif- ically, we have (cid:68) (cid:69) : e−(cid:98)Qα : . te(cid:90) (cid:98)QA = R ts and † a 1(x)a1(x)dx ≡ R(cid:98)Q1, (cid:98)QB = T(cid:98)Q1. (24) (25) The particles in the incoming channels are in thermal equilibrium with the reservoirs. The scattering matrix thus allows us to map the evalution of an expectation value in a non-equilibrium state onto the evaluation of an equilibrium average. Note that the second contact is irrelevant since we are only concerned with the transport above the Fermi level. The single-channel ITP can be evaluated by means of (19) the determinant formula30, Πα(ts, te) = det(I − Qα), (26) 5 FIG. 3. (Color online) Idle time probability (ITP), first passage time distribution, and waiting time distribution (WTD). (a) Results for a static voltage applied to contact 1, V1(t) = V , and no voltage applied to contact 2, V2(t) = 0. The QPC is tuned to T = 0.3. We consider the electrons that are transmitted into outgoing channel B. The time is given in units of ¯τ = h/(eV ). (b) Results for a series of Lorentzian-shaped voltage pulses of unit charge applied to contact 1, see Eq. (27). The width of the pulses is Γ = 0.02T . The time is given in units of the period of the pulses T . where Qα contains the single-particle matrix elements of the operator (cid:98)Qα and I is the identity matrix. General expressions for a conductor with n channels are given in Appendix A. We now compare the WTD for a static voltage and that of periodic voltage pulses applied to contact 1. With a constant voltage, the ITP is a function of the time differ- ence only and we can directly employ Eq. (17) to obtain the WTD. For the time-dependent case, we consider a series of Lorentzian-shaped voltage pulses motivated by recent experiments.35,36 For pulses of unit charge, the voltage reads (see also Appendix B) 2Γ ∞(cid:88) eV (t) = (t − jT )2 + Γ2 , j=−∞ (27) where Γ parameterizes the width of the pulses. Each pulse creates a clean single-electron excitation above the Fermi level without accompanying electron-hole pairs. Such excitations were first proposed theoretically by Lev- itov and co-workers7,42,43 and later named levitons fol- lowing their experimental realization.35,36 Importantly, a time-dependent voltage has the effect of adding a phase to the single-particle states in the contact.50 We can treat this additional phase as a scattering phase which is picked up after the particles leave the contact but before they arrive at the scatterer, i. e. −i e(cid:82) t t0 SVi(t) = e dt(cid:48)Vi(t(cid:48)). (28) Here t0 is the time when the voltage is switched on. In this way, the contacts can be treated as equilibrium reser- voirs at the same chemical potential and all the effects due to the time-dependent driving are captured by the phase Eq. (28). In the energy basis, the effect of the pe- riodic voltage is that particles can absorb or emit energy quanta of size Ω, where Ω = 2π/T is the frequency of the driving. The Floquet scattering matrix51 encodes the quantum mechanical amplitudes for these processes and relates operators before and after the scattering phase as ai(E) = Sna(cid:48) i(E−n). (29) ∞(cid:88) n=−∞ The operator a(cid:48) i(E) annihilates an equilibrium particle in contact i, while ai(E) annihilates a particle incident on the scatterer. We have also defined the energies En = E + nΩ. (30) The Sn's are obtained by Fourier transforming Eq. (28). Based on Eq. (21), the ai(E) operators can be related to the bα(E) operators and the ITP can be calculated analogously to the static case using Eq. (26) (see also Appendix A). Finally, to evaluate the WTD, we employ Eqs. (17, 18). Figure 3 shows the ITP, the first passage time distribu- tion, and the WTD for the static voltage as well as for the train of voltage pulses. Interestingly, the two ITPs are nearly indistinguishable despite the very different volt- ages applied to contact 1. In both cases, the ITP decays monotonously with time from Π(0) = 1 at τ = 0. Turn- ing next to the first passage time distribution, some struc- ture starts to be visible and the two voltage cases can now be distinguished. Finally, considering the WTDs, we see how they clearly differentiate between the two voltages. For the static voltage, the electronic wave packets are strongly overlapping and the WTD is rather structure- less. By contrast, the application of Lorentzian-shaped voltage pulses leads to the emission of well-localized elec- tron wave packets with a small overlap, giving raise to clear peaks at multiples of the period of the driving T . For a fully transmitting QPC (T = 1), the WTD essen- tially consists of a single peak around τ (cid:39) T (see also Ref. 31). For a QPC with a finite transmission, as in Fig. 3, levitons may reflect back on the QPC, giving rise to the series of peaks in the WTD. We note that the period and the width of the Lorentzian-shaped pulses, which determine the overall structure of the WTD, are easily tunable in an experiment. Independently of the applied voltage, the WTD is sup- pressed to zero at τ = 0. This is a consequence of the Pauli principle which forbids two electrons to occupy the same state. For a QPC with non unit transmission T , the mean waiting time takes the form where the value at full transmission, (cid:104)τ(cid:105)T =1, is given by ¯τ defined in Eq. (2) for the static voltage and by the period of the driving T for the levitons. distributed.40 The joint probability distributions then factorize as Win(τ1, . . . , τn) ≈ n(cid:89) Win(τi). Consequently, in Laplace space, we have i=1 and W (0)(z) = Win(z) W (1)(z) = [Win(z)]2, 6 (35) (36) (37) (38) (cid:104)τ(cid:105)T = (cid:104)τ(cid:105)T =1 T , and generally (31) W (n)(z) = [Win(z)]n+1, making use of the fact the n-resolved WTDs under the renewal assumption are convolutions in the time domain. Laplace transforming Eq. (32), we now obtain a geo- metric series which can be resummed as31 ∞(cid:88) E. Renewal theory W(z) = T Rn[Win(z)]n+1 ∞(cid:88) It is instructive to investigate the influence of the QPC in more detail. In this connection we also introduce the concept of a renewal process. The effect of the QPC can be understood by resolving the WTD with respect to the number of reflections on the QPC that have occurred.31 Specifically, we expand the WTD as W(τ ) = T RnW (n)(τ ), (32) n=0 where W (n)(τ ) is the WTD between transmitted elec- trons given that n reflections have occurred during the waiting time τ . These n-resolved WTD's can be related to the joint probability distributions Win(τ1, . . . , τn) for n successive waiting times between incoming electrons, corresponding to the joint WTDs at full transmission. We have for example W (0)(τ ) = Win(τ ) and (cid:90) τ 0 W (1)(τ ) = Win(τ(cid:48), τ − τ(cid:48))dτ(cid:48), (33) (34) where we have integrated over the time at which the re- flection occurs. Similar expressions can be formulated for the n-resolved WTDs of higher orders in terms of the joint probability distributions for the incoming electrons. In the following section we extend our formalism to calculations of joint WTDs. However, at this point we can proceed using a renewal assumption: We make the assumption that successive waiting times are uncor- related and thus statistically independent and equally n=0 TWin(z) 1 − RWin(z) . = (39) Returning to the time domain using an inverse Laplace transform, this expression provides us with a direct test of the renewal assumption of uncorrelated waiting times. We recall that Win(τ ) is the WTD at full transmission. Figure 4 shows a comparison between exact results for the WTD and the approximation based on renewal the- ory. Both for a static voltage and for levitons generated by periodic voltage pulses, the renewal assumption cap- tures the main qualitative features of the WTDs. How- ever, upon closer inspection it becomes clear that the agreement is not perfect. This is an indication that the renewal assumption is not correct and that subsequent electron waiting times are correlated. With this in mind, we go on to develop a theory of joint WTDs. III. GENERALIZED IDLE TIME PROBABILITY We now generalize the concepts from the last section to an arbitrary number of channels and to an arbitrary number of successive waiting times. To this end, we in- troduce a generalized ITP from which we obtain the joint WTDs as well as other distributions. In the following sec- tion we illustrate our formalism with specific examples. We consider a scatterer which is connected to Ni in- coming and No outgoing channels. The generalized ITP is the probability that no particles are detected in any of the outgoing channels during the channel-dependent time intervals [ts α]. The generalized ITP reads α, te (cid:68) : e−(cid:80)No α=1 (cid:98)Qα : (cid:69) Π(ts 1, te 1; . . . ; ts No , te No ) = , (40) 7 FIG. 4. (Color online) Comparison with renewal theory. Exact results (solid lines) for two different transmissions of the QPC are shown together with the approximation in Eq. (39) (dashed lines) based on a renewal assumption of uncorrelated waiting times. (a) Results for a static voltage applied to contact 1, V1(t) = V , and no voltage applied to contact 2, V2(t) = 0. The QPC is tuned to T = 0.3 (green lines) or T = 0.7 (blue lines). We consider the electrons that are transmitted into outgoing channel B. The time is given in units of (cid:104)τ(cid:105) = h/(T eV ). (b) Results for a series of Lorentzian-shaped voltage pulses of unit charge applied to contact 1. The width of the pulses is Γ = 0.02T . The time is given in units of the period of the pulses T . having defined projectors for each channel b† α(x)bα(x)dx. (41) (cid:98)Qα = (cid:90) te α ts α The evaluation of the generalized ITP is discussed in Ap- pendix A. The introduction of individual starting and ending times for each channel allows us to compute a variety of WTDs. In each channel, the idle time inter- val [ts α] can be modified and detection events can be inserted by differentiation with respect to the time argu- ments. The single-channel ITP can always be recovered by letting the length of the intervals in all other channels go to zero α, te α;{ts i(cid:54)=α}), Π(ts α, te where the curly brackets imply ∀ i. The operators (cid:98)Qα α) = Π(ts i(cid:54)=α = te α, te (42) in Eq. (41) count the number of particles in the spatial intervals [ts α]. For this reason, the generalized ITP is closely related to the joint particle number statistics in spatial subregions.52,53 α, te A. Multiple channels For the sake of simplicity, we consider only two out- going channels labeled as A and B. The generaliza- tion to more channels is straightforward. With two outgoing channels, the ITP has four time arguments Π(ts B). We consider two different types of WTDs: The two-channel WTD is the distribution of waiting times between detections in either of the two channels. The cross-channel WTD is the distribution of waiting times between a detection in one channel and B, te A, te A; ts the next detection in the other channel. These WTDs generally have two time arguments but become one-time quantities for driven systems by averaging over a period following Eq. (20). We first discuss the two-channel WTD. This is the con- ditional probability density of detecting an electron at time te in either channel, given that the last prior detec- tion happened at the earlier time ts in any of the two channels. This WTD follows from the generalized ITP by differentiation with respect to the time arguments I(ts)WAB(ts, te) = −∂ts ∂teΠ(ts, te; ts, te), (43) A = ts where we have set ts = ts B, since we do not differentiate between the two channels, and I = IA + IB is the sum of the particle currents in each channel. If the two channels are uncorrelated, the ITP factorizes as32 B and te = te A = te Πuc(ts A, te A; ts B, te B) = ΠA(ts A, te A)ΠB(ts B, te B), (44) where Πα is the ITP in channel α and the superscript uc stands for uncorrelated. In this case, the two-channel WTD takes on a particularly illuminating form I(ts)W uc AB(ts, te) = IA(ts)WA(ts, te)ΠB(ts, te) + FB(ts, te) ¯FA(ts, te) + A ↔ B. (45) The first term represents contributions where both de- tections happen in channel A, while no detections occur in channel B. The second term corresponds to contribu- tions where the first detection happens in channel A at time ts and the next detection occurs in channel B at time te. Finally, the term A ↔ B indicates that the roles of the two channels can be interchanged. In contrast to the single-channel WTD, the Pauli prin- ciple does not force the two-channel WTD to vanish at short times (detections in the two channels can occur si- multaneously). Evaluating Eq. (43) at ts = te, we find WAB(ts, ts) = 2 (cid:104) IA(ts) IB(ts)(cid:105) I(ts) , (46) showing that the two-channel WTD at ts = te is (twice) the coincidence rate in the two channels divided by the total particle current at ts. The factor of two accounts for the fact that either of the two detections can be in- terpreted as being the first one. For two uncorrelated channels, the expectation value factorizes and we find W uc AB(ts, ts) = 2 IA(ts)IB(ts) I(ts) , (47) where Iα is the average particle current in channel α. Next we discuss the cross-channel WTD. This is the conditional probability density of waiting until te for the first detection in channel B to happen after a detection has occurred in channel A at the earlier time ts. We note that additional detection may occur in channel A after the time ts. The cross-channel WTD follows from the ITP as A, ts; ts, te)(cid:12)(cid:12)ts (cid:69) B(te) : e−(cid:98)QB : bB(te)bA(ts) ∂teΠ(ts . A A=ts IA(ts)WA→B(ts, te) = − ∂ts = † † A(ts)b (cid:68)b A = te (48) A = ts after taking the derivatives since We set ts additional detections in channel A may occur following the first detection. For two uncorrelated channels, we recover the first passage time distribution in channel B since the detection in channel A only defines ts without influencing channel B W uc A→B(ts, te) = FB(ts, te). (49) For ts = te, we again obtain the coincidence rate in the two channels. However, this time without the factor of two since the event in channel A by definition is the first one WA→B(ts, ts) = (cid:104) IA(ts) IB(ts)(cid:105) IA(ts) . (50) For uncorrelated channels this expression reduces to IB(ts) in accordance with Eq. (49). In the following section, we evaluate the two WTDs and show how waiting times may be correlated. B. Multiple times We are now in position to formulate our theory of joint WTDs. The joint distribution of n waiting times is the conditional probability to find a given sequence of n wait- ing times. As we will see, the joint WTD can be obtained 8 from the multichannel ITP defined in Eq. (40) by intro- ducing auxiliary channels. For the sake of brevity, we consider the case of just two successive waiting times in a single channel. The extension to several channels or waiting times is straightforward. α , = ∂te β ∂te α β =te β =tm α, te α; ts β, te Π(ts β, te α; ts α, te β = te To find the joint WTD, we consider the two-channel ITP Π(ts β), where β denotes an auxiliary chan- nel. Eventually, we set ts β = tm and α = β and skip the channel index. Specifically, we express the joint WTD as I(ts)W(ts, tm, te) = ∂ts β)(cid:12)(cid:12)ts (cid:68)b†(ts)b†(tm)b†(te) : e−(cid:98)Q : b(te)b(tm)b(ts) (cid:69) where the operator (cid:98)Q is given in Eq. (5). Using the aux- iliary channel a detection event is inserted at time tm between the starting ts and the end time te of the inter- val [ts, te]. In a similar way, additional detection events within the interval can be introduced. Equation (51) is a central result for the joint distribution of two succes- sive electron waiting times in a single conduction channel. Based on this expression we calculated the joint WTD in Fig. 1 and we will make further use of it in the follow- ing section, where we illustrate our method with specific examples. by integrating over the period T , cf. Eq. (20), For a driven conductor, a two-time WTD is obtained (51) W(τ1, τ2) = (cid:104)τ(cid:105) T (cid:90) T 0 I(ts)W(ts, ts + τ1, ts + τ1 + τ2)dts. In addition, we recover the standard WTD by integrating over the time at which the last detection event occurred W(ts, tm) = dteW(ts, tm, te). (52) ∞(cid:90) tm As discussed in Sec. II E, the joint WTD appears in the expansion of the WTD for a QPC. Returning to the results in Fig. 4, the ITP reads (cid:68) : e−T (cid:98)Q1 : (cid:69) (cid:68) : e(R−1)(cid:98)Q1 : (cid:69) , = (53) Π(ts, te) = where (cid:98)Q1 in Eq. (24) acts on the incoming channel 1. We can formally expand this expression in either T or R.30,31 To zeroth order in T , the ITP is unity as no electrons are transmitted through the QPC. The n'th order term in the expansion yields the reduction of the ITP due to the probability that n particles were transmitted through the QPC. To zeroth order in R, the ITP equals the ITP for a fully transmitting QPC. The n'th order term in the expansion in R equals the increase in the ITP due to the probability that n particles were reflected. Here we perform an expansion in R, and by differenti- ating the ITP with respect to ts and te we obtain W(ts, te) = T RnW (n)(ts, te) (54) ∞(cid:88) n=0 9 FIG. 5. (Color online) Joint waiting time distributions for a static voltage. (a) Results for the voltage V1(t) = V applied to contact 1 and no voltage applied to contact 2, V2(t) = 0. The QPC is fully transmitting T = 1 and we consider the electrons that are transmitted into outgoing channel B. The time is given in units of (cid:104)τ(cid:105) = h/(T eV ). (b) Difference ∆W(τ1, τ2) = W(τ1, τ2) − W(τ1)W(τ2) between exact results and results for uncorrelated waiting times. Positive correlations are indicated with red, while areas of negative correlations are blue. (c,d) Similar results for a half-transmitting QPC, T = 1/2. where IV. CORRELATED WAITING TIMES W (n)(ts, te) = te(cid:90) ts 1 n! dt1 ··· dtnWin(ts, t1,··· , tn, te) (55) is the WTD given that n reflections on the QPC have occurred and Win(ts, t1,··· , tn, te) is the joint WTD for n + 2 detection events in the incoming channel 1. Each term in the expansion is the probability density for n par- ticles to be reflected (corresponding to the prefactor Rn) followed by a transmission (corresponding to the pref- actor T ). We integrate over all possible times that the reflections can occur and the factor 1/n! corrects for mul- tiple counting of reflections. We see that joint WTDs occur already in the expansion of the WTD for a QPC. By averaging over ts and making a renewal assumption, we can recover the results from Sec. II E. We now illustrate our formalism for joint WTDs using the setup in Fig. 2. First we consider the partitioning of electrons emitted from one source. In the second example we discuss an electronic analog of the Hong-Ou-Mandel interferometer, where electrons from different input chan- nels interfere on the QPC.54 -- 59 A. Single-source partitioning We consider the setup in Fig. 2 with contact 2 grounded and contact 1 biased with a constant voltage or a train of Lorentzian-shaped voltage pulses. Figures 3 and 4 show single-channel WTDs for this setup. We now go on to calculate two-channel and joint WTDs. The two-channel WTD is the distribution of waiting times between successive electrons irrespective of the channel in which they are detected. As the QPC merely distributes the incoming electrons into the two outgoing channels, we expect simply to recover the WTD of the incoming electrons from channel 1. Indeed, using our 10 FIG. 6. (Color online) Joint waiting time distribution for levitons. (a) Results for a series of Lorentzian-shaped voltage pulses of unit charge applied to contact 1. The width of the pulses is Γ = 0.05T and the QPC is tuned to T = 1/2. The time is given in units of the period of the pulses T . (b) Difference ∆W(τ1, τ2) = W(τ1, τ2) − W(τ1)W(τ2) between exact results and results for uncorrelated waiting times. Positive correlations are indicated with red, while areas of negative correlations are blue. formalism we find from Eq. (43) that within our formalism we readily find, cf. Eq. (48), WA→B(ts, te) = WB(ts, te) (56) = RT IA(ts) † † 1(ts)a a (cid:68) (cid:69) 1(te) : e−T (cid:98)Q1 : a1(te)a1(ts) , (57) I(ts)WAB(ts, te) = −∂ts∂te = −∂ts∂te (cid:68) (cid:69) : e−(cid:98)QA−(cid:98)QB : (cid:69) (cid:68) : e−(cid:98)Q1 : , where the integrations in the (cid:98)Q operators run from ts addition, we have used Eqs. (24) and (25) to relate (cid:98)Qα to (cid:98)Q1 combined with Eq. (22). For a constant voltage, to te and I(ts) is the average current in channel 1. In the WTD was evaluated in Ref. 30 and found to be well- approximated by a Wigner-Dyson distribution reflecting Fermi correlations between the incoming electrons. For a train of Lorentzian-shaped voltage pulses, the WTD was evaluated in Ref. 31 and found to be peaked around the period of the driving with small satellite peaks due to the finite overlap of the voltage pulses. As we see here, the QPC has no effect on the WTD if we consider de- tections irrespective of the channel where they happen. This argument only applies if the QPC partitions elec- trons emitted by a single source. In that case, detections in the outgoing channels are not statistically indepen- dent, and the suppression of the WTD at τ = 0 remains. Next, we consider the cross-channel WTD. This is the distribution of waiting times between a detection in one outgoing channel and the next detection in the other channel. Since the QPC just randomly partitions the incoming electrons into the outgoing channels, we ex- pect the cross-channel WTD to equal the single-channel WTD. The cross-channel WTD is conditioned on the first detection happening in channel A, whereas the single- channel WTD is conditioned on the first detection hap- pening in channel B. In either case, the reflection or transmission of the first particle does not influence the particles that traverse the QPC at a later time. Indeed, having used IA(ts)/R = IB(ts)/T = I(ts). For the partitioning of electrons emitted from a sin- gle source, we see that the two-channel WTD and the cross-channel WTD do not provide additional informa- tion compared with the WTD itself. This is a direct con- sequence of the operator proportionality (cid:98)QA ∝ (cid:98)QB and it does not hold when both contacts emit electrons above the Fermi sea as we will see in the following example. We now consider just one of the outgoing channels and calculate the joint probability distribution of finding two successive waiting times τ1 and τ2 using Eq. (51). For a constant voltage V , the joint WTD at full transmission is shown in Fig. 1. As mentioned in the introduction, the joint WTD is well-approximated by the generalized Wigner-Dyson distribution in Eq. (3). We find that the joint WTD is symmetric with respect to an exchange of the waiting times, τ1 ↔ τ2. This symmetry implies that the WTD does not change if we invert the spatial argu- ments of all the b operators in Eq. (51). The symmetry in the WTD is thus a consequence of the spatial inversion symmetry of the many-body wavefunction. In Fig. 5, we show joint WTDs both for full trans- mission and at half transmission. To highlight possible correlations, we also show the difference ∆W(τ1, τ2) = W(τ1, τ2) − W(τ1)W(τ2), (58) between the joint WTD and a factorized WTD cor- responding to uncorrelated waiting times.40 The figure clearly demonstrates that electron waiting times are cor- related. The probability of observing a waiting time which is shorter (longer) than the mean waiting time (cid:104)τ(cid:105) is reduced if the previous waiting time was already shorter (longer) than the mean waiting time. On the other hand, a short waiting time will likely be followed by a long waiting time and vice versa. This conclusion holds for both values of the transmission. A similar analysis can be carried out for levitons emit- In ted by a train of Lorentzian-shaped voltage pulses. Fig. 6 we show W(τ1, τ2) and ∆W(τ1, τ2) for levitons inci- dent on a QPC tuned to half transmission. Again, we ob- serve a symmetry under the interchange of waiting times due to the spatial inversion symmetry of the fermionic many-body state. The joint WTD displays a lattice- like structure in contrast to the results for the dc-biased source in Fig. 5. This reflects the regular on-demand na- ture of the leviton source. Also here, the probability to observe two short or two long waiting times after each other is suppressed, while a short (long) waiting time is likely followed by a long (short) waiting time. Due to the external driving, the sum of two subse- quent waiting times is likely to equal a multiple of the period. If a particular waiting time is shorter than the average waiting time, this will not affect the absolute emission time of the next electron. Consequently, the next waiting time will most likely be longer. Thus, a strong regularity in the absolute electron emission times leads to strong dependences of the waiting times on each other. For the case that the voltage pulses overlap more and more until the limit of a constant voltage is reached, the regularity in the emission times is provided by the Pauli principle. Moreover, this effect is independent of the QPC transmission. In the low transmission regime (not shown), however, electron emission becomes increas- ingly rare and transport resembles a Poisson process.30 Correlations between waiting times will then eventually be negligible. B. Hong-Ou-Mandel interferometry As a second application, we consider the electronic analog of the Hong-Ou-Mandel experiment developed in quantum optics.54 -- 59 In this case, a driven on-demand source in each incoming channel emits single electrons onto the QPC in Fig. 2 with a possible time delay ∆τ between the emissions from the two sources. Such exper- iments have recently been realized with two mesoscopic capacitors59 and two leviton sources.35 Here we treat the periodic emission of levitons onto the QPC. Figure 7a shows the distribution of waiting times be- tween detection events in either of the two outgoing chan- nels. With zero time delay, incoming levitons antibunch on the QPC and there is a large probability of detect- ing the two outgoing levitons nearly simultaneously. The peak in the WTD around the period T corresponds to the waiting time between the last leviton in one cycle and the first one in the next cycle. Since the probabil- ity of measuring the waiting time between two levitons within one cycle is equal to the probability of measuring 11 the waiting time between levitons in different cycles, the areas underneath the two peaks are the same. In general, for a given time delay ∆τ , we find peaks in the WTD at ∆τ and at T − ∆τ , corresponding to the waiting times between levitons within the same cycle and the waiting between levitons in different cycles. In the special case ∆τ = T /2, the two peaks merge into one, except for a small remaining feature at τ = T . This fea- ture is related to the satellite peak seen in the WTD for levitons in a single channel with perfect transmission.31 The WTD decays strongly for waiting times beyond the period, since all levitons will be detected independently of the QPC. This reflects the high reliability of the leviton sources which emit exactly one electron per cycle. This example demonstrates how the two-channel WTD pro- vides information about the synchronization of the two sources. Importantly, this detailed characterization does not depend on the transmission of the QPC. In Fig. 7b we turn to the distribution of waiting times between detections in different channels. This distribu- tion is rather similar to the WTD for a single channel, shown in the left inset for comparison. Unlike the single- channel WTD, the cross-channel WTD is not suppressed to zero at τ = 0, since detections can occur simultane- ously in the two channels. For finite delay times between the two sources, the particles can go into the same out- going channel, and both WTDs show a peaked structure even for waiting times larger than the period. Due to the Pauli principle, two levitons arriving si- multaneously at the QPC will go into different outgo- ing channels with unit probability, independently of the transmission. Thus, for zero time delay all results are in- dependent of the QPC transmission T and reduce to the results for full transmission. As the time delay between the two sources is increased, the overlap between levitons arriving at the QPC decreases, and simultaneous detec- tions in the outgoing channels become increasingly rare. For ∆τ = T /4, the cross-channel WTD essentially shows four peaks within the first period. The peak at τ = 0 is a relict of the fermionic antibunching due to the finite overlap of the pulses. The peaks at ∆τ and at T − ∆τ are caused by two successive particles entering opposite arms of the interferometer, while the peak at T corre- sponds to two successive particles entering the same arm and the next particle entering the opposite arm. For the maximal detuning ∆τ = T /2, the WTD resembles that of just a single channel. Due to the Pauli principle the cross-channel WTD of a single source is suppressed for zero waiting times. By constrast, with two sources we observe a local maximum due to the antibunching of electrons that arrive simulta- neously at the QPC. The right inset of Fig. 7b shows the cross-channel WTD at τ = 0 as a function of the time de- lay, with a maximum for zero time delay in analogy with the Hong-Ou-Mandel peak found in the zero-frequency current cross-correlations.58,59 We see that the two-channel and cross-channel WTDs contain different information about the emission pro- 12 FIG. 7. (Color online) Distributions of waiting times for a Hong-Ou-Mandel experiment with levitons. Two sources emit levitons of pulse width Γ = 0.05T toward a QPC with transmission T = 1/2. The tunable time delay between the sources is denoted as ∆τ . (a) WTD for waiting times between events occurring in either of the two outgoing channels for different values of the time delay. (b) WTD for waiting times between detection events in different channels. The left inset shows the WTD for just a single channel. The right inset shows the WTD at τ = 0 as a function of the time delay ∆τ between the sources. cesses. The two-channel WTD mostly contains informa- tion about the sources alone, in particular their synchro- nization. The cross-channel WTD contains additional information about the QPC and shows more prominent signatures of fermionic antibunching. V. CONCLUSIONS We have developed a general framework for calculat- ing joint WTDs in electronic multi-channel conductors. The central building block of our formalism is the gener- alized idle time probability, i. e. the joint probability for no detections to occur in the outgoing channels. By cal- culating the joint WTD for a single conduction channel, we have explicitly demonstrated that the electron wait- ing times in coherent conductors are correlated due to the fermionic statistics encoded in the many-body state. Drawing on the analogy between random matrices and free fermions, we have shown that the joint WTD for a fully transmitting conduction channel is well- approximated by a generalized Wigner-Dyson distribu- tion. In contrast to a renewal process with uncorrelated waiting times, we find that the probability of observ- ing a long (short) waiting time following a short (long) waiting time is increased, while finding two long or two short waiting times in succession is less likely. This holds both for electrons coming from a dc-biased contact and for levitons emitted on top of the Fermi sea by applying Lorentzian-shaped voltage pulses to the contact. Correlations between electrons in different outgoing channels also show up in the distributions of waiting times between detections in different channels. We have defined multi-channel and cross-channel WTDs and illus- trated these concepts for a QPC in a chiral setup where electrons are injected in either one or both incoming channels. In a fermionic Hong-Ou-Mandel experiment, where electrons interfere on the QPC, the two-channel WTD provides information about the synchronization of the sources, while the cross-channel WTD shows signa- tures of the scatterer and the fermionic antibunching. Our formalism relies on the ability to detect single elec- trons above the Fermi level. A quantum theory of such a detector is currently being developed. It would also be interesting to make use of WTDs to detect entanglement in electronic conductors, for instance by formulating a set of Leggett-Garg inequalities for the joint WTDs. ACKNOWLEDGMENTS We thank M. Albert and G. Haack for many useful discussions. CF is affiliated with Centre for Quantum Engineering at Aalto University. PH gratefully acknowl- edges the hospitality of McGill University where part of the work was done. The work was supported by the Swiss NSF. Appendix A: Computation of the generalized ITP To compute the generalized ITP in Eq. (40), we map the outgoing operators onto the incoming ones using the Floquet scattering matrix51,60 SF,αβ(E, En)aβ(En), (A1) ∞(cid:88) Ni(cid:88) bα(E) = n=−∞ β=1 where Ω is the frequency of the external driving and we have defined the energies En = E + nΩ, (A2) for integer n. Since the incoming equilibrium particles are described by a Slater determinant, we may evalu- ate the quantum statistical average in Eq. (40) and we thereby obtain a determinant formula30 -- 32 of the form ) = det(I − Q(ts No , te 1, te 1; . . . ; ts No particle matrix elements of the operator(cid:80) Π(ts )). Here I is the identity matrix and Q contains the single- The matrix Q has a block form with elements in the α (cid:98)Qα. 1; . . . ; ts No 1, te , te No (incoming) channel space [Q(ts 1, te 1, . . . , ts No , te No )]αβ = Kαβ (A3) with the Kαβ being matrices in energy space. In the general case of a time-dependent scatterer with many channels described by the Floquet scattering matrix in Eq. (A1), these matrices have the elements [Kαβ]E,E (cid:48) = † S F,γα(Em, E)SF,γβ(E (cid:48) n, E (cid:48) ) No(cid:88) ∞(cid:88) ∞(cid:88) γ=1 m=−(cid:98)E/Ω(cid:99) n=−(cid:98)E /Ω(cid:99) × Θ(−E)Θ(−E (cid:48) where we have defined e−iE(ts γ +te κ π γ − ts γ )/2 sin(E(te E γ)/2) . γ, te γ)]E = [K(ts (A5) In addition, the floor function is denoted as (cid:98)·(cid:99) and we have discretized30,32,47 the transport window [EF , EF + eV ] into compartments of width κ = eV /N , where N is the number of particles. We always consider the limit N → ∞. The matrix has a block form with respect to the incoming channels, while we sum over the indices of the outgoing channels. The result in Eq. (A4) generalizes the earlier expression for a single channel31 to many channels. Appendix B: Levitons 13 Here we briefly discuss the Floquet scattering theory for the periodic creation of levitons.7,42,43,61 Applying a series of periodic Lorentzian voltage pulses of unit charge, ∞(cid:88) j=−∞ eV (t) = 2Γ (t − t0 − jT )2 + Γ2 , (B1) to one of the reservoirs leads to the creation of one clean electron excitation per period -- a leviton -- without ac- companying electron-hole pairs. Here, the period of the pulses is denoted as T = 2π/Ω and the parameter t0 can be adjusted to shift the pulses in time. In the adiabatic limit, where the variation of the voltage is slow com- pared to the scattering time, the elements of the Floquet scattering matrix can be obtained by Fourier transform- ing the time dependent scattering phase in Eq. (28) with V (t) defined in Eq. (B1). We then obtain61 This Floquet scattering matrix relates particles incident on the scatterer to equilibrium particles through Eq. (29). Together with the scattering matrix of the scatterer itself, it yields the Floquet scattering matrix used for the calcu- lation of the generalized idle time probability in Eq. (A4). For the Hong-Ou-Mandel calculations, the time delay be- tween the sources can be tuned by changing the parame- ter t0 in one of the matrix elements in the channel space. (cid:48) )[K(ts γ, te γ)]Em−E , (cid:48) n (A4) Sn = einΩt0 −2e−nΩΓ sinh(ΩΓ) e−ΩΓ 0 if n > 0 if n = 0 otherwise . (B2) ∗ These authors contributed equally to the present work. 1 Ya. M. Blanter and M. Buttiker, "Shot noise in mesoscopic conductors," Phys. Rep. 336, 1 (2000). 2 G. B. Lesovik, "Excess quantum noise in 2D ballistic point contacts," JETP Lett. 49, 594 (1989). 3 M. Buttiker, "Scattering theory of thermal and excess noise in open conductors," Phys. Rev. Lett. 65, 2901 (1990). 4 T. Martin and R. Landauer, "Wave-packet approach to noise in multichannel mesoscopic systems," Phys. Rev. B 45, 1742 (1992). 5 A. Kumar, L. Saminadayar, D. C. Glattli, Y. Jin, and B. Etienne, "Experimental test of the quantum shot noise reduction theory," Phys. Rev. Lett. 76, 2778 (1996). 6 L. S. Levitov and G. B. Lesovik, "Charge distribution in quantum shot noise," JETP Lett. 58, 230 (1993). 7 L. S. Levitov, H. Lee, and G. B. Lesovik, "Electron count- ing statistics and coherent states of electric current," J. Math. Phys. 37, 4845 (1996). 8 Yu. V. Nazarov, ed., Quantum Noise in Mesoscopic Physics (Kluwer, Dordrecht, 2003). 9 A. Shelankov and J. Rammer, "Charge transfer counting statistics revisited," Europhys. Lett. 63, 485 (2003). 10 M. Vanevi´c, Yu. V. Nazarov, and W. Belzig, "Elemen- tary events of electron transfer in a voltage-driven quan- tum point contact," Phys. Rev. Lett. 99, 076601 (2007). 11 M. Vanevi´c, Yu. V. Nazarov, and W. Belzig, "Elementary charge-transfer processes in mesoscopic conductors," Phys. Rev. B 78, 245308 (2008). 12 A. G. Abanov and D. A. Ivanov, "Allowed charge transfers between coherent conductors driven by a time-dependent scatterer," Phys. Rev. Lett. 100, 086602 (2008). 13 A. G. Abanov and D. A. Ivanov, "Factorization of quantum charge transport for noninteracting fermions," Phys. Rev. B 79, 205315 (2009). 14 D. Kambly and D. A. Ivanov, "Statistics of quantum trans- fer of noninteracting fermions in multiterminal junctions," Phys. Rev. B 80, 193306 (2009). 15 D. Kambly, C. Flindt, and M. Buttiker, "Factorial cumu- lants reveal interactions in counting statistics," Phys. Rev. B 83, 075432 (2011). 16 I. Klich and L. S. Levitov, "Quantum noise as an entan- glement meter," Phys. Rev. Lett. 102, 100502 (2009). 17 H. F. Song, C. Flindt, S. Rachel, I. Klich, and K. Le Hur, "Entanglement entropy from charge statistics: Exact rela- tions for noninteracting many-body systems," Phys. Rev. B 83, 161408 (2011). 18 H. F. Song, S. Rachel, C. Flindt, I. Klich, N. Laflorencie, and K. Le Hur, "Bipartite fluctuations as a probe of many- body entanglement," Phys. Rev. B 85, 035409 (2012). 19 A. Petrescu, H. F. Song, S. Rachel, Z. Ristivojevic, C. Flindt, N. Laflorencie, I. Klich, N. Regnault, and K. Le Hur, "Fluctuations and entanglement spectrum in quantum hall states," J. Stat. Mech. , P10005 (2014). 20 K. H. Thomas and C. Flindt, "Entanglement entropy in dynamic quantum-coherent conductors," Phys. Rev. B 91, 125406 (2015). 21 H. Forster and M. Buttiker, "Fluctuation relations without microreversibility in nonlinear transport," Phys. Rev. Lett. 101, 136805 (2008). 22 K. Saito and Y. Utsumi, "Symmetry in full counting statis- tics, fluctuation theorem, and relations among nonlinear transport coefficients in the presence of a magnetic field," Phys. Rev. B 78, 115429 (2008). 23 M. Esposito, U. Harbola, and S. Mukamel, "Nonequi- librium fluctuations, fluctuation theorems, and counting statistics in quantum systems," Rev. Mod. Phys. 81, 1665 (2009). 24 T. Brandes, "Waiting times and noise in single particle transport," Ann. Phys. 17, 477 (2008). 25 S. Welack, S. Mukamel, and Y. Yan, "Waiting time distributions of electron transfers through quantum dot Aharonov-Bohm interferometers," Europhys. Lett. 85, 57008 (2009). 26 M. Albert, C. Flindt, and M. Buttiker, "Distributions of waiting times of dynamic single-electron emitters," Phys. Rev. Lett. 107, 086805 (2011). 27 L. Rajabi, C. Poltl, and M. Governale, "Waiting time dis- tributions for the transport through a quantum-dot tun- nel coupled to one normal and one superconducting lead," Phys. Rev. Lett. 111, 067002 (2013). 28 B. Sothmann, "Electronic waiting-time distribution of a quantum-dot spin valve," Phys. Rev. B 90, 155315 (2014). 29 K. H. Thomas and C. Flindt, "Electron waiting times in non-markovian quantum transport," Phys. Rev. B 87, 121405 (2013). 30 M. Albert, G. Haack, C. Flindt, and M. Buttiker, "Elec- tron waiting times in mesoscopic conductors," Phys. Rev. Lett. 108, 186806 (2012). 31 D. Dasenbrook, C. Flindt, and M. Buttiker, "Floquet the- ory of electron waiting times in quantum-coherent conduc- tors," Phys. Rev. Lett. 112, 146801 (2014). 32 G. Haack, M. Albert, and C. Flindt, "Distributions of electron waiting times in quantum-coherent conductors," Phys. Rev. B 90, 205429 (2014). 33 K. H. Thomas and C. Flindt, "Waiting time distributions of noninteracting fermions on a tight-binding chain," Phys. Rev. B 89, 245420 (2014). 34 G. F`eve, A. Mah´e, J.-M. Berroir, T. Kontos, B. Pla¸cais, D. C. Glattli, A. Cavanna, B. Etienne, and Y. Jin, "An on-demand coherent single-electron source," Science 316, 14 1169 (2007). 35 J. Dubois, T. Jullien, F. Portier, P. Roche, A. Cavanna, Y. Jin, W. Wegscheider, P. Roulleau, and D. C. Glat- tli, "Minimal-excitation states for electron quantum optics using levitons," Nature 502, 659 (2013). 36 T. Jullien, P. Roulleau, B. Roche, A. Cavanna, Y. Jin, and D. C. Glattli, "Quantum tomography of an electron," Nature 514, 603 (2014). 37 M. Albert and P. Devillard, "Waiting time distribution for trains of quantized electron pulses," Phys. Rev. B 90, 035431 (2014). 38 J. D. Fletcher, P. See, H. Howe, M. Pepper, S. P. Giblin, J. P. Griffiths, G. A. C. Jones, I. Farrer, D. A. Ritchie, T. J. B. M. Janssen, and M. Kataoka, "Clock-controlled emission of single-electron wave packets in a solid-state circuit," Phys. Rev. Lett. 111, 216807 (2013). 39 R. Thalineau, A. D. Wieck, C. Bauerle, and T. Meunier, "Using a two-electron spin qubit to detect electrons flying above the fermi sea," (2014), arXiv:1403.7770. 40 D. R. Cox, Renewal Theory (Chapman and Hall, 1962). 41 D. Herman, T. T. Ong, G. Usaj, H. Mathur, and H. U. Baranger, "Level spacings in random matrix theory and coulomb blockade peaks in quantum dots," Phys. Rev. B 76, 195448 (2007). 42 D. A. Ivanov, H. W. Lee, and L. S. Levitov, "Coher- ent states of alternating current," Phys. Rev. B 56, 6839 (1997). 43 J. Keeling, I. Klich, and L. S. Levitov, "Minimal excitation states of electrons in one-dimensional wires," Phys. Rev. Lett. 97, 116403 (2006). 44 J.-P. Brantut, J. Meineke, D. Stadler, S. Krinner, and T. Esslinger, "Conduction of ultracold fermions through a mesoscopic channel," Science 337, 1069 (2012). 45 R. Vyas and S. Singh, "Waiting-time distributions in the photodetection of squeezed light," Phys. Rev. A 38, 2423 (1988). 46 S. Saito, J. Endo, T. Kodama, A. Tonomura, A. Fukuhara, and K. Ohbayashi, "Electron counting theory," Phys. Lett. A 162, 442 (1992). 47 F. Hassler, M. V. Suslov, G. M. Graf, M. V. Lebedev, G. B. Lesovik, and G. Blatter, "Wave-packet formalism of full counting statistics," Phys. Rev. B 78, 165330 (2008). 48 G. Giuliani and G. Vignale, Quantum Theory of the Elec- tron Liquid (Cambridge University Press, 2005). 49 G.-M. Tang, F. Xu, and J. Wang, "Waiting time distri- bution of quantum electronic transport in the transient regime," Phys. Rev. B 89, 205310 (2014). 50 M. H. Pedersen and M. Buttiker, "Scattering theory of photon-assisted electron transport," Phys. Rev. B 58, 12993 (1998). 51 M. Moskalets and M. Buttiker, "Floquet scattering theory of quantum pumps," Phys. Rev. B 66, 205320 (2002). 52 A. Shelankov and J. Rammer, "Counting statistics of in- terfering Bose-Einstein condensates," Europhys. Lett. 83, 60002 (2008). 53 J. Rammer and A. Shelankov, "Counting quantum fluctua- tions of particle density," Ann. Phys. 524, 163 -- 174 (2012). and S. Tarucha, "Quantum interference in electron collision," Nature 391, 263 -- 265 (1998). 54 R. C. Liu, B. Odom, Y. Yamamoto, 55 G. Burkard, D. Loss, and E. V. Sukhorukov, "Noise of entangled electrons: Bunching and antibunching," Phys. Rev. B 61, R16303 (2000). 56 S. Ol'khovskaya, J. Splettstoesser, M. Moskalets, and M. Buttiker, "Shot noise of a mesoscopic two-particle col- lider," Phys. Rev. Lett. 101, 166802 (2008). 57 V. Giovannetti, D. Frustaglia, F. Taddei, and R. Fazio, "Electronic Hong-Ou-Mandel interferometer for multi- mode entanglement detection," Phys. Rev. B 74, 115315 (2006). 58 T. Jonckheere, J. Rech, C. Wahl, and T. Martin, "Electron and hole Hong-Ou-Mandel interferometry," Phys. Rev. B 86, 125425 (2012). 59 E. Bocquillon, V. Freulon, J.-M. Berroir, P. Degiovanni, B. Pla¸cais, A. Cavanna, Y. Jin, and G. F`eve, "Coher- 15 ence and indistinguishability of single electrons emitted by independent sources," Science 339, 1054 (2013). 60 M. V. Moskalets, Scattering matrix approach to non- stationary quantum transport (Imperial College Press, 2012). 61 J. Dubois, T. Jullien, C. Grenier, P. Degiovanni, P. Roul- leau, and D. C. Glattli, "Integer and fractional charge lorentzian voltage pulses analyzed in the framework of photon-assisted shot noise," Phys. Rev. B 88, 085301 (2013). 

1604.01749

1

1604

Quantum Wigner molecules in semiconductor quantum dots and cold-atom optical traps and their mathematical symmetries

Strong repelling interactions between a few fermions or bosons confined in two-dimensional circular traps lead to particle localization and formation of quantum Wigner molecules (QWMs) possessing definite point-group space symmetries. These point-group symmetries are "hidden" (or emergent), namely they cannot be traced in the circular single-particle densities (SPDs) associated with the exact many-body wave functions, but they are manifested as characteristic signatures in the ro-vibrational spectra. An example, among many, are the few-body QWM states under a high magnetic field or at fast rotation, which are precursor states for the fractional quantum Hall effect. The hidden geometric symmetries can be directly revealed by using spin-resolved conditional probability distributions, which are extracted from configuration-interaction (CI), exact-diagonalization wave functions. The hidden symmetries can also be revealed in the CI SPDs by reducing the symmetry of the trap (from circular to elliptic to quasi-linear). In addition the hidden symmetries are directly connected to the explicitly broken-symmetry (BS) solutions of mean-field approaches, such as unrestricted Hartree-Fock (UHF). A companion step of restoration of the broken symmetries via projection operators applied on the BS-UHF solutions produces wave functions directly comparable to the CI ones, and sheds further light into the role played by the emergence of hidden symmetries in the exact many-body wave functions. Illustrative examples of the importance of hidden symmetries in the many-body problem of few electrons in semiconductor quantum dots and of few ultracold atoms in optical traps (where unprecedented control of the interparticle interaction has been experimentally achieved recently) will be presented.

cond-mat.mes-hall

cond-mat

Proceedings of the 16th International Conference on Computational and Mathematical Methods in Science and Engineering, CMMSE 2016 4 -- 8 July, 2016. Quantum Wigner molecules in semiconductor quantum dots and cold-atom optical traps and their mathematical symmetries Constantine Yannouleas1 and Uzi Landman1 1 School of Physics, Georgia Institute of Technology, Atlanta, GA 30332-0430, USA emails: Constantine.Yannouleas@physics.gateh.edu, Uzi.Landman@physics.gateh.edu Abstract Strong repelling interactions between a few fermions or bosons confined in two- dimensional circular traps lead to particle localization and formation of quantum Wigner molecules (QWMs) possessing definite point-group space symmetries. These point- group symmetries are "hidden" (or emergent), namely they cannot be traced in the circular single-particle densities (SPDs) associated with the exact many-body wave func- tions, but they are manifested as characteristic signatures in the ro-vibrational spectra. An example, among many, are the few-body QWM states under a high magnetic field or at fast rotation, which are precursor states for the fractional quantum Hall effect. The hidden geometric symmetries can be directly revealed by using spin-resolved con- ditional probability distributions, which are extracted from configuration-interaction (CI), exact-diagonalization wave functions. The hidden symmetries can also be re- vealed in the CI SPDs by reducing the symmetry of the trap (from circular to elliptic to quasi-linear). In addition the hidden symmetries are directly connected to the ex- plicitly broken-symmetry (BS) solutions of mean-field approaches, such as unrestricted Hartree-Fock (UHF). A companion step of restoration of the broken symmetries via projection operators applied on the BS-UHF solutions produces wave functions directly comparable to the CI ones, and sheds further light into the role played by the emergence of hidden symmetries in the exact many-body wave functions. Illustrative examples of the importance of hidden symmetries in the many-body problem of few electrons in semiconductor quantum dots and of few ultracold atoms in optical traps (where un- precedented control of the interparticle interaction has been experimentally achieved recently) will be presented. Key words: Wigner molecule, emergent point-group symmetries, broken symmetries, symmetry restoration, projection operator, unrestricted Hartree Fock, configuration in- teraction, 2D semicoductor quantum dots, trapped ultracold atoms, fermions, bosons c(cid:13)CMMSE ISBN: 978-84-608-6082-2 Quantum Wigner molecules and symmetries Figure 1: Unrestricted Hartree-Fock electron density in a 2D parabolic QD for N = 19 electrons and total-spin projection Sz = 19/2, exhibiting breaking of the circular symmetry at RW = 5 and zero magnetic field. The electrons are (partially) localized in a (1,6,12) multi-ring structure. which exhibits point-group symmetries. Remaining parameters are: parabolic confinement, ω0 = 5 meV; effective mass m∗ = 0.067me. Distances are in nanometers and the electron density in 10−4 nm−2. 1 Introduction This talk focuses on novel, somewhat exotic, types of clusters of few fermions or bosons. In particular, we discuss clusters of electrons in manmade (artificial) quantum dots (QDs) created through lithographic and gate-voltage techniques at semiconductor interfaces, and clusters of neutral ultracold atoms (either bosonic or fermionic) in harmonic optical traps. These cluster systems exhibit interesting emergent physical behavior arising from spon- taneous breaking of spatial and/or spin symmetries at the mean-field level of theoretical treatment [1, 2]; symmetry breaking (SB) is defined as a circumstance where a lower energy solution of the Schrodinger equation is found that is characterized by a lower symmetry than that of the full many-body Hamiltonian of the few-body system. Such SB in circu- lar traps directly reflects the localization of of particles in cluster arrangements exhibiting point-group symmetries instead of the continuous rotational symmetry expected from the many-body Hamiltonian [2]. A prominent example is the formation of finite electron crys- tallites (referred to as semi-classical Wigner molecules, SCWMs) in two-dimensional (2D) QDs (see Fig. 1). Symmetry breaking at the mean-field level is also manifested in the tran- sition [3, 4], induced by increasing the interatomic repulsive contact-interaction strength, of the ground state of neutral atoms in a parabolic or toroidal 2D trap to a rotating bosonic quantum Wigner molecule (QWM). An example is presented in Fig. 2, where the hierar- chy of the successive approximations (broken symmetry UHF → symmetry restoration) is illustrated, leading to the symmetry-restored, fully-quantal wave function (QWM) in Fig. 2 (c,d); the single-particle density (SPD) for the intermediate BS-UHF (SCWM) wave func- c(cid:13)CMMSE ISBN: 978-84-608-6082-2 -60060-60060026ElectronDensityx (nm)y (nm)UHFYannouleas and Landman Single-particle densities and CPDs for N = 8 neutral repelling bosons in a Figure 2: rotating 2D toroidal trap with reduced rotational frequency Ω/ω0 = 0.2 and Rδ = 50. The 0(r − r0)2/2, is centered at a radius r0 = 3l0. (a) Gross-Pitaevski confining potential, mω2 SPD. (b) UHF SPD exhibiting breaking of the circular symmetry. (c) QWM SPD exhibiting circular symmetry. (d) CPD for the QWM wave function (resulting from the method of symmetry restoration), revealing the hidden point-group symmetry in the intrinsic frame of reference. The fixed observation point is denoted by a white dot. The QWM ground-state angular momentum is Lz = 16. Lengths in units of the oscillator length l0. The vertical scale is the same for (b), (c), and (d), but different for (a). tion is plotted in Fig. 2(b). Note that the point-group symmetry is not visible in the SPD after the step of symmetry restoration is executed, i.e., it becomes hidden [see Fig. 2(c)], but it is revealed via a conditional probability distribution (CPD); see Fig. 2(d). The CPD gives the probability of finding a particle with spin σ at position r given that another one (referred to as the fixed particle) with spin σ0 is located at r0. The degree of particle localization is controlled by the Wigner parameter that specifies the strength of the interparticle repulsion relative to the zero-point kinetic energy, i.e., RW = Z2e2/(κl0ω0) [2, 3] for a Coulomb repulsion and Rδ = gm/(2π2) [2, 3, 4] for a contact-potential (Dirac- delta) repulsion; Z is the charge of the particle, κ is the dielectric constant, ω0 is the frequency of the harmonic trap, l0 =(cid:112)/(mω0) is the oscillator length, m is the particle mass, and g is the strength of the contact interaction. Of great value in analyzing the physics associated with the hidden symmetries is the evolution of the lowest-energy band in the energy spectra (yrast band) as a function of the two successive approximations (broken symmetry UHF → symmetry restoration). Figs. 3(a,b) present the evolution of yrast spectra (as a function of the rotational frequency Ω of the trap) that are associated with the class of wave functions portrayed in Fig. 2. The most prominent trend is that the ground-state angular momenta of the symmetry-restored wave functions do not assume all the possible 2D values, but are restricted to stepwise values Lz = N k, k = 0, 1, 2, . . ., with N = 8, i.e., they change in steps of N = 8, where N is c(cid:13)CMMSE ISBN: 978-84-608-6082-2 Quantum Wigner molecules and symmetries Figure 3: Properties of N = 8 neutral repelling bosons in a rotating 2D toroidal trap as a function of the reduced rotational frequency Ω/ω0. The confining toroidal potential is centered at a radius r0 = 3l0, and the interaction-strength parameter was chosen as Rδ = 50. (a) QWM ground-state energies, EP RJ . The term QWM corresponds to projected (PRJ) wave functions that preserve the total angular momentum (symmetry restoration). The inset shows the range 0 ≤ Ω/ω0 ≤ 0.3. The numbers denote ground-state magic angular (b) Energy difference EP RJ − EU BHF , where the subscript "UBHF" stands momenta. for unrestricted bosonic Hartree-Fock. (c) Total angular momenta associated with (i) the QWM ground states [thick solid line (showing steps and marked as PRJ); online black] and (ii) the broken-symmetry UBHF solutions (smooth thin solid line; online red). the number of particles [see Fig. 3(c) and the inset in Fig. 3(a)]. Such stepwise angular momenta are usually referred to as "magic" and the associated ground states of enhanced stability [see Fig. 3(b)] are finite-size precursors of the bulk fractional quantum Hall states; see also Section 3 below. 2 Group theoretical analysis of symmetry breaking in unre- stricted Hartree Fock We mention here the case of N = 3 fully spin polarized (Sz = 3/2) electrons in the absence of a magnetic field (B) and for RW = 10 (κ = 1.9095). Fully spin polarized UHF determinants c(cid:13)CMMSE ISBN: 978-84-608-6082-2 hΕ/ ω 0Ω/ω0hΕ/ ω 0LZ08163224EPRJUBHF−E(b)(a)(c)UBHFPRJPRJE 40 80 8 4 0−4−8 0 5 6 7 8−0.4−0.3−0.2−0.1 0 0.3 0.2 0.1 9 0.2 0 0.4 0.6Yannouleas and Landman Figure 4: The UHF case exhibiting breaking of the circular symmetry for N = 3 electrons and total spin projection Sz = 3/2 at RW = 10 and at zero magnetic field. (a-c): real orbitals (modulus square). (d): the corresponding electron density (ED). The choice of the remaining parameters is: ω0 = 5 meV and effective mass m∗ = 0.067me. Distances are in nanometers. The real orbitals are in 10−3 nm−1 and the total ED in 10−4 nm−2. The arrows indicate the spin projection (Sz = 1/2) for each orbital. preserve the total spin, but for this value of RW the lowest in energy UHF solution is one with broken circular symmetry. As it has been mentioned earlier, broken rotational symmetry does not imply no space symmetry, but a lower point-group symmetry [5]. In Fig. 4 we display the UHF symmetry-violating orbitals (a−c) whose energies are (a) 44.801 meV, (b) and (c) 46.546 meV, namely the two orbitals (b) and (c) with the higher energies are degenerate in energy. Overall the BS orbitals (a−c) drastically differ from the orbitals of the independent particle model. In particular, they are associated with specific sites (within the QD) forming an equilateral triangle, and thus they can be described as having the structure of a linear combination of "atomic" (site) orbitals (LCAOs). Such LCAO molecular orbitals (MOs) are familiar in natural molecules, and this analogy supports the term "semi-classical electron (or Wigner) molecules" for characterizing the BS-UHF solutions. We notice here that the LCAO orbitals in Fig. 4 are familiar in Organic Chemistry and are associated with the theoretical description of Carbocyclic Systems, and in particular the molecule C3H3 (cyclopropenyl, see, e.g., Ref. [6]). The important point of course is not the uniqueness or not of the 2D UHF orbitals, but the fact that they transform according to the irreducible representations of specific point groups, leaving both the UHF determinant and the associated electron densities invariant. The electron density (ED) portrayed in Fig. 4(d) remains invariant under certain geo- c(cid:13)CMMSE ISBN: 978-84-608-6082-2 -40040-40040030-30030-30030-301680-40040-40040-40040-40040-40040-40040ED(a)(b)(c)(d)Quantum Wigner molecules and symmetries Table 1: Character table for the cyclic group C3 [ε = exp(2πi/3)] C3 A E(cid:48) E(cid:48)(cid:48) E C3 C2 3 1 1 ε∗ 1 ε 1 1 ε ε∗ metrical symmetry operations, namely those of an unmarked, plane and equilateral triangle. They are: (1) The identity E; (2) The two rotations C3 (rotation by 2π/3) and C2 3 (rota- tion by 4π/3); and (III) The three reflections σI through the three vertical planes, one passing through each vertex of the triangle. These symmetry operations for the unmarked equilateral triangle constitute the elements of the group C3v [6, 7]. v , and σIII v, σII v One of the main applications of group theory in Chemistry is the determination of the eigenfunctions of the Schrodinger equation by simply using symmetry arguments alone. This is achieved by constructing the so-called symmetry-adapted linear combinations (SALCs) of AOs. A widely used tool for constructing SALCs is the projection operator (cid:88) R P µ = nµG χµ(R) R, (1) where R stands for any one of the symmetry operations of the molecule, and χµ(R) are the characters of the µth irreducible representation of the set of R's. (The χµ's are tabulated in the socalled character tables [6, 7].) G denotes the order of the group and nµ the dimension of the representation. The task of finding the SALCs for a set of three 1s-type AOs exhibiting the C3v sym- metry of an equilateral triangle can be simplified, since the pure rotational symmetry by itself (the rotations C3 and C2 3 , and not the reflections σv's through the vertical planes) is sufficient for their determination. Thus one needs to consider the simpler character table of the cyclic group C3 (see Table 1). From Table 1, one sees that the set of the three 1s AOs situated at the vertices of an equilateral triangle spans the two irreducible representations A and E, the latter one consisting of two associted one-dimensional representations. To construct the SALCs, one simply applies the three projection operators P A, P E(cid:48) to one of the original AOs, let's say the φ1, , and P E(cid:48)(cid:48) P Aφ1 ≈ (1) Eφ1 + (1) C3φ1 + (1) C2 3 φ1 = (1)φ1 + (1)φ2 + (1)φ3 = φ1 + φ2 + φ3, φ1 ≈ (1) Eφ1 + (ε) C3φ1 + (ε∗) C2 3 φ1 = φ1 + εφ2 + ε∗φ3, P E(cid:48) (2) (3) c(cid:13)CMMSE ISBN: 978-84-608-6082-2 Yannouleas and Landman P E(cid:48)(cid:48) φ1 ≈ (1) Eφ1 + (ε∗) C3φ1 + (ε) C2 3 φ1 = φ1 + ε∗φ2 + εφ3. (4) The A SALC in Eq. (2) is real. The two E SALCs [Eq. (3) and Eq. (4)], however, are complex functions and do not coincide with the real UHF orbitals. These complex SALCs agree with BS-UHF orbitals obtained in the case of an applied magnetic field. On the other hand, a set of two real and orthogonal SALCs that spans the E representation can be derived fron Eq. (3) and Eq. (4) by simply adding and substracting the two complex ones. This procedure recovers immediately the real UHF orbitals displayed in Fig. 4. 3 Restoration of circular symmetry: Group structure and sequences of magic angular momenta In the previous section, we discussed how the BS-UHF determinants and orbitals describe indeed 2D molecular stuctures (semi-classical Wigner molecules) in close analogy with the case of natural 3D molecules. However, the study of the WMs at the UHF level restricts their description to the intrinsic (nonrotating) frame of reference. Motivated by the case of natural atoms, one can take a subsequent step and address the properties of collectively rotating QWMs in the laboratory frame of reference. As is well known, for natural atoms, this step is achieved by writing the total wave function of the molecule as the product of the electronic and ionic partial wave functions. In the case of the purely electronic or bosonic WMs, however, such a product wave function requires the assumption of complete decoupling between intrinsic and collective degrees of freedom, an assumption that might be justifiable in limiting cases only. Using the BS UHF solutions, this subsequent step can be addressed by using the post- Hartree-Fock method of restoration of broken symmetries [2, 8] via projection (PRJ) tech- niques. In this section, we will use the PRJ approach to illustrate how certain universal proper- ties of the CI (exact) solutions, i.e., the appearance of magic angular momenta in the exact rotational spectra, [2, 9, 10, 11, 12, 13] relate to the symmetry broken UHF solutions. In- deed, we will demonstrate that the magic angular momenta are a direct consequence of the symmetry breaking at the UHF level and that they are determined fully by the molecular symmetries of the UHF determinant. As an illustrative example, we have chosen the relatively simple, but non trivial case, of N = 3 electrons. For B = 0, both the Sz = 1/2 and Sz = 3/2 polarizations can be considered. We start with the Sz = 1/2 polarization, whose BS UHF solution (let's denote it by ↓↑↑(cid:105)) exhibits a breaking of the total spin symmetry in addition to the rotational symmetry. We first proceed with the restoration of the total spin by noticing that ↓↑↑(cid:105) has a point-group symmetry lower than the C3v symmetry of an equilateral triangle. The C3v symmetry, however, can be readily restored by applying the projection operator in Eq. (1) to ↓↑↑(cid:105) and by using the character table of the cyclic C3 group (see Table 1). Then for the c(cid:13)CMMSE ISBN: 978-84-608-6082-2 Quantum Wigner molecules and symmetries intrinsic part of the many-body wave function, one finds two different three-determinantal combinations, namely and intr(γ0) = ↓↑↑(cid:105) + e2πi/3 ↑↓↑(cid:105) + e−2πi/3 ↑↑↓(cid:105), ΦE(cid:48) intr(γ0) = ↓↑↑(cid:105) + e−2πi/3 ↑↓↑(cid:105) + e2πi/3 ↑↑↓(cid:105), ΦE(cid:48)(cid:48) (5) (6) eigenstates of the square of the total spin operator S2 (S =(cid:80)3 where γ0 = 0 denotes the azimuthal angle of the vertex associated with the original spin- down orbital in ↓↑↑(cid:105). We note that the intrinsic wave functions ΦE(cid:48) intr are i=1 si) with quantum number intr and ΦE(cid:48)(cid:48) s(s + 1) = 3/4, (s = 1/2)). This can be verified directly by applying S2 to them. To restore the circular symmetry in the case of a (0, N ) ring arrangement, one applies the projection operator [2, 8, 14], 2πPI ≡ dγ exp[−iγ( L − I)] , (7) (cid:90) 2π 0 where L = (cid:80)N lj is the operator for the total angular momentum. Notice that the operator PI is a direct generalization of the projection operator in Eq. (1) to the case of the continuous cyclic group C∞ [the phases exp(iγI) are the characters of C∞]. j=1 The projected wave function, ΨP RJ , (having both good total spin and angular momen- tum quantum numbers) is of the form, 2πΨP RJ = (cid:90) 2π 0 dγΦE intr(γ)eiγI , (8) where now the intrinsic wave function [given by Eq. (5) or Eq. (6)] has an arbitrary azimuthal orientation γ. We note that, unlike the phenomenological Eckardt-frame model [13] where only a single product term is involved, the PRJ wave function in Eq. (8) is an average over all azimuthal directions of an infinite set of product terms. These terms are formed by multiplying the UHF intrinsic part ΦE intr(γ) with the external rotational wave function exp(iγI) (the latter is properly characterized as "external", since it is an eigenfunction of the total angular momentum L and depends exclusively on the azimuthal coordinate γ). The operator R(2π/3) ≡ exp(−i2π L/3) can be applied onto ΨP RJ in two different ways, intr or the external part exp(iγI). Using Eq. (5) and namely either on the intrinsic part ΦE the property R(2π/3)ΦE(cid:48) intr = exp(−2πi/3)ΦE(cid:48) intr, one finds, R(2π/3)ΨP RJ = exp(−2πi/3)ΨP RJ , from the first alternative, and R(2π/3)ΨP RJ = exp(−2πIi/3)ΨP RJ , (9) (10) c(cid:13)CMMSE ISBN: 978-84-608-6082-2 Yannouleas and Landman from the second alternative. Now if ΨP RJ (cid:54)= 0, the only way that Eqs. (9) and (10) can be simultaneously true is if the condition exp[2π(I − 1)i/3] = 1 is fulfilled. This leads to a first sequence of magic angular momenta associated with total spin s = 1/2, i.e., I = 3k + 1, k = 0,±1,±2,±3, ... (11) Using Eq. (6) for the intrinsic wave function, and following similar steps, one can derive a second sequence of magic angular momenta associated with good total spin s = 1/2, i.e., I = 3k − 1, k = 0,±1,±2,±3, ... (12) In the fully polarized case, the UHF determinant is denoted as ↑↑↑ (cid:105), and it is already an eigenstate of S2 with quantum number s = 3/2. Thus only the rotational symmetry intr(γ0) = ↑↑↑ (cid:105). Since needs to be restored, that is, the intrinsic wave function is simply ΦA intr, the condition for the allowed angular momenta is exp[−2πIi/3] = 1, R(2π/3)ΦA which yields the following magic angular momenta, intr = ΦA I = 3k, k = 0,±1,±2,±3, ... (13) We note that in high magnetic fields only the fully polarized case is relevant and that only angular momenta with k > 0 enter in Eq. (13) (see Refs. [15, 16]). In this case, in the thermodynamic limit, the partial sequence with k = 2q + 1, q = 0, 1, 2, 3, ... is directly related to the odd filling factors ν = 1/(2q + 1) of the fractional quantum Hall effect [via the relation ν = N (N − 1)/(2I)]. This suggests that the observed hierarchy of fractional filling factors in the quantum Hall effect may be viewed as a signature originating from the point group symmetries of the intrinsic wave function Φintr, and thus it is a manifestation of symmetry breaking at the UHF mean-field level. 4 Summary The analysis presented above concerning the relation between hidden symmetries and emer- gent signatures (e.g., magic amgular momenta) in the spectra of symmetry-restored mean- field wave functions applies also in the case of configuration-interaction, exact many body wave functions; see the review in Ref. [2]. The CI method requires large-scale, parallel computations, but it has the advantage of providing benchmark results due to the achieved high quantitative accuracy. We refer to this combination of mean-field and CI analysis as computational microscopy [17]. Recently, we have incorporated in our CI computer codes the option of Dirac-delta contact interactions, in addition to the long-range Coulomb one. Thus we have been able to analyze the wave function anatomy of a few untracold fermionic (6Li) atoms in single c(cid:13)CMMSE ISBN: 978-84-608-6082-2 Quantum Wigner molecules and symmetries and double optical traps, where the formation of quantum Wigner molecules can be asso- ciated with the emergence of Heisenberg antiferromagnetic behavior [17, 18, 19] and the creation of highly entangled states; e.g., the celebrated Bell states for two 6Li atoms. (Such behavior was earlier predicted in the case of strongly repelling electrons in double quantum dots [18].) The unprecedented experimental control of the interparticle interaction (from zero to infinite strength) achieved in the case of a few trapped ultracold atoms is enabling investigations of fundamental physics (such as high Tc and 1D and 2D magnetism) from a bottom-up perspective. In addition, in analogy with the field of 2D semiconductor double and triple QDs, it promises technological applications in the area of quantum information and quantum computing. Our computational-microscopy approach can be used to investi- gate the universal behavior in the strongly correlated regime of both ultracold fermionic or bosonic trapped atoms and confined electrons. Acknowledgements This work is supported by the Office of Basic Energy Sciences of the US Department of Energy (Grant No. FG05- 86ER45234). References [1] C. Yannouleas and U. Landman, Spontaneous symmetry breaking in single and molecular quantum dots, Phys. Rev. Lett. 82 (1999) 5325-5328; (E) Phys. Rev. Lett. 85 (2000) 2220 [2] C. Yannouleas and U. Landman, Symmetry breaking and quantum correlations in finite systems: Studies of quantum dots and ultracold Bose gases, and related nuclear and chemical methods, Rep. Prog. Phys. 70 (2007) 2067-2148 [3] I. Romanovsky, C. Yannouleas and U. Landman, Crystalline Boson phases in harmonic traps: Beyond the Gross-Pitaevskii mean field, Phys. Rev. Lett. 93 (2004) 230405 [4] I. Romanovsky, C. Yannouleas, L.O. Baksmaty and U. Landman, Bosonic molecules in rotating traps, Phys. Rev. Lett. 97 (2006) 090401 [5] C. Yannouleas and U. Landman, Group theoretical analysis of symmetry breaking in two-dimensional quantum dots, Phys. Rev. B 68 (2003) 035325 [6] F.A. Cotton, Chemical Applications of Group Theory, Wiley, New York, 1990 [7] A.B. Wolbarst, Symmetry and Quantum Systems, Van Nostrand Reinold, New York, 1977 c(cid:13)CMMSE ISBN: 978-84-608-6082-2 Yannouleas and Landman [8] P. Ring and P. Schuck, The Nuclear Many-body Problem, Springer-Verlag, New York, 1980 [9] S.M. Girvin and T. Jach, Interacting electrons in two-dimensional Landau levels: Results for small clusters, Phys. Rev. B 28 (1983) 4506-4509 [10] P.A. Maksym and T. Chakraborty, Quantum dots in a magnetic-field: Role of electron-electron interactions, Phys. Rev. Lett. 65 (1990) 108-111 [11] W.Y. Ruan, Y.Y. Liu, C.G. Bao and Z.Q. Zhang, Origin of magic angular mo- menta in few-electron quantum dots, Phys. Rev. B 51 (1995) 7942-7945 [12] T. Seki, Y. Kuramoto and T. Nishino, Origin of magic angular momentum in a quantum dot under strong magnetic field, J. Phys. Soc. Jpn. 65 (1996) 3945-3951 [13] P.A. Maksym, Eckardt frame theory of interacting electrons in quantum dots, Phys. Rev. B 53 (1996) 10871-10886 [14] C. Yannouleas and U. Landman, Strongly correlated wave functions for artificial atoms and molecules, J. Phys.: Condens. Matter 14 (2002) L591-L598 [15] C. Yannouleas and U. Landman, Trial wave functions with long-range coulomb correlations for two-dimensional N-electron systems in high magnetic fields, Phys. Rev. B 66 (2002) 115315 [16] C. Yannouleas and U. Landman, Quantum dots in high-magnetic fields: Rotating- Wigner-molecule versus composite-fermion approach, Phys. Rev. B 68 (2003) 035326 [17] B.B. Brandt, C. Yannouleas and U. Landman, Double-well ultracold-fermions computational microscopy: Wave-function anatomy of attractive-pairing and Wigner- molecule entanglement and natural orbitals, Nano Lett. 15 (2015) 7105 [18] Ying Li, C. Yannouleas and U. Landman, Artificial quantum-dot helium molecules: Electronic spectra, spin structures and Heisenberg clusters, Phys. Rev. B 80 (2009) 045326 [19] S. Murmann, F. Deuretzbacher, G. Zurn, J. Bjerlin, S.M. Reimann, L. Santos, T. Lompe and S. Jochim, Antiferromagnetic Heisenberg spin chain of a few cold atoms in a one-dimensional trap, Phys. Rev. Lett. 115 (2015) 215301 c(cid:13)CMMSE ISBN: 978-84-608-6082-2 

1804.09556

1

1804

Determination of the spin-lifetime anisotropy in graphene using oblique spin precession

We determine the spin-lifetime anisotropy of spin-polarized carriers in graphene. In contrast to prior approaches, our method does not require large out-of-plane magnetic fields and thus it is reliable for both low- and high-carrier densities. We first determine the in-plane spin lifetime by conventional spin precession measurements with magnetic fields perpendicular to the graphene plane. Then, to evaluate the out-of-plane spin lifetime, we implement spin precession measurements under oblique magnetic fields that generate an out-of-plane spin population. We find that the spin-lifetime anisotropy of graphene on silicon oxide is independent of carrier density and temperature down to 150 K, and much weaker than previously reported. Indeed, within the experimental uncertainty, the spin relaxation is isotropic. Altogether with the gate dependence of the spin lifetime, this indicates that the spin relaxation is driven by magnetic impurities or random spin-orbit or gauge fields.

cond-mat.mes-hall

cond-mat

ARTICLE Received 12 Nov 2015 Accepted 28 Mar 2016 Published 9 May 2016 OPEN Determination of the spin-lifetime anisotropy in graphene using oblique spin precession Bart Raes1, Jeroen E. Scheerder2, Marius V. Costache1, Fre´de´ric Bonell1, Juan F. Sierra1, Jo Cuppens1, Joris Van de Vondel2 & Sergio O. Valenzuela1,3 DOI: 10.1038/ncomms11444 We determine the spin-lifetime anisotropy of spin-polarized carriers in graphene. In contrast to prior approaches, our method does not require large out-of-plane magnetic fields and thus it is reliable for both low- and high-carrier densities. We first determine the in-plane spin lifetime by conventional spin precession measurements with magnetic fields perpendicular to the graphene plane. Then, to evaluate the out-of-plane spin lifetime, we implement spin precession measurements under oblique magnetic fields that generate an out-of-plane spin population. We find that the spin-lifetime anisotropy of graphene on silicon oxide is independent of carrier density and temperature down to 150 K, and much weaker than previously reported. Indeed, within the experimental uncertainty, the spin relaxation is isotropic. Altogether with the gate dependence of the spin lifetime, this indicates that the spin relaxation is driven by magnetic impurities or random spin-orbit or gauge fields. 1 Catalan Institute of Nanoscience and Nanotechnology (ICN2), CSIC and The Barcelona Institute of Science and Technology, Campus UAB, Barcelona 08193, Spain. 2 INPAC-Institute for Nanoscale Physics and Chemistry, Department of Physics and Astronomy, KU Leuven, Celestijnenlaan 200D, B-3001 Leuven, Belgium. 3 Institucio´ Catalana de Recerca i Estudis Avanc¸ats (ICREA), Barcelona 08070, Spain. Correspondence and requests for materials should be addressed to B.R. (email: bart.raes@icn.cat) or to S.O.V. (email: SOV@icrea.cat). NATURE COMMUNICATIONS 7:11444 DOI: 10.1038/ncomms11444 www.nature.com/naturecommunications 1 ARTICLE NATURE COMMUNICATIONS DOI: 10.1038/ncomms11444 Identifying the main microscopic process for spin relaxation in graphene stands as one of the most fascinating puzzles for the graphene and spintronics communities1–10. Conventional Elliot–Yafet and Dyakonov–Perel relaxation mechanisms, which present opposite scalings between the spin lifetime ts and the momentum scattering time tp, have often been utilized to gain insight into the spin-relaxation processes in graphene, but this type of analysis has yielded contradictory results1,3. For single- layer graphene, there are reports that are consistent with the Elliot–Yafet scaling, the Dyakonov–Perel scaling or even a combination of both11–14. For bilayer and multilayer graphene, the situation is similarly confusing11,15,16. To study the scaling, tp is tuned by either changing the carrier density n, by means of an external gate, or by introducing adatoms or molecules onto the graphene layer17,18. Nevertheless, this tuning will unavoidably lead to the modification of other parameters, relevant for spin transport. For example, in the case of spin-orbit-dominated spin relaxation, the nature and strength of the spin-orbit field can be energy (or charge density n) dependent, while the introduction of adatoms can actively mask or alter pre-existing spin-relaxation processes in an unintended way. In this context, a fingerprint of the spin-orbit mechanism that can be determined as a function of n and the type and density of the adatoms can provide valuable information about the induced modifications. The spin-lifetime anisotropy, which can be quantified by the ratio between the spin lifetimes for perpendi- cular and parallel spin components to the graphene plane z  ts?=tsjj, is precisely such a fingerprint. This anisotropy is determined by the preferential direction of the spin-orbit fields that cause the spin relaxation1. For spin-orbit fields preferentially in the graphene plane (for example, due to Rashba-type interaction), we expect zo1 (refs 5,19,20). For spin-orbit fields out-of-plane (for example, from ripples5 or flexural distortions21), we expect z41. There is no general relation for adatoms; depending on the element, they may induce spin-orbit fields with different directions and lead to specific signatures on the dependence of z with energy22. For example, in the case of fluorine adatoms, z is predicted to be equal to 0.5 for negative energies, but to depart markedly from this value for positive energies23. Finally, the presence of local magnetic moments with random orientation, originating from magnetic impurities or hydrogen adatoms, can overshadow the spin–orbit interaction, yielding z¼ 1 (refs 24,25). Despite their inherent interest, measurements of the spin- lifetime anisotropy are scarce26,27. Besides, the method to quantify the anisotropy requires intense out-of-plane magnetic fields B>41 T (perpendicular to the graphene plane) and, therefore, it is expected to be useful for sufficiently large n only, owing to the large magnetoresistive effects that are present in graphene at low carrier densities (see ref. 27 for further details). The actual value of n beyond which the method would be suitable is sample dependent and, presumably, increases with the mobility of graphene. it Here, we demonstrate an approach that overcomes the above limitation. The concept is based on spin precession measurements under oblique magnetic fields that generate an out-of-plane spin population, which is further used to evaluate the out-of-plane spin lifetime. The key of the method is to focus on the non-precessing spin component along the magnetic field, which will relax faster than the in-plane spin component, if zo1, or slower than the in-plane spin component, if z41. We fabricate graphene devices with high mobility on silicon oxide substrates, however, the concept is general and can be implemented in any system that is susceptible to an anisotropic response. Our experiments demonstrate that the spin-relaxation anisotropy of graphene on silicon oxide is independent of carrier density and temperature down to 150 K, and much weaker than previously reported26. Altogether with the gate dependence of the spin lifetime, this indicates that the spin relaxation is driven either by random magnetic impurities24,25 or by random spin-orbit fields or gauge fields1,20. These findings open the way for systematic anisotropy studies with tailored impurities and on different substrates. This information is crucial to find a route to increase the spin lifetime in graphene towards its intrinsic limit and, as such, has important implications for both fundamental science and technological applications. tsjj spin devices, Results Measurement concept. Our measurement scheme (Fig. 1) allows us to determine z at low magnetic fields (BB0.1 T), circumventing spurious magnetoresistive phenomena. In our is first approach, based on non-local determined using conventional spin precession measurements, as shown schematically in Fig. 1a. This is performed by applying an out-of-plane magnetic field B>, which causes the spins to precess exclusively in-plane as they diffuse from the injector (F1) towards the detector (F2). Because the magnetic field required to generate the spin precession is relatively weak (typically in the range of 0.1 T), the magnetizations of the injector and detector are assumed to remain in-plane over the whole-field range, due to magnetic shape anisotropy. The measurement of ts? (Fig. 1b) relies on the fact that an oblique magnetic field28, characterized by an angle b (Fig. 1b, inset), causes the spins to precess out-of-plane as they diffuse towards F2. Therefore, the precession dynamics for 0obo90° becomes sensitive to both tsjj and ts?. As B increases, the spin component perpendicular to the magnetic field dephases due to diffusive broadening. Eventually, for B larger than the dephasing field Bd, only the component parallel to the field contributes to the non-local signal, which is picked up at the detector electrode. Dephasing greatly simplifies the data interpretation; the effective spin lifetime of the resulting component parallel to the field, tsb, follows a simple relationship with z (see the 'Methods' section): ð1Þ  ¼ cos2 bð Þþ 1 z  sin2 bð Þ tsb tsjj 1 : By first measuring tsjj with the standard spin precession measurements (Fig. 1a) and then studying the dephased non-local signal as a function of b, we extract z using equation 1. Device design and characterization. In our devices (Fig. 2a), we is exfoliated onto a pþ þ Si/SiO2 (440 nm) use graphene that substrate from highly oriented pyrolytic graphite. Two inner ferromagnetic Co electrodes and two outer normal Pd electrodes contacting the graphene flake are fabricated using a single-electron- beam lithography step and shadow evaporation29 (see Supplementary Fig. 1). Before metallization, an amorphous carbon layer is created between all contacts and the graphene flake by electron-beam overexposure of the contact area30,31. Shadow evaporation minimizes contamination from multiple lithographic steps. The amorphous carbon creates a resistive interface between the metals and the graphene (typically of about 10 kO) that suppresses contact-induced spin relaxation and the conductivity mismatch problem32, and that (associated with Co) is highly spin-polarized30,33 (see the 'Methods' section for further details). We first characterise the graphene charge transport properties by means of standard four-terminal local measurements, from which we electron/hole mobility an m¼ 1.7 104 cm2 V 1 s 1 carrier density n0¼ 1.5 1011 cm 2, which sets the region where electron-hole puddles result in an inhomogeneous carrier density (see the average and a estimate residual 2 NATURE COMMUNICATIONS 7:11444 DOI: 10.1038/ncomms11444 www.nature.com/naturecommunications NATURE COMMUNICATIONS DOI: 10.1038/ncomms11444 ARTICLE a I F1 B⊥ ♢ b I F1 B ♢ F2 F2 Figure 1 Spin precession and spin-lifetime anisotropy measurement principle. (a) Conventional spin precession experiment in which the magnetic field B> is applied perpendicular to the graphene plane (purple arrow). A charge current (straight black arrows) through one of the ferromagnetic electrodes (F1) injects spins having an orientation parallel to the magnetization direction, which is fixed along the easy (that is, long) axis of the injector electrode. The injected spins (red arrows) undergo Larmor precession around B> while diffusing towards the detector electrode (F2). The precession angle f changes with the strength of B>, thus modulating the detected signal at F2. In this case, the precession is exclusively in the graphene plane and only sensitive to the parallel relaxation time (see inset). (b) Schematic illustration of the oblique-spin precession experiment proposed in this article. The magnetic field B is applied in a plane that contains the easy axis of the ferromagnetic electrodes and that is perpendicular to the substrate. For an oblique field, that is, ba0, 90° (see inset), the spins precess out-of-plane as they diffuse towards F2. In this situation, the effective spin lifetime is sensitive to both parallel and perpendicular spin lifetimes, tsjj and ts?, and the spin-relaxation anisotropy can be experimentally obtained. a E2 E1 E4 E3 Vnl L I E1 E2 E3 E4 b ) 0 = B ( q s R / q s R c ) Ω ( l n R 1.1 1.0 0.0 0.2 0.4 B⊥ (T) 0.6 0.5 0.0 –0.5 –10 0 10 B (mT) Figure 2 Device schematics and spin transport. (a) Schematic drawing of the lateral non-local spin device geometry showing both the outer normal metallic electrodes (E1 and E4), and the inner ferromagnetic injector (E2) and detector (E3) electrodes. Wiring is shown in the non-local configuration, in which a current I is applied between E1 and E2 and the non-local voltage Vnl is measured between E3 and E4. An optical image of the non-local device used for the measurements shown in this paper is also shown. L¼ 11 mm. (b) Normalized graphene square resistance Rsq as a function of B. Vg VCNP¼ 2.5, 42.5, 22.5 V (top to bottom). The magnetoresistance is largest for Vg VCNP. (c) Non-local resistance RnlVnl/I as a function of a magnetic field applied B in-plane for up and down magnetic field sweeps at gate voltages Vg such that Vg VCNP¼ 25 V, with VCNP the position of the CNP. T¼ 300 K and I¼ 10 mA. 'Methods' section, Supplementary Fig. 2 and Supplementary Note 1). Figure 2b shows the normalized graphene resistance as a function of the strength of a perpendicular magnetic field B>, in the low field range, for various gate voltages Vg applied to the pþ þ Si substrate. The magnetoresistance presents a peak at the charge-neutrality point (CNP)34, Vg¼ VCNP, but remains below 3% for all gate voltages when B>r0.2 T (for further details see Supplementary Fig. 3 and Supplementary Note 2). The spin- transport properties in the graphene plane are then investigated using the non-local configuration, in which the current path is separated from the voltage detection circuit as shown in Fig. 2a. The current I is driven between E1 and E2 and the non-local voltage Vnl is measured between E3 and E4. Figure 2c shows typical non-local resistance RnlVnl/I measurements as a function of the applied in-plane magnetic field. Sharp steps from positive- to-negative Rnl are observed at ±7 and ±15 mT, when the relative magnetization of the Co electrodes switches from parallel (kk,mm) (km) configuration, or vice versa. #" Figure 3a shows the dependence of DRnl¼R nl on Vg; the observed variation nearby the CNP is associated with a faster spin relaxation (see below) and a resistive, but not tunnelling, interface between the metal and graphene4,35. to antiparallel "" nl R Conventional non-local spin precession. A typical precession curve for magnetic fields perpendicular to the graphene plane is shown in Fig. 3b. We use these measurements to evaluate tsjj and p the spin-diffusion coefficient Ds as a function of Vg. To this end, we fit the data to the solution of the Bloch equations4,36. The results for tsjj and Ds are presented in Fig. 3c, altogether with the spin-relaxation length lsjj¼ . The spin-relaxation times and lengths in our device are as large as tsjj¼0:45 ns and lsjj¼5:8 mm, respectively, and increase away from the CNP. These values compare well with state-of-the-art studies of spin transport in non-encapsulated graphene on h-BN14, and on suspended graphene37,38. ffiffiffiffiffiffiffiffiffiffiffi tsjjDs the position of To avoid magnetoresistive phenomena, shown in Fig. 2b, we note that complete dephasing of the spin component perpendi- ffiffiffi cular to the magnetic field, at the detector electrode, should occur at Bdt0.1 T. This condition is met at a p large-enough injector-detector separation L. For typical graphene devices it is sufficient that L  ls, where ls is the graphene 2 ffiffiffi spin-relaxation length (see Supplementary Note 3). The separa- tion of the Co electrodes in the device of Fig. 2, L¼ 11 mm, p ensures that this condition is fulfilled, as L is larger than the ls¼8:2 mm. As expected, Fig. 3b demon- maximum value of 2 strates that diffusive broadening completely suppresses the observation of spin precession at BdB0.1 T. Using the fitting in Fig. 3b, we also evaluate the tilting angle g of the electrode NATURE COMMUNICATIONS 7:11444 DOI: 10.1038/ncomms11444 www.nature.com/naturecommunications 3 ARTICLE NATURE COMMUNICATIONS DOI: 10.1038/ncomms11444 a ) Ω ( l n R Δ 0.6 0.4 0.2 0.0 b ) Ω ( l n R T = 300 K I = 10 μA Vg – VCNP = –17.5 V 0.2 0.1 0.0 T = 300 K I = 10 μA c ) s n ( s τ ) 1 – s 2 m ( s D ) m μ ( s ♮ 0.4 0.2 0.08 0.04 6 4 2 T=150 K T=300 K –20 20 0 Vg – VCNP (V) 40 –0.1 –0.2 –0.1 0.0 0.1 0.2 –20 0 20 40 60 B⊥ (T) Vg – VCNP (V) Figure 3 Spin precession under perpendicular magnetic field. (a) Non-local resistance Rnl change, DRnl¼R"" nl , as a function of gate voltage Vg. The values of DRnl are extracted from Rnl versus in-plane magnetic field measurements and reflect the change in the signal when the relative magnetization of p ffiffiffiffiffiffiffiffiffiffiffi the ferromagnetic electrodes switches from parallel (kk,mm) to antiparallel (km) configuration. The line is a guide to the eye. (b) Conventional spin precession measurements with perpendicular magnetic field B> for up (open symbols) and down (solid symbols) sweeps. Complete spin dephasing occurs for field strengths larger than BdB0.1 T. (c) In-plane spin-relaxation time tsjj, spin-diffusion coefficient Ds and spin-relaxation length lsjj¼ , obtained from the spin precession measurements in b. The dashed lines are a guide to the eye. The measurements in a,b are performed in the non-local configuration. The error bars are smaller than the symbol size. The errors correspond to the standard error of mean obtained from the fitting for Ds and lsjj, and the propagated error for tsjj. nl R#" tsjjDs magnetization out of the graphene plane (see Supplementary Fig. 4 and Supplementary Note 4). The results were verified with anisotropic magnetoresistance measurements in a Co wire (Supplementary Fig. 5 and Supplementary Note 5). We find that g remains below B5° for B Bd, which guarantees that the extracted spin lifetimes are a reliable estimate for tsjj. Finally, the data discussed so far was acquired at 300 K but the same conclusions apply to the measurements at 150 K. As observed in Fig. 3c, only weak variations are observed as the temperature is decreased. a 0.2 0.1 ) Ω ( l n R 0.0 30° 45° 60° 75° T = 300 K Vg – VCNP = –17.5 V Oblique non-local spin precession. Having determined the in-plane spin-transport properties and demonstrated the high quality of the graphene sample, we proceed with the oblique- precession experiments to extract z. Figures 4 and 5 present the main results of our work. Figure 4a shows spin precession measurements at T¼ 300 K for a representative set of b values. As before, it is observed that the precessional motion dephases at BdB0.1 T. For B4Bd, Rnl is determined by the remanent non- precessional spin component that lies along the magnetic field Þ, depends on b and is direction. Its magnitude, Rb nearly constant with increasing B. Figure 4b shows Rb nl versus b at fixed B¼ 175 mT (marked by the vertical line in Fig. 4a) for the indicated values of Vg. For an isotropic system (z¼ 1), nl¼Rb;iso Þ and b*¼ b g(b, B) Rb is the angle between the magnetization of the electrodes and the magnetic field, taking into consideration the small tilting g. The factor cos2(b*) accounts for the projection of the injected spins along the direction of the magnetic field and the subsequent projection along the direction of the detector magnetization. nl  Rnl B4Bd nl¼Rnl B¼0 Þ, where R0 nlcos2 bð nl ¼R0 ð ð Determination of the spin-lifetime anisotropy. For the general case of an anisotropic system, we determined Rb nl by solving the Bloch equations including both tsjj and ts?, with z not necessarily equal to unity (see the 'Methods' section). We found that s ffiffiffiffiffiffiffiffiffiffiffiffiffi ! # p ffiffiffiffiffiffiffiffiffiffiffiffiffi Þ f z; bð " exp L ljj nl ¼ Rb 1 f z; bð Þ 1 Rb;iso nl ; ð2Þ where f z; bð straightforward to verify that Rb Þ¼tsb=tsjj is given by equation 1. For z¼ 1, it is , regardless of the value of b. nl¼Rb;iso nl 90° 0.10 B (T) 0.15 0.20 T = 300 K 0.00 0.05 b ) Ω ( l n R 0.3 0.2 0.1 0.0 47.5 V 37.5 V 17.5 V 7.5 V –20 0 20 60 80 100 40 ♢ (deg) Figure 4 Spin precession measurements under oblique magnetic fields. (a) Representative subset of experimental spin precession curves for b¼ 30°, 45°, 60°, 75°, 90°. The precession data are acquired at T¼ 300 K and Vg VCNP¼ 17.5 V, using an injector current of I¼ 10 mA, after preparing a parallel configuration of the electrodes. The horizontal dashed line is the non-local resistance at B¼ 0, Rnl(B¼ 0), which coincides with Rnl at b¼ 0° in the parallel configuration. (b) Angular dependence of Rnl at fixed magnetic field B¼ 175 mT4Bd for representative Vg. This magnetic field is shown with a vertical dashed line in a; Vg VCNP¼ 47.5, 37.5, 17.5, 7.5 V. 4 NATURE COMMUNICATIONS 7:11444 DOI: 10.1038/ncomms11444 www.nature.com/naturecommunications NATURE COMMUNICATIONS DOI: 10.1038/ncomms11444 ARTICLE a l n 0 R / 1.0 0.8 0.6 Vg – VCNP = –47.5 V Vg – VCNP = –37.5 V Vg – VCNP = –17.5 V 1.8 1.5 1.2 l n R 0.4 0.2 0.0 0.0 0.8 0.5 0.2 T = 300 K 0.2 0.4 0.6 0.8 1.0 cos2 ♢* b 2.0 1.5 ζ 1.0 0.5 0.0 T = 150 K T = 300 K Spin-orbit field ⊥ graphene plane Spin-orbit field graphene plane –20 –10 0 10 20 30 40 50 Vg – VCNP (V) Figure 5 Spin-lifetime anisotropy ratio f. (a) Representative data from Fig. 4b normalized to 1, as a function of cos2 b*, where b*¼ b g(b, B). The grey nl for the indicated values of z. The black straight line corresponds to z¼ 1. (b) Extracted anisotropy ratio z as a function of Vg for T¼ 300 K lines represent Rb (solid circles) and T¼ 150 K (open squares). The blue solid lines in a,b mark the expected z¼ 0.5 for in-plane spin-orbit fields. The grey area indicates 0.9ozo1.03. The error bars correspond to the standard error of mean from the fits in a. The main source of error is the noise of the measurements; propagated uncertainties in lsjj, b and g produce a marginal contribution to the error in z. The presented results are all obtained from the same sample. nl=R0 nl=R0 nl cos2 bð Þ and, therefore, Rb Figure 5a shows the data in Fig. 4b normalized to 1, Rb nl, as a function of cos2(b*). The representation of Fig. 5a helps visualize any deviation from the isotropic case. As discussed above, for z¼ 1, nl¼R0 Rb nl versus cos2(b*) results in a straight line. According to equation 2, the response lies above the straight line for z41 and below it for zo1. In Fig. 5a, we represent the predicted response for specific values of z from 0.2 to 1.8. We find that the experimental results (open symbols) are in excellent agreement with a straight line behaviour, thus zB1. The data is well-enclosed by the lines that correspond to z¼ 0.8 and z¼ 1.2, thereby defining rough lower and upper limits for z. Figure 5b shows z as a function of Vg for T¼ 150 K and 300 K by fitting the data to equation 2. All of the fitted values of z fall between B0.9 and B1.03 with no apparent dependence on either Vg or T. forces the electrodes to align to the field, A comparison of the above results with those obtained with the methods shown in refs 26,27, can be achieved by applying a large the magnetization perpendicular magnetic field that direction of therefore enabling the injection of spins perpendicular to the graphene plane. When complete rotation of the magnetization is obtained, Rnl should saturate to a constant value from which ts? can be extracted. Unfortunately, we found that such a method could not be implemented in our sample, even for the largest carrier densities n that we have studied. Indeed, although an apparent saturation of Rnl is observed at BB1 T, further increase of the magnetic field demonstrates that Rnl is actually not independent of B, presenting a monotonous decrease, no matter the value of n. In the Supplementary Note 2 and Supplementary Fig. 3, we present the magnetoresistance effects in the graphene, while in Supplementary Note 6 and Supplementary Fig. 6 we discuss the results with perpendicularly magnetized electrodes and a criterion to evaluate the method applicability. regarding the dominant mechanism of Discussion Considering the obtained anisotropy (Fig. 5b) and the gate dependence of ts and Ds (Fig. 3c), we can draw several conclusions spin relaxation in our graphene samples. The temperature indepen- dence of all of the involved parameters demonstrates the small relevance of phonon scattering on the spin relaxation. In contradiction to the experiments, phonons are also predicted to larger than tens of nanoseconds1,21,39. lead to z41 and to ts Because zB1, we can also rule out mechanisms associated with spin-orbit fields that are exclusively in-plane or out-of-plane of a decreasing the graphene. Rashba spin–orbit coupling caused by (heavy) adatoms, such as gold, or the substrate, and the spin–orbit coupling caused by hydrogen-like adatoms (excluding magnetic moments)23,40 predict spin lifetime when approaching the CNP, as observed in the experiments (Fig. 3c). In the former, the CNP is due to a new spin-pseudospin entanglement mechanism specific to graphene20,41. the faster relaxation originates from resonances close to the CNP that enhance the effect of the spin–orbit coupling23. However, in both cases z¼ 0.5 should hold strictly, in disagreement with our results. the faster relaxation at In the latter, intrinsic spin–orbit The anisotropy can be weaker for other adatoms that induce resonances, when the coupling becomes important at off-resonance energies. Such is the case for fluor-like adatoms, although here z is expected to depend on energy and significantly deviate from 0.5 only at positive energies. In addition, spin-relaxation mechanisms associated with spin–orbit coupling from hydrogen- and fluorine-like adatoms predict ts to be larger than 100 ns for reasonable adatom coverage and therefore are unable to explain the ns-time scale observed in the experiments23. Besides Rashba-like spin-orbit fields, adatoms can introduce intrinsic-like spin-orbit fields, depending on the element under consideration. It is in principle possible to tailor any value of z for random distributions of specific adatoms. Moreover, increasing gauge fields, associated with strain, topological defects, or ripples can also result in a transition from zo1 to z41 and mask a z¼ 0.5 relation5. Their presence combined with spin-pseudospin entanglement could therefore lead to zB1 and, at the same time, explain the magnitude and carrier-density dependence of ts. Although it appears unlikely to find exactly z¼ 1 as in Fig. 5, experiments in graphene encapsulated with h-BN, which presumably reduces strain and contamination, are in favour of this interpretation27. In these experiments, the reported z is of the order of 0.7 at large n. However, measurements as a function of n and at low magnetic fields are necessary to discard possible artefacts (Supplementary Note 6). As discussed above, resonant scattering due to hydrogen-like adatoms can explain the observed dependence of ts on carrier density but not its magnitude. The reason is that, even though the resonances enhance the spin-flip scattering, the spin–orbit interaction is not effective because the resonance width (5 meV) is larger than the spin-orbit energy (1 meV) (ref. 23). Combining resonant scattering with local magnetic moments, which introduce the much larger exchange energy Jj j 0:4 eV, can lead to much shorter spin lifetimes while preserving the carrier- density dependence24. Calculations on hydrogenated graphene NATURE COMMUNICATIONS 7:11444 DOI: 10.1038/ncomms11444 www.nature.com/naturecommunications 5 ARTICLE s that capture the effect of magnetic moments predict spin lifetimes in the range of the experimental results with about 1 p.p.m. of hydrogen24,25. Because local magnetic moments are paramagnetic at such low concentrations, no preferential spin orientation is expected and thus z¼ 1. Within this model, chemisorbed hydrogen produces two peaks in t 1 : one above and one below the Dirac point with a separation of about 100 meV. To reproduce the experimental results, a smearing of the peaks of 110 meV is introduced, which is justified by the presence of charge fluctuations24,25. Spin anisotropy measurements at low temperatures in encapsulated graphene would therefore be the ultimate test for this model. Charge fluctuations should be much smaller in that case, in the range of 10 meV, allowing to resolve the two peaks and leading to a much sharper change in ts as a function of carrier density, and a maximum ts at the Dirac point, while z should remain equal to 1. These predictions could also be tested in bilayer graphene, where the energy broadening would be greatly reduced due to its larger density of states42. We have thus demonstrated spin-lifetime anisotropy measure- ments in graphene and discussed them in light of current theoretical knowledge. We used a measurement technique that provides reliable information on spin dynamics at low magnetic fields, and in a broad carrier-density range that was not accessible before. We show that only a very limited number of models can explain our results, and we provide a route based on our methods to discriminate between them using graphene spintronic devices that are within reach of the current state of the art. In addition, the microscopic properties of the graphene used in devices originating from different laboratories are not necessarily equivalent, owing to the graphite source or the processing steps that have been used. Therefore, it is plausible that experimental results in one research group might not be directly reproduced in another1,11–14,17,18. This underscores the importance of developing advanced spin-transport characterization techniques and the systematic implementation using samples of different origin. Spin-relaxation anisotropy measurements on specific substrates and with a controlled number of deposited adatoms will be crucial to increase the spin lifetime towards the theoretical limit, to find ways of controlling the spin lifetime, and to ultimately develop unprecedented approaches for the emergence of spin-based information processing protocols relying on graphene2,43,44. Methods Device fabrication. Graphene flakes are obtained by mechanically exfoliating highly oriented pyrolytic graphite (SPI Supplies) onto a p-doped Si substrate covered with 440 nm of SiO2. The thickness of the SiO2 layer is chosen to have optimum optical contrast between single- and multi-layer graphene flakes, which were discriminated by optical means after contrast calibration with Raman measurements. The width of the flake in the device shown in Fig. 2 is about wB1 to 1.5 mm. An amorphous carbon (aC) interface is created between all contacts and the graphene flake by electron-beam (e-beam)-induced deposition before the fabrication of the contacts30,31. The aC deposition allows us to obtain large spin-signals by suppressing the conductivity mismatch problem, and the contact-induced spin relaxation (spin-sink effect); it is done by an e-beam overexposure of the contact area, with a dose about 30 times the typical working dose of e-beam resists. We define the contact electrodes in our devices in a single e-beam lithography step using shadow evaporation to minimize processing contamination29. This results in devices with relatively large mobilities, which are in the range of 1 m2 V 1 s 1 and thus compare well with similar devices on h-BN14. The shadow mask and evaporation steps are shown schematically in Supplementary Fig. 1. The mask is made from a 300-nm thick resist (MMA/PMMA) layer using MIBK:IPA (1:3) developer (the resists and the developer are from Microchem). All materials were deposited by e-beam evaporation in a chamber with a base pressure of B10 8 Torr. We first deposit the outer electrodes of Ti/Pd by angle deposition. Selecting an angle of ±45° from the normal to the substrate assures that no image of the lithographically produced lines reserved for the Co electrodes are deposited during the evaporation of Ti/Pd; indeed, Ti and Pd deposit onto the sidewalls of the lithography mask and are later on removed by lift-off29,45. Subsequently, Co is deposited under normal incidence. Co is in direct contact with the aC/graphene interface only for the inner electrodes, while for the outer electrodes the Ti/Pd layer is deposited in between. An undercut produced in the MMA layer assures that the deposited Co area on top of the outer Ti/Pd contacts is smaller by a factor of a few per cent. The nominal width wo and deposited thicknesses dTi and dPd NATURE COMMUNICATIONS DOI: 10.1038/ncomms11444 of the identical outer Ti/Pd electrodes are wo¼ 1,500 nm, and dTi¼ 5 nm and E3¼ 140 nm and dPd¼ 10 nm. The inner Co contacts have widths wi thickness dCo¼ 30 nm. The centre to centre distance between the Co contacts is L¼ 11 mm; their difference in width sets distinct coercive field strengths that allow us to control the relative orientation of their magnetization by an external magnetic field. E2¼ 150 nm and wi Electrical characterization. The devices are wired to a chip carrier that is placed in a variable temperature cryostat. Before start measuring, the sample space is first flushed with helium gas and then pumped, reaching an actual pressure of o6.5 10 4 Torr at 300 K and 1.5 10 5 Torr at 150 K. We characterise the graphene charge transport properties by means of three- and four-terminal measurements. The contact resistances are of the order of 10 kO or larger, as determined by three-terminal measurements. The devices are homogeneous and we have not observed any significant shifts of the CNP in the different regions between each pair of contacts. For the four-terminal local measurements a current I is driven between the outer Pd electrodes, E1 and E4, and the voltage measured between the inner Co electrodes, E2 and E3 (Fig. 2a). The characterization results are shown in Supplementary Fig. 2 and briefly discussed in Supplementary Note 1. Bloch-diffusion model including spin-lifetime anisotropy. The contribution of the residual spin component parallel to the field, sBjj (sb in the main text), to the non-local voltage is derived within the spin-diffusion model including spin anisotropy, which is quantified by the anisotropy ratio z. In the isotropic limit (z¼ 1), this model has been successfully used to describe spin precession phenomena in lateral devices46, and in graphene when only the in-plane spin lifetime is relevant. Þ¼ðsx; sBjj ; sB?Þ, on application of a homogeneous dc magnetic field B¼ (0, B, 0) can be described within a rotated cartesian axis system characterized by the unit vectors ð^ex; ^eBjj ; ^eB?Þ (see Supplementary Fig. 7) by, The spatio-temporal evolution of the spin density, s x; t ð s s @s @t
s; ¼ Dsr2sþ gcsB t 1 ð3Þ where Ds is a scalar matrix with all diagonal entries equal to the spin-diffusion is a symmetric (3 3) matrix describing spin relaxation constant Ds, while t 1 with entries determined by the parallel and perpendicular spin lifetimes, tsjj and ts?, and the angle b of the applied magnetic field B. The injected spins feel a torque N due to the presence of the magnetic field, N¼ gcs B, which results in a precessional evolution of the spin density. The constant pre-factor, gc¼ gmB/ is the Bohr magneton and g is the so-called g-factor. In the case of exfoliated graphene, we can consider the g-factor to be equal to the free-electron value and to be field independent and isotropic. : , is the gyromagnetic ratio of the carriers, where mB In the limiting case in which the precessional motion is fully dephased, the spin density perpendicular to the field direction is suppressed, and the spatio-temporal evolution of sBjj , governed by equation 3, simplifies to, @2sBjj @x2 ¼ sBjj tsb @sBjj @t ð4Þ  þ Ds  with sb ¼ 1 t 1 tsjj cos2 bð Þþ 1 z sin2 bð Þ ¼ 1 tsjj f z; bð Þ 1: ð5Þ The above equation is equivalent to that governing the spin diffusion when only one spin lifetime is relevant, which here is given by the effective spin lifetime tsb that is dependent on the direction of the field. The solution to this equation is well- known36,46; the non-local voltage Vnl at the detector is, q ffiffiffiffiffiffiffi r ffiffiffiffiffiffi Vnl ¼ aI tsb Ds e L2 tsb Ds cos b gi ½ ð Þ Š cos b gd ½ ð Þ Š; ð6Þ where L indicates the distance between injector and detector and I the magnitude of the injector current. The factor cos(b gi) accounts for the projection of the injected spins along the direction of the magnetic field, corrected by a small tilting of the magnetization of the injector electrode, which is given by the angle gi. Similarly, the detector picks up the projection of the spin density along its own magnetization, whose orientation is corrected by an angle gd, leading to the cos(b gd) factor (see also Supplementary Notes 4 and 5 and Supplementary Figs 4 and 5). The factor a depends on the square resistance of the graphene sheet and the effective polarization of the ferromagnetic electrodes. The effective polarization is a complex function, usually unknown, that depends on the materials and nature of the contacts involved (tunnelling, transparent, pinholes). Therefore, it is typically assumed to be a fitting parameter. If we normalize the above result equation 6 to the value at B¼ 0 taking into ffiffiffiffiffiffiffiffiffiffiffiffiffi account equation 5 and considering gi¼ gd¼ g, we obtain, p Þ 1 1 f z;bð ffiffiffiffiffiffiffi p q
  cos2 b g ð Þ ð7Þ ffiffiffiffiffiffiffiffiffiffiffiffiffi Þ f z; bð e L2 tsjj Ds nl Bð Þ Rb ð Rnl B ¼ 0 Þ ¼ a Bð Þ ð Þ a B ¼ 0 where Rnl¼ Vnl/I. At low-enough magnetic fields, the magnetoresistance can be disregarded and a(B)/a(B¼ 0)¼ 1 (see Supplementary Note 2). By replacing l2jj¼tjjDs, we obtain 6 NATURE COMMUNICATIONS 7:11444 DOI: 10.1038/ncomms11444 www.nature.com/naturecommunications NATURE COMMUNICATIONS DOI: 10.1038/ncomms11444 equation 2. Therefore, for the isotropic case, where z¼ 1 and f(z, b)¼ 1, the pre-factor in equation 7 is equal to one and the angular dependence of Rnl simply follows cos2(b g). In contrast, when za1, the cos2(b g) dependence is no longer valid. Because tsjj and Ds can be independently determined from the conventional spin precession measurements (b¼ 90°), equation 7 provides a straightforward means of extracting the spin-lifetime anisotropy ratio z as a single fitting parameter from the experimental measurement of the angular dependence of Rb nl. References 1. Han, W., Kawakami, R. K., Gmitra, M. & Fabian, J. Graphene spintronics. Nat. Nanotechnol. 9, 324–340 (2014). 2. Seneor, P. et al. Spintronics with graphene. MRS Bull. 37, 1245–1254 (2012). 3. Roche, S. & Valenzuela, S. O. Graphene spintronics: puzzling controversies and challenges for spin manipulation. J. Phys. D 47, 094011 (2014). 4. Tombros, N., Jozsa, C., Popinciuc, M., Jonkman, H. T. & van Wees, B. J. Electronic spin transport and spin precession in single graphene layers at room temperature. Nature 448, 571–574 (2007). 5. Huertas-Hernando, D., Guinea, F. & Brataas, A. Spin-orbit-mediated spin relaxation in graphene. Phys. Rev. Lett. 103, 146801 (2009). 6. Zhang, P. & Wu, M. W. Electron spin relaxation in graphene with random Rashba field: comparison of the D'yakonov-Perel' and Elliott-Yafet-like mechanisms. New J. Phys. 14, 033015 (2012). 7. Ochoa, H., Castro Neto, A. H. & Guinea, F. Elliot-Yafet mechanism in graphene. Phys. Rev. Lett. 108, 206808 (2012). 8. Dlubak, B. et al. Highly efficient spin transport in epitaxial graphene on SiC. Nat. Phys. 8, 557–561 (2012). 9. Drogeler, M. et al. Nanosecond spin lifetimes in single-and few-layer graphene- hBN heterostructures at room temperature. Nano Lett. 14, 6050–6055 (2014). 10. Kamalakar, M. V., Groenveld, C., Dankert, A. & Dash, S. P. Long distance spin communication in chemical vapour deposited graphene. Nat. Commun. 6, 6766 (2015). 11. Han, W. & Kawakami, R. K. Spin relaxation in single-layer and bilayer graphene. Phys. Rev. Lett. 107, 047207 (2011). 12. Jo´zsa, C. et al. Linear scaling between momentum and spin scattering in graphene. Phys. Rev. B 80, 241403 (2009). 13. Volmer, F. et al. Role of MgO barriers for spin and charge transport in Co/MgO/graphene nonlocal spin-valve devices. Phys. Rev. B 88, 161405 (2013). 14. Zomer, P. J., Guimaraes, M. H. D., Tombros, N. & van Wees, B. J. Long- distance spin transport in high-mobility graphene on hexagonal boron nitride. Phys. Rev. B 86, 161416 (2012). 15. Yang, T.-Y. et al. Observation of long spin-relaxation times in bilayer graphene at room temperature. Phys. Rev. Lett. 107, 047206 (2011). 16. Maassen, T., Dejene, F. K., Guimaraes, M. H. D., Jo´zsa, C. & van Wees, B. J. Comparison between charge and spin transport in few-layer graphene. Phys. Rev. B 83, 115410 (2011). 17. Pi, K. et al. Manipulation of spin transport in graphene by surface chemical doping. Phys. Rev. Lett. 104, 187201 (2010). 18. Han, W. et al. Spin relaxation in single-layer graphene with tunable mobility. Nano Lett. 12, 3443–3447 (2012). 19. Fabian, J., Matos-Abiague, A., Ertler, C., Stano, P. & Zutic´, I. Semiconductor spintronics. Acta Phys. Slovaca 57, 565–907 (2007). 20. Van Tuan, D., Ortmann, F., Soriano, D., Valenzuela, S. O. & Roche, S. Pseudospin- driven spin relaxation mechanism in graphene. Nat. Phys. 10, 857–863 (2014). 21. Fratini, S., Gosa´lbez-Martnez, D., Merodio Ca´mara, P. & Ferna´ndez-Rossier, J. Anisotropic intrinsic spin relaxation in graphene due to flexural distortions. Phys. Rev. B 88, 115426 (2013). 22. Pachoud, A., Ferreira, A., O zyilmaz, B. & Castro Neto, A. H. Scattering theory of spin-orbit active adatoms on graphene. Phys. Rev. B 90, 035444 (2014). 23. Bundesmann, J., Kochan, D., Tkatschenko, F., Fabian, J. & Richter, K. Theory of spin-orbit-induced spin relaxation in functionalized graphene. Phys. Rev. B 92, 081403 (2015). 24. Kochan, D., Gmitra, M. & Fabian, J. Spin relaxation mechanism in graphene: resonant scattering by magnetic impurities. Phys. Rev. Lett. 112, 116602 (2014). 25. Soriano, D. et al. Spin transport in hydrogenated graphene. 2D Mater. 2, 022002 (2015). 26. Tombros, N. et al. Anisotropic spin relaxation in graphene. Phys. Rev. Lett. 101, 046601 (2008). 27. Guimaraes, M. H. D. et al. Controlling spin relaxation in hexagonal BN-encapsulated graphene with a transverse electric field. Phys. Rev. Lett. 113, 086602 (2014). 28. Motsnyi, V. F. et al. Optical investigation of electrical spin injection into semiconductors. Phys. Rev. B 68, 245319 (2003). 29. Costache, M. V., Bridoux, G., Neumann, I. & Valenzuela, S. O. Lateral metallic devices made by a multiangle shadow evaporation technique. J. Vac. Sci. Technol. B 30, 04E105 (2012). 30. Neumann, I., Costache, M. V., Bridoux, G., Sierra, J. F. & Valenzuela, S. O. Enhanced spin accumulation at room temperature in graphene spin valves with amorphous carbon interfacial layers. Appl. Phys. Lett. 103, 112401 (2013). ARTICLE 31. Sierra, J. F., Neumann, I., Costache, M. V. & Valenzuela, S. O. Hot-carrier Seebeck effect: diffusion and remote detection of hot carriers in graphene. Nano Lett. 15, 4000–4005 (2015). 32. Schmidt, G., Ferrand, D., Molenkamp, L. W., Filip, A. T. & van Wees, B. J. Fundamental obstacle for electrical spin injection from a ferromagnetic metal into a diffusive semiconductor. Phys. Rev. B 62, R4790–R4793 (2000). 33. Djeghloul, F. et al. Highly spin-polarized carbon-based spinterfaces. Carbon 87, 269–274 (2015). 34. Cho, S. & Fuhrer, M. S. Charge transport and inhomogeneity near the minimum conductivity point in graphene. Phys. Rev. B 77, 081402R (2008). 35. Han, W. et al. Tunneling spin injection into single layer graphene. Phys. Rev. Lett. 105, 167202 (2010). 36. Johnson, M. & Silsbee, R. H. Coupling of electronic charge and spin at a ferromagnetic-paramagnetic metal interface. Phys. Rev. Lett. 37, 5312–5325 (1988). 37. Guimaraes, M. H. D. et al. Spin transport in high-quality suspended graphene devices. Nano Lett. 12, 3512–3517 (2012). 38. Neumann, I. et al. Electrical detection of spin precession in freely suspended graphene spin valves on cross-linked poly(methyl methacrylate). Small 9, 156–160 (2013). 39. Droth, M. & Burkard, G. Acoustic phonons and spin relaxation in graphene nanoribbons. Phys. Rev. B 84, 155404 (2011). 40. Castro Neto, A. H. & Guinea, F. Impurity-induced spin-orbit coupling in graphene. Phys. Rev. Lett. 103, 026804 (2009). 41. Van Tuan, D., Ortmann, F., Cummings, A. W., Soriano, D. & Roche, S. Spin dynamics and relaxation in graphene dictated by electron-hole puddles. Sci. Rep. 6, 21046 (2016). 42. Kochan, D., Irmer, S., Gmitra, M. & Fabian, J. Resonant scattering by magnetic impurities as a model for spin relaxation in bilayer graphene. Phys. Rev. Lett. 115, 196601 (2015). 43. Dery, H. et al. Nanospintronics based on magnetologic gates. IEEE Trans. Electron Dev. 59, 259262 (2012). 44. Roche, S. et al. Graphene spintronics: the European Flagship perspective. 2D Mater. 2, 030202 (2015). 45. Valenzuela, S. O. & Tinkham, M. Spin-polarized tunneling in room-temperature mesoscopic spin valves. Appl. Phys. Lett. 85, 5914–5916 (2004). 46. Jedema, F. J., Heersche, H. B., Filip, A. T., Baselmans, J. J. A. & van Wees, B. J. Electrical detection of spin precession in a metallic mesoscopic spin valve. Nature 416, 713–716 (2002). Acknowledgements We thank A.W. Cummings and S. Roche for a critical reading of the manuscript and D. Torres for his help in developing Fig. 1. This research was partially supported by the European Research Council under Grant Agreement No. 308023 SPINBOUND, by the Spanish Ministry of Economy and Competitiveness, MINECO (under Contract No. MAT2013-46785-P and Severo Ochoa No. SEV-2013-0295), and by the Secretariat for Universities and Research, Knowledge Department of the Generalitat de Catalunya. M.V.C., J.F.S. and J.C. acknowledge support from the Ramo´n y Cajal, Juan de la Cierva and Beatriu de Pino´s programs, respectively. F.B. acknowledges funding from the People Programme (Marie Curie Actions) of the European Union's Seventh Framework Programme FP7/2007-2013/ under REA Grant Agreement No. 624897. J.E.S. and J.V.d.V. acknowledge funding from the Methusalem Funding of the Flemish Government and the Research Foundation-Flanders (FWO). Author contributions B.R. and S.O.V. planned the measurements and wrote the paper. B.R. fabricated the samples. J.E.S. and B.R. performed the measurements. B.R. modelled the results with input from all the authors. S.O.V. supervised the experiment. All authors commented on the manuscript. Additional information Supplementary Information accompanies this paper at http://www.nature.com/ naturecommunications Competing financial interests: The authors declare no competing financial interests. Reprints and permission information is available online at http://npg.nature.com/ reprintsandpermissions/ How to cite this article: Raes, B. et al. Determination of the spin-lifetime anisotropy in graphene using oblique spin precession. Nat. Commun. 7:11444 doi: 10.1038/ncomms11444 (2016). This work is licensed under a Creative Commons Attribution 4.0 International License. The images or other third party material in this article are included in the article's Creative Commons license, unless indicated otherwise in the credit line; if the material is not included under the Creative Commons license, users will need to obtain permission from the license holder to reproduce the material. To view a copy of this license, visit http://creativecommons.org/licenses/by/4.0/ NATURE COMMUNICATIONS 7:11444 DOI: 10.1038/ncomms11444 www.nature.com/naturecommunications 7 

1209.3599

2

1209

Inelastic shot noise characteristics of nanoscale junctions from first principles

We describe an implementation of ab-initio methodology to compute inelastic shot noise signals due to electron-vibration scattering in nanoscale junctions. The method is based on the framework of non-equilibrium Keldysh Green's functions with a description of electronic structure and nuclear vibrations from density functional theory. Our implementation is illustrated with simulations of electron transport in Au and Pt atomic point contacts. We show that the computed shot noise characteristics of the Au contacts can be understood in terms of a simple two-site tight-binding model representing the two apex atoms of the vibrating nano-junction. We also show that the shot noise characteristics of Pt contacts exhibit more complex features associated with inelastic interchannel scattering. These inelastic noise features are shown to provide additional information about the electron-phonon coupling and the multichannel structure of Pt contacts than what is readily derived from the corresponding conductance characteristics.We finally analyze a set of Au atomic chains of different lengths and strain conditions and provide a quantitative comparison with the recent shot noise experiments reported by Kumar et al. [Phys. Rev. Lett. 108, 146602 (2012)].

cond-mat.mes-hall

cond-mat

Inelastic shot noise characteristics of nanoscale junctions from first principles R. Avriller1, 2 and T. Frederiksen2, 3 1Univ. Bordeaux, LOMA, UMR 5798, F-33400 Talence, France CNRS, LOMA, UMR 5798, F-33400 Talence, France 2Donostia International Physics Center (DIPC) -- UPV/EHU, Paseo Manuel de Lardizabal 4, E-20018 Donostia-San Sebastian, Spain 3IKERBASQUE, Basque Foundation for Science, E-48011, Bilbao, Spain (Dated: July 29, 2021) We describe an implementation of ab-initio methodology to compute inelastic shot noise signals due to electron-vibration scattering in nanoscale junctions. The method is based on the framework of non-equilibrium Keldysh Green's functions with a description of electronic structure and nuclear vibrations from density functional theory. Our implementation is illustrated with simulations of electron transport in Au and Pt atomic point contacts. We show that the computed shot noise characteristics of the Au contacts can be understood in terms of a simple two-site tight-binding model representing the two apex atoms of the vibrating nano-junction. We also show that the shot noise characteristics of Pt contacts exhibit more complex features associated with inelastic interchannel scattering. These inelastic noise features are shown to provide additional information about the electron-phonon coupling and the multichannel structure of Pt contacts than what is readily derived from the corresponding conductance characteristics. We finally analyze a set of Au atomic chains of different lengths and strain conditions and provide a quantitative comparison with the recent shot noise experiments reported by Kumar et al. [Phys. Rev. Lett. 108, 146602 (2012)]. I. INTRODUCTION The signatures of vibrational modes in the shot noise properties of nanoscale junctions have been the subject of active theoretical investigations [1 -- 12]. Recently, it was shown in Refs. [6 -- 8], that under applying a bias volt- age eV larger than the typical phonon energy ω0, the activation of phonon emission in a junction at low tem- peratures is responsible for a threshold behavior of the shot noise versus voltage characteristics. More specifi- cally, depending on the electronic transmission probabil- ity of the junction, the correction to the shot noise sig- nals induced by electron-phonon (e-ph) interactions was shown to exhibit jumps in the voltage derivative that are either positive or negative, as the result of a sub- tle interplay between one-electron tunneling events and correlated two-electrons processes [13]. This behavior of the inelastic shot noise signal was recently shown to be strongly dependent on fluctuations in the occupation of the locally excited vibrational mode. Under certain con- ditions, this phenomenon might lead to a strong feedback of the dynamics of the oscillator on the electronic noise properties [10, 11] and the corresponding nonlinear effect in the shot noise could be of interest for the characteri- zation of heating effects at the nanoscale. In parallel to this theoretical activity, the first non- equilibrium shot noise measurements were recently per- formed on gold (Au) nano-junctions that unravel clear signatures of the excitation of local vibrational modes [13]. Those measurements confirm qualitatively the predictions of Ref. [6 -- 8] concerning the existence of a crossover from positive to negative correction to the Fano factor upon phonon excitation, when decreasing the elec- tron transmission factor τ of the junction from unity. However, the experimental shot noise characteristics ex- hibit some unexplained features which seem to be out of the range of "single level, single vibrational mode" mod- els. For instance, the position of the crossover was shown to be shifted compared to the theory from τ ≈ 0.85 to τ ≈ 0.95 [13]. The above mentioned experiments exemplify the rele- vance of exploring new methods that allow to compute quantitatively the inelastic shot noise signals from first principles. The aim of the present paper is thus two-fold: First, to document our implementation of such a frame- work into the Inelastica [14 -- 16] ab-initio code based on Siesta [17] and TranSiesta [18]. To this end, we adopt the results derived in Refs. [19, 20] based on non- equilibrium Keldysh Green's functions. Secondly, we ap- ply our implementation to discuss the inelastic shot noise signals in atomic point contacts of Au and Pt [21, 22] with all parameters extracted from atomistic calculations. De- pending on the material, a different number of conduc- tance channels are available for the electron transport [23 -- 25] and -- as a consequence -- also the inelastic trans- port properties are qualitatively different [26]. In particu- lar, this allows us to highlight and analyze the additional complexities that arise from interchannel scattering un- der realistic conditions. We finally analyze calculations for a set of Au atomic chains of different lengths and strain conditions and compare the findings with the re- cent shot noise experiments [13]. The organization of the paper is the following. In Sec. II, we present the ab-initio methodology we have im- plemented in order to compute the correction to the shot noise induced by e-ph interactions. In Sec. III we present results obtained for different Au and Pt atomic point con- tact, and calculations for Au atomic chains bridging the electrodes is investigated in Sec. IV. Our conclusions are presented in Sec. V. II. METHODOLOGY B. Electronic structure methods 2 In this section we outline the methodology used to per- form first-principles calculations of shot noise character- istics. A. Model We consider the standard partitioning scheme in which an interacting device region D couples to two reservoirs of noninteracting electrons, namely left L and right R leads. This system is described by the following spin degenerate Hamiltonian H = HD + HL,R + HT . (1) Here the device region D, in which the e-ph interactions are assumed to be strictly localized, is described by a Hamiltonian of the form HD = H (0) H (0) e + H (0) ph + He-ph, d† dj, i H (0) ij (2) (3) H (0) e =Xi,j ph =Xλ He-ph =Xλ Xi,j ωλb† λ bλ, M λ ij d† i dj(b† λ + bλ), (4) (5) where All parameters in the above Hamiltonian are extracted from self-consistent calculations with Siesta [17] for the TranSiesta setup [18] according to the Inelastica scheme [14 -- 16]. With {φi} denoting the full nonorthog- onal basis set of atomic orbitals, the Fermion opera- tors satisfy the anti-commutation relations { d† j} = { di, dj} = 0 and { di, d† j} = Sij, where Sij = hφiφji is the overlap matrix (and similar for the lead Fermion operators). Using boldface notation throughout this manuscript for the electronic space, we let H(0) repre- sent the matrix elements of the Kohn-Sham Hamiltonian in the device region, Vα the coupling elements between device and lead α, S the overlap matrix, and Mλ the e-ph coupling matrix (obtained by finite differences [15]) corresponding to a localized mode λ with energy ωλ. i , d† In the device region D the single-particle noninteract- ing retarded (advanced) Green's function gr(a)(E), i.e., without e-ph interactions, takes the usual form gr(a)(E) = {(E ± i0+)S − H(0) − Σr(a) L (E) − Σr(a) R (E)}−1, (8) Σr(a) α (E) = (Vα)†gr(a) αS (E)Vα (9) e i and b† where d† λ are the electron and phonon creation operators in the device space, respectively. Here H (0) is the single-particle Kohn Sham DFT Hamiltonian de- scribing electrons moving in a static arrangement of the atomic nuclei. In this Hamiltonian, electron-electron in- teractions are taken into account at the mean field level. H (0) ph is the phonon Hamiltonian of free uncoupled har- monic oscillators, and He-ph is the e-ph coupling within the harmonic approximation. The Hamiltonians describing the leads HL,R and the tunnel couplings between leads and device region HT are given by ij c† α,icα,j, H α HL,R = Xα=L,RXi,j HT = Xα=L,RXi,j (cid:16)V α ij c† α,i dj + h.c.(cid:17) , (6) (7) where c† α,i is the electron creation operator in lead α = L, R. Each lead is considered to be in local equilib- rium such that the occupied states are characterized by a Fermi distribution with thermal energy kBT and chemical potential µα. An applied voltage V is as- sumed to shift symmetrically the chemical potentials of the leads with respect to the Fermi level position at equilibrium EF (determined self-consistently in Kohn- Sham DFT by filling the electronic states from below) as µL(R) = EF + (−)eV /2 and to leave the electrostatic potential of the device unchanged. is the retarded (advanced) self-energy due to lead α. Here gr(a) αS (E) represents the corresponding surface Green's function for the isolated lead and is calculated recursively [27]. The retarded and advanced Green's functions are connected through the relation ga(E) = {gr(E)}†. For convenience we also introduce here the level broadening due to lead α Γα(E) = i{Σr α(E) − Σa α(E)}. (10) Finally, as the theory of shot noise characteristics pre- sented in the following section is developed in an orthog- onal basis, we orthogonalize all relevant quantities ac- cording to standard Lowdin transformations [28], i.e., H(0) → eH(0) = S−1/2H(0)S−1/2, Mλ → fMλ = S−1/2MλS−1/2, S → eS = S−1/2SS−1/2 = 1, gr(a) →egr(a) = S1/2gr(a)S1/2, Γα → eΓα = S−1/2ΓαS−1/2, S1/2 = Udiag(√ε1, . . . ,√εn)U−1, S−1/2 = Udiag(1/√ε1, . . . , 1/√εn)U−1, where the transformation matrices (11) (12) (13) (14) (15) (16) (17) are determined through the eigenvalue problem SUi = εiUi for the overlap matrix. C. Nonequilibrium Keldysh Green's functions We summarize in this section how to calculate conduc- tance and shot noise characteristics for nano-junctions described by the electronic structure as outlined in Sec. II B. The concepts underlying this methodology are derived in Ref. [19] and were applied to single-level and single-mode models in Refs. [6 -- 8]. Here we follow the specific generalization to the multi-level and multi-mode model formulated by Haupt et. al. [8, 20]. The two fun- damental approximations are (i) weak e-ph interactions and (ii) the so-called extended wide-band limit (EWBL). 1. Extended wide-band limit (EWBL) As in previous work we adopt the EWBL [8, 14, 15, 20, 29, 30] to perform energy integrations analytically such that explicit results for the mean current and shot noise can be stated. The EWBL consists of approximating the noninteracting retarded and advanced Green's functions as well as the level broadenings with their values at the Fermi energy EF , i.e., gr(a)(E) ≈ gr(a)(EF ) ≡ gr(a), Γα(E) ≈ Γα(EF ) ≡ Γα. (18) (19) Physically this is motivated by the fact that in many real systems the electronic spectral properties typically vary slowly on the scale of a few phonon energies and applied voltages [14, 15, 29, 30]. In the case of atomic gold wires this approximation was successfully tested by one of us in Ref.[15] via a direct comparison to computationally more expensive calculations based on the self-consistent Born approximation. The physical reason is the strong hybridization of device states with the electrode states (life-time broadening on the eV scale) and the existence of only low-energy vibrational modes (on the meV scale). As this situation also applies to point contacts of Au and Pt, we expect the EWBL to be a very good approxima- tion for these systems too. 2. Computing the current characteristics In absence of e-ph interactions (Mλ = 0), the (bare) current I0(V ) is given by the standard Landauer-Buttiker formula [31]. Within EWBL the transmission function becomes energy independent so that the expression for the current-voltage characteristics is simply where I0[2e/h](V ) = Tr{T}eV, T = ΓLgrΓRga (20) (21) is the elastic (bare) transmission matrix of the junction. This expression is similar to the Fisher and Lee formula for the conductance [32]. 3 In presence of weak coupling to the vibrational sub- system (Mλ 6= 0), the above expression for the mean electronic current has to be modified. At second or- der of perturbation theory in the e-ph coupling strength, the correction to the current δI(V ) (within EWBL) can be expressed in terms of products of microscopic factors (system dependent) by voltage-dependent universal func- tions (system independent) [15, 20]. We write the inelas- tic corrections to the current as δI[2e/h](V ) = δIel(V ) + δIinel(V ), δIel[2e/h](V ) =Xλ n(1 + 2nλ BTr{T(1) Tr{T(1) δIinel[2e/h](V ) =Xλ B)Tr{T(0) λ } λ }oeV, λ }gλ(eV ). +2nλ (22) (23) (24) with the microscopic factors given by traces over the fol- lowing quantities (0) (1) T T i (25) λ = ΓL(cid:0)grMλgr λ = ΓLgr(cid:8)MλARMλ ReMλAR + H.c.(cid:1), 2(cid:0)MλAMλgrΓR − H.c.(cid:1)(cid:9)ga, 2(cid:8)U (eV − ωλ) − U (eV + ωλ)(cid:9), (27) (26) − 1 and the voltage dependent universal functions given by gλ(eV ) = eV + U (eV ) = eV coth (βeV /2). (28) In the above equations, gr Re denotes the real part of gr, β = 1/kBT the inverse temperature of the electrodes, and Aα = grΓαga the partial spectral function corre- sponding to lead α. The total spectral function is given by A = AL + AR. We note that the expressions Eqs. (23)-(24) are subject to the following three additional approximations: (i) We have ignored the asymmetric contributions to the con- ductance (with respect to voltage) which are derived in Ref. [20, 33, 34]. These contributions are logarithmically divergent in the zero-temperature limit for eV = ωλ and signal the breakdown of second-order perturbation theory at the inelastic threshold. However, a resumma- tion scheme might renormalize and cure this problem [10]. Out of the threshold region, the logarithmic cor- rections to the conductance are typically orders of mag- nitude smaller than the symmetric contribution. They also vanish in the limit of symmetric couplings to the left and right lead [15], thus justifying our assumption. (ii) We fix the phonon populations nλ B to the equilib- rium values as given by the Bose-Einstein distribution nλ B = 1/[exp(βωλ)−1] (regime of equilibrated phonons). For atomic point contacts and chains this is a reasonable starting point as the vibrations in the nanoscale contact are damped to some extent by coupling to bulk phonons [14, 15, 35, 36]. Furthermore, the effect of phonon heat- ing on the shot noise characteristics is a delicate research topic that is beyond the scope of the present study [10 -- 12]. (iii) Equation (22) corresponds to taking into ac- count only the Fock (exchange) self-energy. Neglecting the Hartree term is indeed a good approximation in most cases of interest: within the EWBL this term provides a voltage independent renormalization of the molecular level positions in the regime of equilibrated phonons and thus displays no features at the phonon emission thresh- old [20]. e.g., τ ≈ τ +Pλ Tr{T(0) The total correction to the current in Eq. (22) is the sum of two contributions. The first one δIel(V ) is due to elastic processes induced by e-ph interactions that renor- malize the bare transmission factor τ to an effective one, λ } at zero temperature. The sec- ond contribution δIinel(V ) originates from inelastic pro- cesses activated by phonon emission. The voltage de- pendence of this contribution is a universal function of voltage gλ(V ) [see Eq. (27)] that exhibits a threshold at eV = ωλ, in the zero temperature limit. The sign and size of this threshold (jump in conductance) is controlled by the inelastic microscopic factors T(1) λ given in Eq. (26). Upon differentiation of Eq. (22) with respect to voltage one obtains the correction of the conductance δG(V ) = ∂V (δI(V )). In the zero temperature limit, δG(V ) is dis- continuous at the inelastic threshold, due to the contri- bution of the inelastic term, see Eq. (24). We thus define the corresponding jump in the inelastic correction to the conductance ∆Gλ ≡ limη→0+{δG(ωλ+η)−δG(ωλ−η)} at the threshold voltage corresponding to mode λ by ∆Gλ[2e2/h] = Tr{T(1) λ }. (29) 3. Computing the shot noise characteristics Shot noise characteristics in absence of coupling to lo- cal vibrational modes (Mλ = 0) was reviewed by Blanter and Buttiker in Ref. [37]. Within EWBL, the correlation function of the current operator evaluated at zero fre- quency, e.g. the (bare) shot noise characteristics S0(V ), is given by the simplified expression [37] 2 β S0[2e2/h](V ) = Tr{T2} + Tr{T(1 − T)}U (eV ). (30) This expression is associated with the noise induced by thermal fluctuations in the electron occupation of the electrode Fermi seas and with fluctuations in the occu- pation of the coherent left- and right-moving scattering states. At second order of perturbation theory in the e-ph cou- pling strength, the expression for the finite temperature correction to the noise δS(V ) in the regime of equili- brated phonons is rather complicated, and derived in de- tail in Ref. [20]. While we have implemented these results and used them in the numerical part of this work, we here just state the simpler result for the zero-temperature limit 4 δS[2e2/h](V ) = δSel(V ) + δSinel(V ), δSel[2e2/h](V ) =Xλ δSinel[2e2/h](V ) =Xλ Trn(1 − 2T)T(0) Trn(1 − 2T)T(1) λ oeV , λ + Qλo (33) (32) (31) ×(eV − ωλ)θ(eV − ωλ), with Qλ = −gaΓLgr(cid:8)MλARΓLARMλ +MλARΓLgrMλgrΓR + H.c.(cid:9). (34) Consistent with our assumptions for the inelastic correc- tions to the mean current (Sec. II C 2) we neglect also for the noise part the asymmetric terms leading to logarith- mic divergences as well as contributions from the Hartree e-ph self-energy. Analogous to the corrections to the current [Eqs. (22)- (28)] also the inelastic noise corrections [Eqs. (31)-(34) in the zero-temperature limit] can be written as products of microscopic factors by universal voltage-dependent func- tions. The first term δSel(V ) [Eq. (32)] represents an elastic correction to the noise. The second term δSinel(V ) [Eq. (33)] is related to inelastic signatures of phonon ac- tivation in the shot noise. Its origin and interpretation is less intuitive than the corresponding expression for the mean current. The part proportional to Tr{(1−2T)T(1) λ } originates from a mean-field contribution to the shot noise while the other part proportional to Tr{Qλ} is related to vertex corrections [20]. A similar (although slightly different) decomposition in terms of one-electron (mean-field like) and two electron (vertex-like) processes was proposed in Ref. [13]. Instead of looking directly at the inelastic corrections in the shot noise it is convenient to analyze the voltage derivative of the shot noise δ S(V ) = ∂V (δS(V )), i.e., the inelastic noise change. In the zero temperature limit, δ Sλ(V ) is discontinuous at the inelastic threshold, due to the contribution of the inelastic term, see Eq. (33). We thus define the corresponding jump in the inelastic correction to the shot-noise ∆ Sλ ≡ limη→0+{δ S(ωλ + η)− δ S(ωλ − η)} at the threshold voltage corresponding to mode λ by ∆ Sλ[2e3/h] = Trn(1 − 2T)T (1) λ + Qλo. (35) III. RESULTS FOR AU AND PT CONTACTS We first consider Au and Pt atomic point contacts, such as those shown Fig. 1, as benchmark systems for our ab-initio calculations. Along the lines of Ref. [38] by one of us, we consider periodic supercells with a 4×4 rep- resentation of either Au(100) and Pt(100) surfaces sand- wiching two pyramids pointing toward each other. The (a) (b) PSfrag replacements Pt contact 5 (c) 25 20 15 10 5 d L ) V e m ( λ ω  PSfrag replacements Au contact (a) x2 x2 25 20 15 10 5 FIG. 1. (Color online) Atomic point contacts of (a) Au atoms (shown in yellow-gold) and (b) Pt atoms (shown in blue-gray) considered in the first-principles transport calculations. The characteristic electrode separation L is measured between the second-topmost surface layers. The distance d characterizes the separation between the two apex atoms. characteristic electrode separation L is measured between the second-topmost surface layers, since the surface layers themselves are relaxed and hence deviate on the decimals from the bulk values. Our Siesta calculations use a single-ζ plus polariza- tion (SZP) basis with a confining energy of 0.01 Ry [corresponding to the 5d and 6(s, p) states of the free atoms], the generalized gradient approximation (GGA) for exchange-correlation, a cutoff energy of 200 Ry for the real-space grid integrations, and the Γ-point approxima- tion for the sampling of the three-dimensional Brillouin zone. The interaction between the valence electrons and the ionic cores is described by standard norm-conserving Troullier-Martins pseudopotentials generated from rela- tivistic atomic calculations. As for the bulk Au and Pt crystals, we set the lattice constants to 4.18 A and 4.02 A, respectively. For each electrode separation L we relax the surface atoms until the residual forces are smaller than 0.02 eV/A and proceed calculating vibrational modes and e-ph cou- plings by finite differences. For simplicity, we here only consider that the two apex atoms can vibrate, leaving us with six characteristic vibrational modes in the device. This assumption is only made to facilitate a fundamen- tal understanding of the inelastic signals. Finally, while electron transport in the supercell approach generally in- volves a sampling over k-points, we approximate in the following all relevant quantities with their values at the Γ-point. We present in Fig. 2 the dependence of the electron transmission and vibrational frequencies, as a function of the electrode separation L. In both cases of Au and Pt junctions, the total transmission τ decreases with L as observed in Fig. 2(b),(d) and covers the range from contact (ballistic limit) to the tunnel regime (low- transmission regime). The total transmission τ = Pi τi 0 14.5 15 15.5 16 16.5 17 3 2 (b) 0 13 13.5 14 14.5 15 15.5 3 2 (d) n o i s s i m s n a r T 1.0 0.5 0.2 0.1 τtot τ1 τ2 τ3 τ4 1.0 0.5 0.2 0.1 14.5 15 15.5 16 16.5 17 13 13.5 14 14.5 15 15.5 L (A) L (A) FIG. 2. (Color online) (a) Vibrational frequencies and (b) electronic transmission at the Fermi energy as a function of electrode separation L for the Au atomic point contact. (c),(d) similar as (a),(b) but for the Pt atomic point con- tact. As indicated in the legends in panel (a) the longitudinal (transverse) eigenmodes are pictured in (a) and (c) with cir- cular (diamond) symbols. The out-of-phase (in-phase) modes are represented by filled (open) symbols. In panels (b) and (d) both the total transmission τ = τtot (large red circles) as well as the transmission τ1, . . . , τ4 for the four most conduct- ing transport eigenchannels (small symbols) are shown. Note that the legends shown in the Au panels apply also to the Pt panels. can be understood as a sum over eigenchannel trans- missions τi for a set of (non mixing) electron scattering states. In Fig. 3 we have visualized the scattering states belonging to the four most transmitting channels (waves incoming from below) [39]. In the case of Au junctions the total transmission is essentially made up of a single channel, i.e., τ ≈ τ1 as seen in Fig. 2(b). This fact can be traced back to the single s-valence of Au [21, 23 -- 25]. The corresponding eigenchannel scattering state ψσ 1 is rotationally symmet- ric (σ-type) as seen in Fig. 3(a). The fact that essentially only one transmission channel contributes to the elastic current can also be appreciated by comparing the ampli- tude of the transmitted part of the different scattering states as it reflects the transmission probability. For the Au contact it is clear that only ψσ 1 penetrates significantly the tunnel gap between the apex atoms. 6 ψσ 1 (r) ψπ 2 (r) ψπ 3 (r) ψσ 4 (r) (a) (b) PSfrag replacements ψσ 1 (r) ψπ 2 (r) ψπ 3 (r) ψσ 4 (r) FIG. 3. (Color online) Isosurface representations of the four most conducting eigenchannel scattering states ψi(r) (incoming from below) for (a) a Au atomic point contact (L = 16.58 A) and (b) a Pt atomic point contact (L = 15.50 A). The states are ordered according to decreasing transmission. Due to the tunnel gap between the two sides the electron scattering states decay rapidly and the transmitted part of the wave on the other electrode is not visible. The blue and red colors represent the sign of the real part of the scattering states (our choice of phase makes the imaginary part negligible for visualization purposes). The isosurface plots reveal different rotational symmetry around an axis connecting the apex atoms. For both contacts one observes that the channels ψσ 3 are π-type states (with a nodal plane through the symmetry axis). 4 are σ-type states (rotationally symmetric) while ψπ 2 and ψπ 1 and ψσ For the Pt junctions on the other hand the total trans- mission has significant contributions from three eigen- channels in the contact regime, cf. Fig. 2(d). As revealed in Fig. 3(b) the symmetry of the most transmitting chan- nel is of σ-type while the following two are of π-type with a nodal plane through the symmetry axis. This mul- tichannel nature reflects the partially filled sd valence shells of Pt [21, 23 -- 25]. A. Au contacts 1. δG(V ) and δ S(V ) characteristics Using the methodology presented in Sec. II C we pro- ceed by studying the inelastic effects in the transport through the considered Au atomic point contacts. Fig- ure 4 shows the curves obtained for the δG(V ) and δ S(V ) = ∂V (δS(V )) characteristics upon phonon exci- tation, for several electrode distances spanning the range from tunnel to contact. As shown in Fig. 4(a), for each of the considered geometries the correction to the reduced conductance δG(V ) exhibits a threshold-like character around eV ∼ ω←→ ≈ 10 meV corresponding to the out-of-phase lon- gitudinal vibrational mode [38]. The signals from the other five modes are so small that they are hardly visi- ble. For the tunneling setups (τ < 1/2) the activation of phonon emission processes above the inelastic threshold opens a new channel for conduction, thus increasing the conductance compared to its elastic background value. Contrary, in the contact regime (τ > 1/2) the activation of inelastic scattering processes reduces the conductance (backscattering), i.e., results in a negative jump. shows Figure 4(b) the corresponding δ S(V ) = ∂V (δS(V )) characteristics, which also exhibit a threshold response for voltages close to the phonon frequency of the "←→" mode. But in contrast to the conductance curves, the finite temperature effect is not only to smoothen the jump but also to produce some small downturn in the vicinity of the inelastic threshold [8, 20]. The sign of the jumps ∆ S←→ is consistent with the predictions based on "single-level, single-mode" models in Refs. [6 -- 8], i.e., that it is negative only in the region 0.15 ≤ τ ≤ 0.85 and Au-contact =0.990 =0.982 =0.866 =0.685 =0.393 =0.293 =0.062 L14.98  L15.38  L15.78  L15.88  L15.98  L16.08  L16.58  (a) 1.2 1.0 0.8 0.6 0.4 0.2 0.0 ) 2 e 2 h 1 0 . 0 ( /
G -0.2 Au-contact =0.990 =0.982 =0.866 =0.685 =0.393 =0.293 =0.062 L14.98  L15.38  L15.78 L15.88  L15.98  L16.08  L16.58  (b) 1.0 0.8 0.6 0.4 0.2 0.0 ) 3 e 2 h 1 0 . 0 ( / S -0.2 0.0 0.005 0.01 0.015 0.02 V (eV) 0.0 0.005 0.01 0.015 0.02 V (eV) FIG. 4. (Color online) Inelastic conductance and noise cor- rections for Au atomic point contacts with different electrode separations L in the regime of equilibrated phonons. (a) Con- ductance corrections δG/τ (V ) induced by e-ph interactions as a function of voltage V . (b) Derivative of the shot noise with respect to voltage δ S/τ (V ) induced by e-ph interactions. For each geometry six eigenmodes are considered as only the two apex atoms are vibrating, cf. Fig. 2(a). The calculations are performed at T = 4.2 K. positive elsewhere. The onset of a negative inelastic cor- rection to shot noise is not particularly intuitive. It was recently observed experimentally in shot noise measure- ments performed on Au nano-junctions and explained in terms of correlated two-electron processes mediated by Pauli principle (Pauli blocking) and e-ph interactions [13]. A more direct way to appreciate these trends is shown in Fig. 5(a). Here the total jumps in the conductance ∆G =Pλ ∆Gλ (blue circles) and derivative of shot noise versus voltage ∆ S = Pλ ∆ Sλ (red diamonds) [over a voltage range that covers all possible phonon excitations] is shown as a function of the bare transmission τ of each considered geometry. For comparison the correspond- ing jumps ∆G←→ and ∆ S←→ associated with just the most active "←→" mode is shown with stars. All the computed data is consistent with sign changes in the in- elastic correction at τ = 1/2 for the conductance and at τ ≈ {0.15, 0.85} for the shot noise. 2. Analytic model To develop an understanding for the calculated am- plitudes and sign changes for ∆G and ∆ S shown in Fig. 5(a), we developed a simple two-site tight-binding model of the vibrating Au contact along the lines of Ref. [38]. This model is more appropriate to describe our physical problem than the single-level model analyzed in Refs. [6 -- 8]. As shown in the inset to Fig. 5(a) we represent each of Au-contact (b) Au-contact 7 2.0 0.0 -2.0 -4.0 -6.0 G e / S 0.2 (a) ) 3 e 2 h 1 0 . 0 ( S  & ) 0.1 0.0 2 e 2 h -0.1 1 0 . 0 ( G -0.2 0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0   FIG. 5. (Color online) Analysis of inelastic features for Au atomic point contacts with varying electrode separation d shown as a function of the total transmission factor τ . (a) Ab- solute values of the total jump in the conductance ∆G (blue circles) and derivative of shot noise versus voltage ∆ S (red diamonds) at zero temperature. Blue-dashed and red-dotted curves are analytic results for ∆G and ∆ S, respectively, ob- tained within the corresponding two-site tight-binding model of the vibrating nano-junction (shown in inset). (b) The ra- tio ∆ S/e∆G (red circles) as a function of the transmission factor at zero temperature. The black-dashed curve represent the corresponding analytic result. Common to both panels: Stars correspond to the inelastic signals when retaining only the contribution of the longitudinal out-of-phase vibrational mode ∆G←→ and ∆ S←→. The parameters for the two-site tight-binding model are m0 = 0.0167Γ, t0 = 0.875Γ, and ǫ0 = EF (see text for details). the two apex atoms in the Au contact by a single orbital and write the noninteracting electronic Hamiltonian of this device region as two−site =(cid:20) ǫ0 t(d) H(0) t(d) ǫ0 (cid:21) , (36) where ǫ0 is the onsite energy of each orbital (chosen to be equal to the Fermi energy EF for simplicity) and t(d) is the hopping term that is modulated when varying the distance d between the electrodes. The hybridization of each orbital with its metallic electrode is described by 0 0(cid:21) , ΓL =(cid:20) Γ 0 0 Γ(cid:21) , ΓR =(cid:20) 0 0 (37) where Γ characterizes the coupling strength to each lead. For the dependence of t on distance d we adopt the simple relationship t(d) = t0 1 + e(d−d0)/D , (38) that interpolates between the tunneling regime (d ≫ d0) for which the hopping decreases exponentially with dis- tance and the contact regime (d ≈ d0) for which it de- creases linearly with distance. In Eq. (38), the parameter   8 d0 is the typical distance where the contact is formed, t0 is an energy scale that provides the prefactor for the ex- ponential decay in the tunnel regime (taken to be larger than Γ/2) and D describes the size of the crossover region between the two regimes. Within the EWBL the elastic transmission for this two- site model is given by τ (d) =(cid:16) Γt(d) (Γ/2)2 + t(d)2(cid:17)2 . (39) As it is reasonable to consider Γ to be independent of d (and to be the largest energy scale of the model), we restrict the value of the hopping to be in the physically relevant branch 0 < t(d) ≤ Γ/2 < t0, for which one sees that the correspondence between τ (d) and t(d) is one to one, i.e., the transmission factor τ (t) is a bijective function of the hopping spanning the range 0 < τ (t) ≤ 1 and simply decreases with d because of the dependence in Eq. (38). Finally, the two atoms are coupled to the out-of-phase longitudinal vibrational mode (←→) (vibrating at a fre- quency ω←→) which modulates the interatomic distance d. The corresponding e-ph coupling matrix M←→ can therefore be determined as the derivative of the electronic Hamiltonian with respect to d, i.e., M←→[t(d)] = m0 t(d)[t0 − t(d)] 1 0(cid:21) . Γ/2(t0 − Γ/2)(cid:20) 0 1 (40) This expression for the e-ph coupling properly captures the physical behavior with d, namely M←→ is propor- tional to t(d) in the tunnel regime (exponential depen- dence on hopping amplitude) and is almost constant in the contact regime (linear dependence on hopping am- plitude). The coupling matrix M←→ depends on two parameters, namely m0 which is the value of the e-ph coupling at unit transmission (obtained for t(d) = Γ/2) and the hopping energy scale t0. The dependence of the coupling matrix with the distance d is being encoded into the one to one relation τ [t(d)] and is thus not explicit in Eq. (40). In the zero temperature limit, the inelastic corrections to the mean current and shot noise for this two-site model can simply be expressed as δI[2e/h](V ) ≈ γ(τ ; m0, t0)n2(1 − τ )eV + (1 − 2τ )(eV − ω←→)θ(eV − ω←→)o, δS[2e2/h](V ) ≈ γ(τ ; m0, t0)n2(1 − τ )(1 − 2τ )eV +(cid:2)1 − 8τ (1 − τ )(cid:3)(eV − ω←→)θ(eV − ω←→)o, (41) (42) where γ(τ ; m0, t0) = τ(cid:16) m0[t0 − t(τ )] Γ/2(t0 − Γ/2)(cid:17)2 , (43) is an effective e-ph coupling constant that depends on all the parameters describing the electronic and vibrational structure of the junction, namely τ , m0 and t0. The comparison of our ab-initio results with Eqs. (41)- (42) is shown in Fig. 5(a). The blue-dashed and red- dotted lines correspond to the results for the analyt- ical jumps ∆G and ∆ S, respectively. We modulated the transmission factor τ by decreasing the value of the hopping term from t(d) = Γ/2 (in the contact regime τ = 1) to t(d) ≈ 0 (in the tunnel regime τ ≈ 0). We found that a reasonable (although not perfect) fit to the ab-initio data points could be achieved by fix- ing the two independent parameters m0 = 0.0167Γ and t0 = 0.875Γ. It is interesting to notice that the simple analytical model of Refs. [6 -- 8] fails to reproduce both the shape and amplitude of the curves in the full range of transmissions τ ∈ [0, 1] (not shown here), mainly be- cause the e-ph coupling strength changes when varying the distance between the electrodes in a way qualitatively provided by Eq. (40). We also note that the analytic re- sults in Fig. 5(a) display clear asymmetries with respect to τ = 1/2. This is because the situations τ → 0 (tunnel limit) and τ → 1.0 (ballistic limit) are physically very dif- ferent (this asymmetry is also present in the single-level models of Refs. [6 -- 8]). Moreover, the inelastic correc- tions, as given by Eqs. (41)- (42)- (43), have nontrivial dependences on τ . For the chosen model parameters we find extrema in ∆G around τ ≈ {0.22, 0.94} and in ∆ S around τ ≈ {0.07, 0.55, 0.99}. A remarkable feature of the two-site tight-binding model is that the ratio of ∆ S[2e3/h] to e∆G[2e2/h] is a universal function of τ and is independent of the effec- tive e-ph coupling strength γ(τ ; m0, t0), i.e., ∆ S[2e3/h] e∆G[2e2/h] = 1 − 8τ (1 − τ ) 1 − 2τ . (44) This result is also found for the single-level model of Refs. [6 -- 8] as a common prefactor containing the details of the electronic structure cancels out. Figure 5(b) shows that our ab-initio data follows quantitatively the analytic results for the ratio ∆ S/e∆G (black-dashed curve). The agreement with the analytical result is even better when considering only the contribution from the out-of-phase longitudinal vibrational mode to the inelastic signals, i.e., ∆G←→ and ∆ S←→ (red stars). B. Pt-contacts 1. δG(V ) and δ S(V ) characteristics Pt : contact regime Pt : contact regime =3.148 =3.014 =2.925 =2.807 L13.50 L14.00 ! L14.25 " L14.50 # (b) 0.6 0.5 ) 0.4 h 3 e 2 1 0 . 0 ( / S 0.3 0.2 0.1 0.0 =3.148 =3.014 =2.925 =2.807 L13.50  L14.00  L14.25  L14.50  0.01 0.02 V (eV) 0.03 0.0 Pt : tunneling regime =0.330 =0.224 =0.094 =0.043 L14.90 & L15.00 ' L15.25 ( L15.50 ) 0.01 0.02 V (eV) (d) 0.8 0.6 ) 3 e 2 h 0.4 1 0 . 0 ( /+ S 0.2 0.0 -0.2 0.03 -0.4 0.0 0.01 0.02 V (eV) Pt : tunneling regime =0.330 =0.224 =0.094 =0.043 L14.90 , L15.00 - L15.25 . L15.50 / 0.01 0.02 V (eV) 0.03 0.03 (a) -0.4 -0.5 ) -0.6 2 e 2 h 1 0 . 0 ( / G -0.7 -0.8 -0.9 -1.0 0.0 (c) 1.0 0.8 ) 0.6 2 e 2 h 1 0 . 0 ( /% G 0.4 0.2 0.0 -0.2 -0.4 0.0 FIG. 6. (Color online) Inelastic conductance and noise cor- rections for Pt atomic point contacts with different electrode separations L in the regime of equilibrated phonons. (a) Con- ductance corrections δG/τ (V ) (in the contact regime) induced by e-ph interactions as a function of voltage V . (b) Derivative of the shot noise with respect to voltage δ S/τ (V ) (in the con- tact regime) induced by e-ph interactions. (c)-(d) as (a)-(b) but for the geometries corresponding to the tunneling regime. For each geometry six eigenmodes are considered as only the two apex atoms are vibrating, cf. Fig. 2(c). The calculations are performed at T = 4.2 K. We present in Fig. 6 the corresponding results for Pt atomic point contacts as was given in Fig. 4 for Au con- tacts. In contrast to the Au case the electronic transport properties of Pt contacts can no longer be understood in terms of a single conducting eigenchannel. In fact, as seen from Fig. 2(d) the contact regime is characterized by three almost open channels, i.e., one σ-type (labelled ψσ 1 ) and two π-type (labelled ψπ 2,3) as visualized in Fig. 3(b). The fourth channel ψσ 4 is included as it turns out that scattering into such closed channels are important to un- 9 derstand the inelastic transport characteristics. As for the Au contact the transmission factor decreases with L and drops suddenly at the point where the contact (chemical bond) breaks (τ ≈ 2.14 for the critical geom- etry L = 14.80 A). Beyond this point the transmission drops exponentially with d signalling the tunnel regime [see Fig. 2(d)]. The case of the contact regime (τ ≈ 3.0) is shown in Fig. 6(a)-(b). The inelastic features in the δG(V ) and δ S(V ) characteristics reveal several steps associ- ated to the excitation of transverse and longitudinal vi- brational modes. In this transport regime, the jumps ∆G(λ) (∆ S(λ)) are not necessarily negative (positive) -- i.e., dominated by inelastic backscattering processes -- as expected for a single-channel system close to the ballistic limit. As shown in Fig. 6(a)-(b), the sign of the jumps can be the other way around (see also the discussion in Sec. III B 2). When the transmission factor decreases, one enters into the tunnel regime (the junction breaks). As shown in Fig. 6(c)-(d), two main inelastic signals are seen and the jumps ∆G(λ) and ∆ S(λ) are always positive in the case of low transmissions. 2. Mode by mode analysis Due to the multichannel nature of the Pt contacts, the δG(V ) and δ S(V ) characteristics cannot be described within the framework of the simple analytical model used for Au contacts in Sec. III A 2. Instead, we can gain an understanding for the characteristics by analyzing the contribution from each vibrational mode to the total in- elastic signals δG(V ) and δ S(V ) for the Pt contacts. For simplicity, we restrict our analysis to the L = 14.50 A ge- ometry representative for the contact regime [Fig. 7(a)- (b)] and to the L = 15.50 A geometry representative for the tunnel regime [Fig. 7(c)-(d)]. These two geometries are shown respectively in Fig. 6 with plain red and blue lines. Table I lists the vibrational modes and energies ωλ, the corresponding inelastic corrections in conductance ∆Gλ and shot noise ∆ Sλ, and a decomposition of the underlying scattering processes among the eigenchannels. The idea is that inelastic scattering can be understood in terms of Fermi's golden rule where scattering occurs from occupied eigenchannel scattering states ψL i (for channel i) originating in the left electrode into empty scatter- ing states ψR (for channel j) originating in the right j electrode (or vice versa, depending on the bias polar- ity) [40, 41]. The total scattering rate, proportional to Pi,j hψL j i2, can therefore be decomposed into intrachannel (i = j) and interchannel (i 6= j) compo- nents. We begin our analysis with the results for the contact geometry (L = 14.50 A) as shown in Fig. 7(a)-(b). The conductance δG(V ) curve is dominated by the inelastic contribution from the longitudinal, out-of-phase vibra- tional mode (labelled ←→) with an energy quantum of i MλψR   $ * (a) 0.19 ) -0.01 2 e 2 h 1 0 . 0 ( /1 G -0.21 -0.41 -0.61 0.0 Pt : L14.50 < =2.807 (b) 0.3 Pt : L14.50 S =2.807 = > ? @ 2 A B C 2 D E F Total T U V W 2 X Y Z 2 [ \ ] Total 4 5 2 9 : ; 2 6 7 8 2 3 0.2 ) 3 e 2 h 1 0 . 0 ( /H S 0.1 0.0 0.03 0.0 I J 2 M N O K L 2 P Q R 0.03 0.01 0.02 V (eV) 0.01 0.02 V (eV) (c) 0.8 Pt : L15.50 j =0.043 (d) 0.6 Pt : L15.50  =0.043 (a) L = 14.50 A -- contact regime Mode λ ←→ ←← ↑↓ ×2 ↑↑ ×2 10 ωλ [meV] ∆Gλ[%∆G] ∆ Sλ[%∆ S] intraa: ψL intera: ψL interb: ψσ i ↔ ψR i ↔ ψR 1,4 ↔ ψπ i (i ≤ 10) [%] j (i 6= j) [%] 2,3 [%] 9.8 16.2 11.6 6.7 116.7 38.4 -38.8 -16.3 72.0 43.0 51.1 -66.1 88 11 0 0 92 66 0 97 70 61 34 0 (b) L = 15.50 A -- tunnel regime Mode λ ←→ ←← ↑↓ ×2 ↑↑ ×2 ωλ [meV] ∆Gλ[%∆G] ∆ Sλ[%∆ S] intraa: ψL intera: ψL interb: ψσ 15.7 16.3 12.0 26.2 65.4 -0.1 61.1 -0.1 29.4 36 46 0 11.8 8.5 9.6 0 97 27 0.6 ) 2 e 2 h 0.4 1 0 . 0 ( /_ G 0.2 0.0 -0.2 0.0 k l m n 2 o p q 2 r s t Total b c ` a 2 g h i 2 d e f 0.4 0.2 0.0 ) 3 e 2 h 1 0 . 0 ( /v S -0.2 0.01 0.02 0.03 0.0 V (eV) ‚ ƒ „ … 2 † ‡ ˆ 2 ‰ Š ‹ Total y z i ↔ ψR i ↔ ψR 1,4 ↔ ψπ 0 99 65 a considering the 10 most transmitting eigenchannels b considering only scattering among eigenchannels 1-4 i (i ≤ 10) [%] 99 j (i 6= j) [%] 0 2,3 [%] 0 TABLE I. Mode by mode analysis of the vibrational frequen- cies ωλ and jumps in the inelastic signals (∆Gλ and ∆ Sλ) for two Pt atomic point contacts. The characteristic elec- trode separation is (a) L = 14.50 A (contact regime) and (b) L = 15.50 A (tunnel regime). Relative contributions from inter- and intrachannel scattering processes to the total scat- tering rate are given in percent. w x 2 ~  € 2 { } 0.03 0.01 0.02 V (eV) FIG. 7. (Color online) Mode by mode analysis of inelastic features for two characteristic Pt atomic point contacts in the contact and tunneling regimes, respectively. (a) Correction to the conductance δG/τ and (b) correction to the derivative of the shot noise with respect to voltage δ S/τ from the six characteristic vibrational modes as a function of voltage eV for L = 14.50 A (contact regime). (c)-(d) as (a)-(b) but for L = 15.50 A (tunnel regime). The calculations are performed at T = 4.2 K. The mode character is shown for each curve with two arrows similar to Fig. 2(c). The total characteristics δG/τ (V ) and δ S/τ (V ) (sum over all vibrational modes) are shown as colored plain lines. ω←→ = 16.2 meV. As reported in Tab. I, ∆G←→ is found to account for 116.7 % of the overall conductance correction. The table also reports that for this partic- ular mode intrachannel inelastic transitions are clearly dominant. In fact, 88 % of the total scattering corre- sponds to intrachannel scattering involving the 10 most transmitting eigenchannels. Furthermore, transitions be- tween the σ-type and π-type states shown in Fig. 3 are symmetry forbidden (0 % scattering) as expected for a longitudinal mode that creates a rotationally symmetric deformation potential along the transport axis. The e-ph coupling matrix M←→ is therefore essentially diagonal in the eigenchannel basis. Finally, since scattering from this mode is essentially intrachannel involving only the three most transmitting eigenchannels, its effect can be ratio- nalized in terms of a simple summation over three inde- pendent single-channel models [6 -- 8] where the resulting jump ∆G←→ is expected to be negative and ∆ S←→ to be positive because the condition τi > 0.85 is satisfied for each of the first three eigenchannels, cf. Fig. 2(d). Indeed this is consistent with the numerics in Fig. 7(a)-(b). As for the remaining vibrational modes the contribu- tions to δG(V ) are rather small compared to the out-of- phase longitudinal mode (←→) as quantified in Tab. I [42]. However, their contributions in the δ S(V ) charac- teristics are -- interestingly -- much more pronounced. As shown in Fig. 7(b) we can identify several inelastic signals in the shot noise of comparable order of magnitude that were not clearly visible in the conductance. An interest- ing case is the negative jump ∆ S↑↑ due to the excitation of the in-phase transverse vibrational modes (↑↑, doubly degenerate). Due to symmetry this mode does not al- low for intrachannel scattering (0 %) and its effect can therefor not be rationalized in terms of single-channel models [6 -- 8]. In fact, this highlights that the e-ph cou- pling matrix M↑↑ is essentially off-diagonal in the eigen- channel basis. Indeed the jump ∆ S↑↑ is negative despite that τi > 0.85 for the three most transmitting eigen- channels, contrary to the understanding derived from a single-channel picture. The clear signature in the shot noise from the ↑↑ mode is therefore a prominent demon- 0 G ^ u stration that information about the e-ph coupling may be extracted from noise measurements despite that the mode is essentially passive in the corresponding conduc- tance characteristics. We complete our analysis by considering also a repre- sentative case in the tunneling regime (L = 15.50 A) shown in Fig. 7(c)-(d). The δG(V ) and δ S(V ) char- acteristics both exhibit positive jumps of comparable heights. The dominant inelastic features are located at voltages corresponding to ω←→ = 15.7 meV and ω↑↓ = 12.0 meV that are associated with the excitation of the out-of-phase longitudinal mode (←→) and of the two degenerate out-of-phase transverse modes (↑↓ ×2), respectively. According to Tab. I the ←→ mode again gives rise only to intrachannel transitions (99% of the total scattering involves intrachannel scattering for the 10 most transmitting eigenchannels). The e-ph coupling matrix M←→ is therefore almost diagonal in the eigen- channel basis and the physics can again be understood in terms of a summation over independent channels. The fact that the jumps ∆G←→ and ∆ S←→ are both positive is consistent with the picture from single-level models as the condition τi < 0.15 is satisfied for all eigenchannels, cf. Fig. 2(d). Considering the out-of-phase transverse modes (↑↓ ×2) we again find that the main inelastic processes are inter- channel transitions (see Table I) and therefore that the e-ph coupling matrix M↑↓×2 is off-diagonal in the eigen- channel basis. In essence, transitions occur between the σ-type and π-type states shown in Fig. 3 as expected for a transversal mode due to symmetry. Again, single-level models are inadequate to describe the physics and there- fore do not predict the sign of the jumps ∆G↑↓ and ∆ S↑↓. According to our numerics we find that both jumps are positive. IV. RESULTS FOR AU ATOMIC CHAINS Au atomic chains constitute benchmark systems for the exploration of inelastic signatures in the transport characteristics [15, 29, 30, 35, 43] and most recently the effect of phonon emission in the electronic shot noise was reported for the first time [13]. To complete our study we present in this section realistic calculations of inelas- tic signals in the δG(V ) and δ S(V ) characteristics for a set of Au atomic chain geometries, shown in Fig. 8, corresponding to different lengths and strain conditions. Our calculations were carried out in the same way as described in Sec. III for the Au and Pt atomic contacts, except that now we allow all atoms bridging the Au(111) surfaces to vibrate, i.e., the dynamical region consists of 15 vibrating atoms. Again, for each electrode separation L we relax the dynamical atoms and the topmost Au(111) layer until the residual forces are smaller than 0.02 eV/A. All the structures considered here are stable in the sense that ωλ > 0 for all vibrational modes (no imaginary frequency modes exist in the calculations). 11 In order for us to explore the location of a possible sign change in δ S(V ) we intended to generate very different geometries with transmission factors spanning as large an interval as possible. For all the considered structures, shown in Fig. 8, we obtained the range τ = 0.911 − 1.02 which is significantly more narrow than reported in the experiments of Ref. [13]. Within our treatment it appears difficult to construct geometries where the transmission deviates significantly from unity (unless the chain is bro- ken). The generic behavior for the Au atomic chains -- such as the plain red curve for the geometry L = 27.00 A in Fig. 9(a) (for which τ = 0.991) -- is a main inelas- tic threshold in the conductance curve δG(V ) located at voltages eV ≈ 10 meV. The corresponding negative jump in conductance is due to inelastic backscattering events associated with the activation of the out-of-phase longitudinal (alternating bond length) vibrational mode of the atomic chain (the zone-boundary phonon mode with wave length q ≈ 2kF ) [35, 43]. This active mode is sketched qualitatively with arrows in Fig. 8. The size of the jump is of order 1%, which is significantly larger than the corresponding signals in the case of Au contacts, cf. Fig. 4(a). This is consistent with tight-binding models [30, 44] which have shown that the inelastic conductance correction is roughly proportional to the number of vi- brating atoms in the chain. When considering the noise properties one observes a positive jump of similar mag- nitude in the δ S(V ) characteristics as seen in Fig. 9(b). In order to compare the output of our numerical calcu- lations to the experimental results of Kumar et al. [13], we represented in Fig. 9(c) the total shot noise charac- teristics S(V ) = S0(V ) + δS(V ) expressed in the reduced scale Y (V ) = (S(V )−S(V = 0))/S(V = 0), as a function of the reduced parameter X(V ) = βeV /2 coth(βeV /2). As in Ref. [13], the Y (X) curves are well approximated by piecewise linear functions. When eV is below the main inelastic threshold, the slope of the Y (X) curve gives the Fano factor F−. Upon phonon excitation, the slope changes to F+ thus defining a modified Fano factor (due to the inelastic correction to shot noise δSinel(V ) [Eq. (33)]) and a relative jump of Fano factor ∆F/F = (F+ − F−)/F− (see Ref. [13]). We represent on Fig. 9(d) the dependence of the relative jump of Fano factor ∆F/F (red points) as a function of the renormalized transmis- sion factor τ ≈ τ +Pλ Tr{T(0) λ } (which coincides with the zero bias conductance). Each point is associated to a given geometry shown in Fig. 8 and is computed by nu- merical differentiation of the corresponding Y (X) curve. The experimental points from Kumar et al. [13] are shown in the range τ = 0.9 − 1.0 as blue crosses. We first remark that for transmission factors τ > 0.95 the computed points agree quantitatively with the ex- perimental ones. This exemplify the predictive power of ab initio methods for generating quantitative results for transport calculations without any adjustable parameter. Although our results are fully consistent with the experi- mental data from Ref. [13] in the regime τ > 0.95, we are L = 18.20 A L = 19.20 A L = 20.00 A L = 20.20 A L = 20.50 A L = 21.00 A 12 L = 30.20 A L = 27.20 A L = 27.00 A L = 26.50 A L = 22.00 A L = 21.20 A PSfrag replacements FIG. 8. (Color online) Stable Au atomic chain geometries considered in the first-principles transport calculations. The characteristic electrode separation L, measured between the second-topmost surface layers, is indicated over each geometry. The active out-of-phase longitudinal (alternating bond length) vibrational mode is sketched qualitatively with arrows in one case. unable to identify a sign change in the correction to the shot noise, which was reported experimentally to occur around τ ≈ 0.95 instead of the expected theoretical value τ ≈ 0.85 (see Fig. 9(d) blue crosses). In order to try to extrapolate the position of the crossover as obtained in the output of our calculations, we summarize in Fig. 10(a) the dependence of the size of the total jumps ∆G and ∆ S as a function of the transmis- sion factor τ of the Au atomic chains. This plot clearly shows that statistically the jumps in conductance ∆G (blue points) and shot noise ∆ S (red points) have oppo- site sign and very similar magnitude. The more detailed correlation between ∆G and ∆ S can be characterized by considering the ratio ∆ S/e∆G as shown in Fig. 10(b). Here it is observed that the absolute value of the ratio tends to increase with τ . However, the computed ratios (red points) do not fall exactly on the prediction of an- alytical models (black dashed curve) such as the result in Eq. (44). Rather it appears as if the analytics is an upper bound to the computed ratios [45]. The data that agrees the best with the analytic ratio corresponds to the cases of even-numbered, linear atomic chains L = 27.00 A and L = 27.20 A. Our data in Fig. 10(b) actually suggests that the crossover could occur at a lower value than the τ ≈ 0.85 value from analytic models. An interesting future exten- sion of our results would be to try to include disorder- induced conductance fluctuations in the treatment, as this effect played an important role in the interpreta- tion of the experiments in Ref. [13]. Such a development with a semi-empirical model of disorder could be achieved through a multi-scale methodology [46], but at the cost of giving up the parameter-free scheme presented here for simulating the inelastic transport characteristics of nanodevices. V. CONCLUSIONS We have implemented into the Inelastica [14 -- 16] code a methodology [19, 20] that allows to compute quan- titatively inelastic shot noise signals from first princi- Au-nanowire L20.50 Ž L21.00  L26.50  L27.00 ‘ L27.20 ’ =0.997 =0.977 =0.992 =0.991 =0.985 Au-nanowire =0.997 =0.977 =0.992 =0.991 =0.985 L20.50 • L21.00 – L26.50 — L27.00 ˜ L27.20 ™ (b) 1.5 1.2 ) 0.9 3 e 2 h 1 0 . 0 ( /” S 0.6 0.3 0.0 -0.3 (a) 3.0 ) 3 e 2 h 2.0 1 0 . 0 ( S ¥ & ) 1.0 0.0 2 e 2 h 1 0 . 0 ( G 0.03 -1.0 -2.0 0.01 0.02 V (eV) Au-nanowire =0.997 =0.993 =0.987 =0.978 L20.50  L27.00 ž L27.20 Ÿ L21.00 0.01 0.02 V (eV) Au-nanowire Exp L20.50 L21.20 L22.00 L26.50 L27.00 L27.20 L20.00 L21.00 0.03 0.0 (d) 350 ) % ( F / F 300 250 200 150 100 50 0 (a) 0.3 0.1 ) -0.1 2 e 2 h 1 0 . 0 ( / G -0.3 -0.5 -0.7 -0.9 -1.1 0.0 (c) 0.7 0.6 0.5 ) 0 ( S / } ) 0 ( S 0.4 0.3 )œ V ( S { = 0.2 ) V ( Y 0.1 0.0 1 6 X = š 11 16 21 26 31 eV/2) eV/2coth( › 0.9 0.92 0.94 0.96 0.98 1.0 ¡ FIG. 9. (Color online) Inelastic conductance and noise char- acteristics for Au atomic chains shown in Fig. 8 in the regime of equilibrated phonons. (a) δG/τ (V ) and (b) δ S/τ (V ) char- acteristics computed at T = 4.2 K. (c) Total shot noise S(V ) = S0(V )+δS(V ) expressed in the reduced scale Y (V ) = (S(V ) − S(V = 0))/S(V = 0) as a function of the reduced pa- rameter X(V ) = βeV /2 coth(βeV /2) computed at T = 4.2 K. The dashed curves are the extrapolation of the low voltage re- duced noise characteristics Y−(X) = F−(X − 1), with F− the Fano factor (see the text for details). (d) The relative jump of Fano factor ∆F/F = (F+ −F−)/F− (red points) as a function of the renormalized transmission factor τ ≈ τ + Pλ Tr{T(0) λ }, as extracted numerically from the Y (X) characteristics. For comparison, the blue crosses are the experimental results of Kumar et al. [13]. ples. The method was illustrated for Au and Pt atomic point contacts as well as Au atomic chains described at the atomistic level. We showed that the Au con- tact constitutes a benchmark system that can be un- derstood in terms of the "single level, single vibrational mode" models [6 -- 8]. To be quantitative we rationalized computed inelastic corrections to the conductance and shot noise characteristics in terms of a "two-site" tight- binding model for the vibrating Au contacts. However, in the case of Pt contacts the multichannel nature compli- cates the physics and simple analytic models were shown Au-nanowire Au-nanowire 13 L18.20 L19.20 L20.20 L30.20 L20.50 L21.20 L22.00 L26.50 L27.00 L27.20 L20.00 L21.00 (b) -0.3 -0.4 -0.5 -0.6 G -0.7 e / S -0.8 -0.9 -1.0 -1.1 -1.2 0.9 0.94 0.98 1.02 0.9 0.94 0.98 1.02 £ ¦ FIG. 10. (Color online) Inelastic conductance and noise cor- rections for Au atomic chains shown in Fig. 8 in the regime of equilibrated phonons. (a) Total jumps ∆G (blue points) and ∆ S (red points) at zero temperature as a function of the transmission factor. (b) The ratio ∆ S/e∆G (red points) as a function of the transmission factor. The black dashed curve is the prediction of analytical models for the ratio, cf. Eq. (44). to be inadequate. Interestingly, we found that for Pt con- tact the coupling to transverse vibrational modes gives rise to features in the inelastic shot noise signals (origi- nating in interchannel scattering processes) that are not readily visible in the conductance. This opens up an al- ternative approach to characterize the e-ph couplings and the underlying multichannel electronic transport prob- lem. We also analyzed Au atomic chain configurations in order to compare directly with the experimental results of Ref. [13] for the inelastic effects in the shot noise. While the simulated shot noise behavior was found to agree quantitatively with the experiment in the ballistic regime (0.95 < τ ≈ 1) we were unable to identify the sign change in the shot noise correction that was reported in Ref. [13]. The techniques for characterizing inelastic effects in shot noise appears as a promising way to gain deeper insight into transport properties of nanoscale devices. It is our hope that the methodology presented in this paper will stimulate further research in this direction. ACKNOWLEDGEMENTS The authors are grateful to Federica Haupt, Jan M. van Ruitenbeek, and Alfredo Levy Yeyati for their careful reading and comments on the manuscript. Œ “ ¢ ¤ § 14 [1] V. L. Gurevich and A. M. Rudin, Phys. Rev. B 53, 10078 [27] M. P. L. Sancho, J. M. L. Sancho, and J. Rubio, (Apr 1996) [2] J.-X. Zhu and A. V. Balatsky, Phys. Rev. B 67, 165326 (Apr 2003) J. Phys. F: Met. Phys. 15, 851 (1985) [28] P.-O. Lowdin, J. Chem. Phys. 18, 365 (1950) [29] J. K. Viljas, J. C. Cuevas, F. Pauly, and M. Hafner, Phys. [3] B. Dong, H. L. Cui, X. L. Lei, and N. J. M. Horing, Rev. B 72, 245415 (2005) Phys. Rev. B 71, 045331 (Jan 2005) [30] L. de la Vega, A. Martin-Rodero, N. Agraıt, and [4] M. Galperin, A. Nitzan, and M. A. Ratner, Phys. Rev. A. Levy Yeyati, Phys. Rev. B 73, 075428 (FEB 2006) B 74, 075326 (2006) [5] J. T. Lu and J.-S. Wang, Phys. Rev. B 76, 165418 (2007) [6] T. L. Schmidt and A. Komnik, Phys. Rev. B 80, 041307 [31] M. Buttiker, Phys. Rev. Lett. 57, 1761 (Oct 1986) [32] D. S. Fisher and P. A. Lee, Phys. Rev. B 23, 6851 (Jun 1981) (Jul 2009) [33] R. Egger and A. O. Gogolin, Phys. Rev. B 77, 113405 [7] R. Avriller and A. Levy Yeyati, Phys. Rev. B 80, 041309 (Mar 2008) (Jul 2009) [34] O. Entin-Wohlman, Y. Imry, and A. Aharony, [8] F. Haupt, T. Novotn´y, and W. Belzig, Phys. Rev. Lett. Phys. Rev. B 80, 035417 (Jul 2009) 103, 136601 (Sep 2009) [35] T. Frederiksen, M. Brandbyge, N. Lorente, and A.-P. [9] Y.-C. Chen and M. Di Ventra, Phys. Rev. Lett. 95, Jauho, Phys. Rev. Lett. 93, 256601 (2004) 166802 (Oct 2005) [36] M. Engelund, M. Brandbyge, and A.-P. Jauho, Phys. [10] D. F. Urban, R. Avriller, and A. Levy Yeyati, Rev. B 80, 045427 (2009) Phys. Rev. B 82, 121414 (Sep 2010) [37] Y. Blanter and M. Buttiker, Physics Reports 336, 1 [11] T. Novotn´y, F. Haupt, and W. Belzig, Phys. Rev. B 84, (2000) 113107 (Sep 2011) [38] T. Frederiksen, N. Lorente, M. Paulsson, and M. Brand- [12] T.-H. Park and M. Galperin, Phys. Rev. B 84, 205450 byge, Phys. Rev. B 75, 235441 (Jun. 2007) (Nov 2011) [39] M. Paulsson and M. Brandbyge, Phys. Rev. B 76, 115117 [13] M. Kumar, R. Avriller, A. L. Yeyati, and J. M. van (2007) Ruitenbeek, Phys. Rev. Lett. 108, 146602 (Apr 2012) [14] M. Paulsson, T. Frederiksen, and M. Brandbyge, Phys. Rev. B 72, 201101 (NOV 2005), ISSN 1098-0121 [40] M. Paulsson, T. Frederiksen, H. Ueba, N. Lorente, and M. Brandbyge, Phys. Rev. Lett. 100, 226604 (Jun 2008) [41] A. Garcia-Lekue, D. Sanchez-Portal, A. Arnau, and [15] T. Frederiksen, M. Paulsson, M. Brandbyge, and A.-P. T. Frederiksen, Phys. Rev. B 83, 155417 (2011) Jauho, Phys. Rev. B 75, 205413 (May 2007) [16] The Inelastica software freely http://inelastica.sourceforge.net/ is available at [17] J. M. Soler, E. Artacho, J. D. Gale, A. Garc´ıa, J. Jun- quera, P. Ordej´on, and D. S´anchez-Portal, Journal of Physics: Condensed Matter 14, 2745 (2002) [18] M. Brandbyge, J. L. Mozos, P. Ordej´on, J. Taylor, and K. Stokbro, Phys. Rev. B 65, 165401 (2002) [42] We note that the observation of transverse modes in the conductance signal was reported in Ref. [26] for multi- channel atomic-size contacts of aluminum. As in our case the excitation of transverse modes appeared as a positive correction to δG(V ), while the excitation of longitudinal modes appeared as a negative correction in the contact regime. [43] N. Agraıt, C. Untiedt, G. Rubio-Bollinger, and S. Vieira, [19] A. O. Gogolin and A. Komnik, Phys. Rev. B 73, 195301 Phys. Rev. Lett. 88, 216803 (2002) (May 2006) [44] T. Frederiksen, M. Brandbyge, N. Lorente, and A.-P. [20] F. Haupt, T. Novotn´y, and W. Belzig, Phys. Rev. B 82, Jauho, J. Comput. Electron. 3, 423 (2004) 165441 (Oct 2010) [21] N. Agraıt, A. Levy Yeyati, and J. M. van Ruitenbeek, Phys. Rep. 377, 81 (2003) [22] C. Untiedt, M. J. Caturla, M. R. Calvo, J. J. Palacios, R. C. Segers, and J. M. van Ruitenbeek, Phys. Rev. Lett. 98, 206801 (2007) [23] M. Brandbyge, M. R. Sørensen, and K. W. Jacobsen, Phys. Rev. B 56, 14956 (1997) [24] E. Scheer, N. Agraıt, J. C. Cuevas, A. Levy Yeyati, B. Lu- doph, A. Martin-Rodero, G. Rubio-Bollinger, J. M. van Ruitenbeek, and C. Urbina, Nature 394, 154 (1998) [25] J. C. Cuevas, A. L. Yeyati, and A. Martin-Rodero, Phys. Rev. Lett. 80, 1066 (1998) [45] We also studied a tight-binding model of an N -site atomic chain that couples to the longitudinal vibrational mode with alternating bond-length character. For an even number of sites (N = 2p) the model has the same output as the simple Eq. (44), whereas for odd number of sites (N = 2p + 1) we observed a different qualitative be- havior of the inelastic signals where the total jump ∆G is always negative (independent of the transmission τ ) and ∆ S changes sign at τ = 0.5 instead of at τ ≈ {0.15, 0.85}. We assign this even-odd effect with N to the vanishing vertex correction to the shot-noise due to the symmetry of the particular electron-phonon matrix elements. [46] R. Avriller, S. Roche, F. Triozon, X. Blase, and S. Latil, [26] T. Bohler, A. Edtbauer, and E. Scheer, New J. Phys 11, Mod. Phys. Lett. 21, 1955 (2007) 013036 (JAN 23 2009) 

1803.01315

3

1803

Systematic derivation of realistic spin-models for beyond-Heisenberg solids

We present a systematic derivation of effective lattice spin Hamiltonians derived from a rotationally invariant multi-orbital Hubbard model including a term ensuring Hund's rule coupling. The Hamiltonians are derived down-folding the fermionic degrees of freedom of the Hubbard model into the proper low-energy spin sector using L\"owdin partitioning, which will be outlined in detail for the case of two sites and two orbitals at each site. Correcting the ground state systematically up to fourth order in the hopping of electrons, we find for spin $S\geq 1$ a biquadratic, three-spin and four-spin interaction beyond the conventional Heisenberg term. Comparing the puzzling energy spectrum of the magnetic states for a single Fe monolayer on Ru(0001), obtained from density functional theory, with the spin Hamiltonians taken at the limit of classical spins, we show that the previously ignored three-spin interaction can be comparable in size to the conventional Heisenberg exchange.

cond-mat.mes-hall

cond-mat

Systematic derivation of realistic spin-models for beyond-Heisenberg solids Peter Grunberg Institut and Institute for Advanced Simulation, Forschungszentrum Julich and JARA, 52425 Julich Germany (Dated: June 18, 2021) Markus Hoffmann∗ and Stefan Blugel We present a systematic derivation of effective lattice spin Hamiltonians derived from a rotation- ally invariant multi-orbital Hubbard model including a term ensuring Hund's rule coupling. The Hamiltonians are derived down-folding the fermionic degrees of freedom of the Hubbard model into the proper low-energy spin sector using Lowdin partitioning, which will be outlined in detail for the case of two sites and two orbitals at each site. Correcting the ground state systematically up to fourth order in the hopping of electrons, we find for spin S ≥ 1 a biquadratic, three-spin and four-spin interaction beyond the conventional Heisenberg term. Comparing the puzzling energy spectrum of the magnetic states for a single Fe monolayer on Ru(0001), obtained from density func- tional theory, with the spin Hamiltonians taken at the limit of classical spins, we show that the previously ignored three-spin interaction can be comparable in size to the conventional Heisenberg exchange. I. INTRODUCTION Magnetic interactions have captivated several gener- ations of condensed matter physicists because of their diversity of physical origins in very different solids, the emergence of a vast spectrum of magnetic structures as a result of their competition and subsequently the many interesting physical phenomena that are arising from those magnetic structures [1 -- 3]. Antiferromagnets with noncoplanar spin-textures and topological magnetization solitons such as skyrmions are current examples of com- plex magnetic structures with a broad spectrum of exotic properties that are of interest for both basic research and applications in spintronics [4]. Understanding the prop- erties of these novel spin-textures has revitalized the field of magnetic interactions. In this context itinerant mag- nets play an important role as the itinerant electrons give rise to these complex magnetic structures and in turn the complex magnetic structures give rise to interesting transport phenomena [5 -- 7]. In a materials specific context, the theoretical descrip- tions of magnetic ground states as well as the dynami- cal or thermodynamical properties of magnetic systems are often made possible by a realistic spin Hamiltonian typically determined by a multi-scale approach: density functional theory (DFT) calculations are mapped onto a classical lattice spin Hamiltonian, i.e. a lattice of classical spins interacting according to spin-models, whose prop- erties are then evaluated carrying out Monte-Carlo or spin-dynamic simulations [8 -- 16]. That is to say that the materials specificity enters through the parameters of the model determined by DFT. The choice of the spin-model itself reflects the choice of materials and the interactions that seem relevant to understand certain properties. For many bulk as well as application customized multilayer and heterostructure systems, the well-known spin S = 1/2-Heisenberg model [17] of quantum spins S is extrapolated to systems with higher quantum spin, S > 1/2, and very often to classical vector spins S pro- viding a parameterization of an effective spin Hamilto- nian successful in describing the required magnetic prop- erties. This holds also true for metallic magnetic ma- terials, in particular those for which the longitudinal spin-fluctuations are unimportant as compared to the transversal ones. These are typically magnets of tran- sition metals with atomic spin moments in the order of 2 µB and more such as for Mn, Fe, Co in their bulk phases, as alloys and multilayers commonly used in spin- tronic devices. In fact, describing typical properties of those magnetic metals one resorts to the classical Heisenberg model of bilinear exchange interactions of the form H1 = −(cid:88)(cid:48) Jij Si · Sj (1) ij between pairs of classical spins S at different lattice sites i, j with exchange interactions Jij whose signs and strengths depend on details of the electronic structure. The spatial dependence of the exchange interaction fol- lows typically the crystal anisotropy imposed by the crys- tal lattice. For metals the Jij can be long-ranged and in part determined by the topology of the Fermi surface, in opposite to insulators, where they are typically short- ranged. A success of this approach is for example the prediction of magnetic structures consistent to experi- ments [18] or the Curie temperatures of bulk ferromag- nets [19, 20]. The minus sign in (1) is just a convention we follow for all spin lattice Hamiltonians throughout the means here and throughout the paper that we are taking the sum over all possible integer sites i and j except for any summations of two equal sites i = j. paper. The notation(cid:80)(cid:48) There are, however, well-known cases where the Heisenberg model is insufficient to describe correctly the magnetic ground state structure or magnon excitations. In these cases [21] one addresses the higher-order spin in- teraction beyond the Heisenberg model. A typical signa- ture of the higher-order spin interaction is the occurrence of particular types of non-collinear states, e.g. canted magnetic states [22] or multi-q states, a superposition of spin-spiral states of symmetry related wave vectors q. A spin-spiral state with a single q-vector [22, 23] is an 9 1 0 2 p e S 7 2 ] l l a h - s e m . t a m - d n o c [ 3 v 5 1 3 1 0 . 3 0 8 1 : v i X r a exact solution of the classical Heisenberg model for a pe- riodic lattice. The higher-order terms couple modes of symmetry equivalent q vectors and can lead to complex magnetic structures of energies lower than the single-q state [24]. One of the most commonly considered extensions of the bilinear Heisenberg form is the addition of the bi- quadratic exchange, a term of the form Bij(Si · Sj)2 . H2 = −(cid:88)(cid:48) (2) ij This term has been motivated by very different mi- croscopic origins, through superexchange [1], magneto- elastic effect [25, 26] or interlayer exchange coupling [27]. Quite generally, according to the algebra of the spin op- erators, any power of scalar products of pairs of quan- tum spins of total spin S at sites i, j, can only have 2S independent powers up to (S i · S j)2S. Thus, for the bi- quadratic term to occur through the interaction of elec- trons requires at least a total spin S = 1 at the lattice sites. As we will see below, as the power of (S i · S j) is related to the order of perturbation theory, the bi- quadratic term [28 -- 36] is the most essential correction to the Heisenberg model for spins S > 1/2 involving two lattice sites. Involving more lattice sites, a systematic extension of the bilinear Heisenberg form is the four-spin interaction, which was derived by Takahashi [37] for a spin 1/2-system treating electrons by a single band Hubbard model. It arises in fourth order perturbation theory of electron hop- ping versus Coulomb interaction [38]. The four-spin in- teraction consists of four-body operators that appear by permuting all spins in a four-membered ring and can be written in the limit of classical spin as Kijkl [(Si · Sj)(Sk · Sl) + (Si · Sl)(Sj · Sk) ijkl −(Si · Sk)(Sj · Sl)] , (3) with the sum over all rings of four sites. Although the higher-order spin models where mostly applied to magnets with localized electrons such as mag- netic insulators [39, 40], comparing DFT results for itin- erant metallic magnets with spin-models reveals their sig- nificance also for these systems. Examples include con- tributions of the biquadratic term to the spin-stiffness of the bulk magnets Fe, Co and Ni [41], the coni- cal spin spirals for a double-layer Mn on W(110) [42], or even three-dimensional non-collinear spin structures on a two-dimensional lattice as in Mn/Cu(111) [24], in Fe/Ir(100) [10] or Fe/Ir(111) [43]. In case of the latter, the 4-spin interaction couples spin spirals with different propagation directions and forms a square lattice of chi- ral magnetic skyrmions of atomic scale size. However, one became recently aware not all systems studied with DFT could be explained purely on the basis of the higher-order interactions discussed above. Two such examples are the theoretically predicted [44] H4 = −(cid:88)(cid:48) 2 and recently experimentally verified [45] so-called up-up- down-down (uudd) state, a multi-q state, in Fe/Rh(111) or a canted uudd state in a RhFe bilayer system on Ir(111) [46]. While an uudd state could in general be stabilized by both considered higher-order interactions independently, the calculated energy spectrum revealed that the main stabilization has to originate from another, hitherto unknown, interaction. Summarizing the spin-models discussed so far we can view the Heisenberg, biquadratic and four-spin model as a two-spin-two-site, four-spin-two-site, and four-spin- four-site interaction, respectively. Heisenberg and four- spin interaction emerge for S = 1/2, the biquadratic one requires at least S = 1. Since typical magnetic Mn, Fe, Co moments at surfaces are in the order of 2 or 3 µB equivalent to S = 1 or S = 3/2, there should be a large number of quasi two-dimensional non-Heisenberg mag- nets, in particular for substrates for which the effective Heisenberg exchange is small due to compensation of Jij of different signs between different neighbors. Further, using this notion of classification, a four- spin-three-site interaction seems missing. Indeed, vari- ous partly phenomenological models of three-spin inter- actions [47] had been proposed or derived to explain ex- periments mostly for insulating magnets [30, 48 -- 53]. In this paper we provide a consistent and system- atic derivation of expressions describing the beyond- Heisenberg higher-order spin interactions resulting from the electron-electron interaction up to the fourth order in the hopping interaction strength of electrons for to- tal spins of size S ≥ 1/2. This includes all possible se- quences of four hopping events of electrons between or- bitals at maximal four sites. The spin-orbit interaction is neglected at this point. The starting point is the rota- tionally invariant multi-orbital Hubbard model assuming half-filling, which will be explained in the next section. The spin-model is derived down-folding the dynamical fermionic degrees of freedom of electrons described by the Hubbard model into the proper low-energy spin sector using Lowdin partitioning [54, 55], which is also known as Schrieffer-Wolf transformation [56, 57]. The Lowdin partitioning is briefly sketched for a dimer of S = 1-spins described by two electron orbitals at both sites. Then, we will present our results for different numbers of sites and orbitals and also for lattices with different space groups like a square lattice as for example for magnetic atoms on a (001)-surface of a fcc crystal, or on a hexagonal lattice like the (111)-surface to adapt the theoretical approach to real systems. Taking the classical spin-limit of the quantum spin-models derived, we reproduce the known spin Hamiltonians above plus the missing three-spin in- teraction H3 = −2 Yijk(Si · Sj)(Sj · Sk) , (4) ijk where the sum goes over triangles of sites. At the end, we will analyze the energy spectrum for various magnetic structures determined by density func- (cid:88)(cid:48) FIG. 1. Schematic representation of the two investigated models. (a) The multi-band Hubbard model. A periodic arrangement of atoms on a lattice is shown. The different orbitals (here two) are illustrated by gray planes located at each atom. Each orbital can host up to two electrons, one spin-up (shown in red) and one spin-down (blue). Additionally, the hopping paths are indicated by green arrows and sites with non-zero on-site energies (proportional to U , U(cid:48) and JH ) are highlighted by green spheres. (b) The extended Heisenberg model in the limit of classical spins (gray arrows) at each lattice site. Direct exchange (Si · Sj) is illustrated by colored arrows. Higher-order interactions couple two of them to form 4-spin interactions involving two (B), three (Y) or four (K) sites as indicated by the springs. 3 t(cid:48) i,α,j,α(cid:48) (c † i,α,σcj,α(cid:48),σ + h.c.) Ui,α ni,α,↑ ni,α,↓ U(cid:48) i,α,α(cid:48) (ni,α,σ ni,α(cid:48),σ + ni,α,σ ni,α(cid:48),¯σ) i,α + + i<j,σ α(cid:54)=α(cid:48) − (cid:88) (cid:88) (cid:88) − (cid:88) − (cid:88) − (cid:88) i,σ α<α(cid:48) i,σ α<α(cid:48) i,α<α(cid:48) i,α<α(cid:48) tional theory for a single Fe monolayer on Ru(0001). Subsequently we will show that the hitherto puzzling re- sults [44] can finally be understood. II. METHODOLOGY A. Multi-band Hubbard model In this section we briefly introduce the Hamiltonian, from which we start our derivations, and define the most important parameters of our model. In the following sec- tion we will then focus on reducing the inherent degrees of freedom of the Hamiltonian to the spin degrees of free- dom in order to derive effective spin models. This Hamil- tonian will then be used to generate spin Hamiltonians for different systems that vary by the number of sites and orbitals and also by the lattice type. Earlier similar investigations [37, 38] typically used the one-band Hubbard model [58 -- 60] as a starting point since it is the simplest model for describing interacting elec- trons on a lattice. For practical magnetic systems, which we have in mind with typical magnetic spin moments on the order of 2 or 3 µB (S = 1 or S = 3/2), we extend our investigation to systems with more than one orbital per site (e.g. d-orbitals of transition metals). Therefore, we work with a generalized Hubbard Hamiltonian, which not only includes the additional hopping terms and Coulomb interactions, but contains also additional terms to ensure Hund's rule coupling. The Hund's terms are included as we are interested in states with a fixed and stable mag- netic moment S per atom or site: H = − (cid:88) i<j,α,σ (cid:16) ti,α,j,α † i,α,σcj,α,σ + h.c. c (cid:17) Ji,α,α(cid:48) ni,α,σ ni,α(cid:48),σ (cid:16) (cid:16) Ji,α,α(cid:48) J(cid:48) i,α,α(cid:48) † † i,α,↑ci,α,↓c i,α(cid:48),↓ci,α(cid:48),↑ + h.c. c † † i,α,↓ci,α(cid:48),↓ + h.c. i,α,↑ci,α(cid:48),↑c c (cid:17) (cid:17) . (5) Here, i and j represent the atomic sites, α and α(cid:48) stand for the orbitals and σ denotes the quantization of the spin † projection of the electron (↑ or ↓). ni,α,σ = c i,α,σci,α,σ defines the number of electrons at site i in orbital α with spin σ. t (t(cid:48)) describes the hopping amplitude between two different sites of the same (different [61]) orbital types. The on-site hopping between different orbitals is not considered as we assume the orbitals to be orthogo- nal with respect to each other (t(cid:48) i,α,i,α(cid:48) = 0). Fig. 1 shows a schematic visualization of the Hubbard as well as the effective spin model. Only on-site Coulomb interactions are taken into ac- count throughout the paper. Having a periodic solid in mind with only one atom type, we assume that the intra-orbital Coulomb interaction between electrons of the same orbitals α is the same for each site, Ui,α = U , as well as the inter-orbital Coulomb interaction between U′-JH U′ U t′ t K B Y J (a) (b) electrons in different orbitals, U(cid:48) i,α,α(cid:48) = J(cid:48) Ji,α,α(cid:48) = JH and J(cid:48) sence of the site dependency. i,α,α(cid:48) = U(cid:48). Analogously, H simplifies due to the ab- B. Lowdin partitioning Here we briefly explain how Lowdin partitioning [54, 55] is used to derive an effective spin Hamiltonian. As an example, we take the smallest interacting system with more than one orbital per site, two sites with two or- bitals each. Assuming half-filled orbitals, we deal with four electrons, that could be distributed among the four available orbitals. Thus, an orbital, s(cid:105), can be occu- pied with s equal to one or two electrons or it can be unoccupied, denoted as ·(cid:105). The possible states sorted according to the angular momentum quantum number m, representing the z-component of the total spin of the system include the following product states: m = 2 :↑,↑,↑,↑(cid:105) m = 1 :↑,↑,↑,↓(cid:105) ,↑,↑,↓,↑(cid:105) ,↑,↓,↑,↑(cid:105) ,↓,↑,↑,↑(cid:105) , ↑↓,↑,↑,·(cid:105) ,↑,↑↓,↑,·(cid:105) ,↑,↑,↑↓,·(cid:105) ,↑↓,↑,·,↑(cid:105) , ↑,↑↓,·,↑(cid:105) ,↑,↑,·,↑↓(cid:105) ,↑↓,·,↑,↑(cid:105) ,↑,·,↑↓,↑(cid:105) , ↑,·,↑,↑↓(cid:105) ,·,↑↓,↑,↑(cid:105) ,·,↑,↑↓,↑(cid:105) ,·,↑,↑,↑↓(cid:105) m = 0 :↑,↑,↓,↓(cid:105) ,↑,↓,↑,↓(cid:105) ,↑,↓,↓,↑(cid:105) ,↓,↑,↑,↓(cid:105) , ↓,↑,↓,↑(cid:105) ,↓,↓,↑,↑(cid:105) ,↑↓,↑,↓,·(cid:105) ,↑↓,↑,·,↓(cid:105) , ↑,↑↓,↓,·(cid:105) ,↑,↑↓,·,↓(cid:105) ,↑↓,↓,↑,·(cid:105) ,↑↓,·,↑,↓(cid:105) , ↑,↓,↑↓,·(cid:105) ,↑,·,↑↓,↓(cid:105) ,↑↓,↓,·,↑(cid:105) ,↑↓,·,↓,↑(cid:105) , ↑,↓,·,↑↓(cid:105) ,↑,·,↓,↑↓(cid:105) ,↓,↑↓,↑,·(cid:105) ,↓,↑,↑↓,·(cid:105) , ·,↑↓,↑,↓(cid:105) ,·,↑,↑↓,↓(cid:105) ,↓,↑↓,·,↑(cid:105) ,↓,↑,·,↑↓(cid:105) , ·,↑↓,↓,↑(cid:105) ,·,↑,↓,↑↓(cid:105) ,↓,·,↑↓,↑(cid:105) ,↓,·,↑,↑↓(cid:105) , ·,↓,↑↓,↑(cid:105) ,·,↓,↑,↑↓(cid:105) ,↑↓,↑↓,·,·(cid:105) ,↑↓,·,↑↓,·(cid:105) , ↑↓,·,·,↑↓(cid:105) ,·,↑↓,↑↓,·(cid:105) ,·,↑↓,·,↑↓(cid:105) ,·,·,↑↓,↑↓(cid:105) m = −1 :↑,↓,↓,↓(cid:105) ,↓,↑,↓,↓(cid:105) ,↓,↓,↑,↓(cid:105) ,↓,↓,↓,↑(cid:105) , ↑↓,↓,↓,·(cid:105) ,↑↓,↓,·,↓(cid:105) ,↑↓,·,↓,↓(cid:105) ,↓,↑↓,↓,·(cid:105) , ↓,↑↓,·,↓(cid:105) ,·,↑↓,↓,↓(cid:105) ,↓,↓,↑↓,·(cid:105) ,↓,·,↑↓,↓(cid:105) , ·,↓,↑↓,↓(cid:105) ,↓,↓,·,↑↓(cid:105) ,↓,·,↓,↑↓(cid:105) ,·,↓,↓,↑↓(cid:105) m = −2 :↓,↓,↓,↓(cid:105) (6) Here, s1, s2, s3, s4(cid:105) = s1(cid:105)s2(cid:105)s3(cid:105)s4(cid:105) means that at site 1 the first (second) orbital is occupied by s1 (s2) and at site 2 the first (second) orbital is occupied by s3 (s4). In general, for a system with n orbitals, the number of states for each value of m is given by(cid:0) (cid:1)2 n n/2+m . Since the z-component of the angular momentum vec- tor operator Sz commutes with the Hamiltonian (5), the Hamiltonian block-diagonalizes in separate subspaces of different m, and the matrix representation of (5) can be calculated for each subspace separately. To support our goal of contracting Hamiltonian (5) of our model to an effective spin Hamiltonian, it is convenient to change the product basis s1, s2, s3, s4(cid:105) to one where the total spin at any site is a good quantum number. For example, for m = 1 the first 4 states are replaced by the following superpositions: 4 (↑,↑,↑,↓(cid:105) + ↑,↑,↓,↑(cid:105)) = 1, 1(cid:105)1, 0(cid:105) (↑,↓,↑,↑(cid:105) + ↓,↑,↑,↑(cid:105)) = 1, 0(cid:105)1, 1(cid:105) (↑,↑,↑,↓(cid:105) − ↑,↑,↓,↑(cid:105)) = 1, 1(cid:105)0, 0(cid:105) (↑,↓,↑,↑(cid:105) − ↓,↑,↑,↑(cid:105)) = 0, 0(cid:105)1, 1(cid:105) , (7) 1√ 2 1√ 2 1√ 2 1√ 2 where we used the notation S1, m1(cid:105)S2, m2(cid:105) with Si be- ing the spin quantum number and mi being the total z-component at site i. We are essentially interested in the subspace spanned by the first two states of (7) as we assume magnetic systems, which have constant magnetic moments (here, S = 1) at each site. Although there is no direct inter- action between these two states, there are indirect inter- actions across states where S is not equal at all sites. These indirect interactions between intermediate states in different subspaces can be downfolded into the sec- tor of interacting spins of constant quantum number at each site using the so-called Lowdin partitioning [54, 55]. Lowdin partitioning can be used because we are dealing with energetically well separated subspaces of spins with different S. This is a consequence of Hund coupling and on-site Coulomb energies that are large with respect to the hopping parameters, as we are discussing transition metals here. The Lowdin partitioning is a tool to decouple these subspaces pertubatively and to map the indirect inter- action between two states of the same subspace over states of the other subspaces to direct interactions be- tween these states with increasing order of the per- turbation. E.g., the indirect interaction ↑,↑,↓,↓(cid:105) ∼t←→ ↑,·,↓,↑↓(cid:105) ∼t←→ ↑,↓,↓,↑(cid:105) is mapped on a direct interac- tion ↑,↑,↓,↓(cid:105) ∼t2/U←−−−→ ↑,↓,↓,↑(cid:105) if terms up to at least second order are taken into account in the Lowdin par- titioning. By going to higher orders also indirect inter- actions including more than two hopping events are con- sidered. These can then relate to interactions with more than two sites. tonian H into two parts, Mathematically, this is achieved by dividing the Hamil- H = H0 + H(cid:48) = H0 + H1 + H2 , (8) a term H0 that contains the on-site contributions, i.e. the repulsive Coulomb interaction and the Hund exchange, and a term H(cid:48), which contains the off-diagonal matrix elements due to the electron hopping, which are treated as a perturbation. Here, H1 contains those terms whose matrix elements couple within the subspaces, whereas H2 describes the coupling between them. The subspaces are decoupled through a canonical transformation [56, 57] H = e− SHe S , (9) where hermiticity of the Hamiltonian implies S† = − S and the generator S of the transformation is chosen such that H becomes block-diagonal. This is achieved writing Eq. (9) in the form of successive applications of commu- tator rules ∞(cid:88) k=0 1 k! H = (cid:104)H, S(cid:105)k = ∞(cid:88) k=0 1 k! (cid:104)H0 + H1 + H2, S(cid:105)k , (10) with [ A , B ]k = [ [ . . . [ [ A , B ] , B ] . . . , B ] , B ] nested k times. Considering the definitions of H1 and H2 [55], this allows then to decouple Eq. (10) into a Hamiltonian term Hd, whose matrix representation is block diagonal and a term Ho with off-block-diagonal matrix elements as shown here: Hd = + 1 1 (cid:104)H0 +H1, S(cid:105)2k ∞(cid:88) (cid:104)H0 +H1, S(cid:105)2k+1 k=0 (2k)! 1 (2k + 1)! (cid:104)H2, S(cid:105)2k+1 (cid:104)H2, S(cid:105)2k (2k + 1)! ∞(cid:88) k=0 + 1 (2k)! ∞(cid:88) ∞(cid:88) k=0 k=0 Ho = (11) The requirement of block diagonalization or Ho = 0, re- spectively, up to a given order k in the perturbation de- termines the generator S and subsequently the effective Hamiltonian Hd. Due to the block-diagonalization of H with respect to the basis of S z, the Lowdin partitioning can be carried out independently for each angular mo- mentum quantum number m. We work out all spin mod- els for either m = 0 or m ± 1/2, depending the systems have integer or half-integer total spins, since these states denote the largest subspaces, and the Lowdin partition- ing becomes least degenerate and the functional forms of the spin Hamiltonians become most obviously distinct. 5 can be expressed by spin operators and thus represents the corresponding spin Hamiltonian or spin model of the system. In the following we present results up to fourth- order perturbation in (11) which permits the investiga- tion of interactions between 2, 3, and 4 sites. We start with spin S = 1/2, i.e. exactly one orbital per site, and then move to S ≥ 1. 1. Spin S=1/2 a. 2 sites, spin S = 1/2 To demonstrate the gen- eral procedure, we first discuss a S = 1/2 dimer, i.e., two sites and only one orbital per site. For m = 0, the Hilbert space is spanned by four possible states ↑,↓(cid:105) ,↓,↑(cid:105) ,↑↓,·(cid:105) ,·,↑↓(cid:105), from which the first two span the subspace of interest with S = 1/2 at both sites. H0 gives the same on-site energies for both states, which we consider the origin of our energy scale. Going up to sec- ond order in the perturbation (the first order vanishes, because there is no direct coupling between the states) additional terms occur which couple the states. Those † † 1,↓c1,↑c2,↓ + h.c., rep- terms are, e.g. proportional to c 2,↑c resenting a hopping ↑,↓(cid:105) ↔ ↓,↑(cid:105). Collecting all those terms and extending the derivation to an infinite lattice of two-site interactions, the resulting Hamiltonian can be written in terms of the spin operators (13) as H2 sites 2ndorder = = 4 t2 U 2 t2 U (cid:88)(cid:48) (cid:88)(cid:48) ij ij c (cid:18) † † j,↓cj,↑ − ni,↑ nj,↓ i,↑ci,↓c S i · S j − ni nj 4 (14) , (cid:19) III. RESULTS A. Derived spin Hamiltonians Recalling that the spin operators S i = (Si,x,Si,y,Si,z) with ni = (ni,↑ + ni,↓) being the total number opera- tor with expectation value ni for electrons at site i. As we only consider the low-energy subspace, charge exci- tations are neglected, and only states with half-filled or- bitals giving rise to maximal S are considered. Thus, the last term in Eq. (14) defines a constant energy shift by n1n2/4 = S1 · S2 = S2 [62]. Thus, by going just up to second order in the perturbation of the hopping terms we obtain the well-known Heisenberg term (1), if we define the exchange parameter J as J = −2t2/U . According to what has been said above, S = 1/2 mod- els with pair-interaction involving electrons hopping be- tween two sites can only exhibit a bilinear spin Hamilto- nian. This is confirmed by the inclusion of fourth order terms in the perturbation (10) (the third order vanishes again), which can be summarized to the following expres- sion (cid:18) (cid:88)(cid:48) ij (cid:19) can be expressed by the electron operators ci,α,σ, c as (12) † i,α,σ (13) (cid:17) (cid:17) † i,α,↑ci,α,↓ + c † i,α,↓ci,α,↑ c † † i,α,↑ci,α,↓ − c i,α,↓ci,α,↑ c (cid:16) (cid:16) (cid:88) (cid:88) (cid:88) α α α Si,x = 1 2 Si,y = − i 2 Si,z = 1 2 (ni,α,↑ − ni,α,↓) , 4thorder = − 8 t4 H2 sites U 3 S i · S j − ni nj 4 . (15) whereby the sum goes over all orbitals α at site i, we show now how the electron Hamiltonian of a particu- lar model system folded down in the proper spin sector No terms of additional spin-spin interactions show up in fourth order perturbation for two site-interactions. This shows that indeed a system of pair interactions of spin- 1/2 sites can be described purely by the Heisenberg in- teraction (1), although the fourth-order term provides a correction of the Heisenberg exchange parameter 6 TABLE I. Calculated prefactors of the Heisenberg exchange and the 4-spin interactions in terms of the model parameters t and U of the Hubbard model (5) taken at single-orbital per site for different numbers of sites with S = 1/2 obtained by going up to 4th order in the Lowdin partitioning. J = − 2 t2 U + 8t4 U 3 . (16) sites J K 0 0 2 U 3 U 3 − 2 t2 − 2 t2 3 4 − 2 t2 5 − 2 t2 6 − 2 t2 8 − 2 t2 U + 8 t4 U + 6 t4 U + 10 t4 U + 20 t4 U + 36 t4 U + 86 t4 U 3 U 3 − 10 t4 U 3 − 10 t4 U 3 − 10 t4 U 3 − 10 t4 U 3 U 3 U 3 The negative sign of the leading term means that the magnetic interaction of a spin-1/2 system is the m = 0 singlet state if t/U < 1/2, which we equate with the antiferromagnetic state. If the system becomes more metallic, the hopping matrix element t increases as well as the number of sites involved. Then the prefactor of the second term increases rapidly with system size (see Tab. I) and the likelihood for ferromagnetic interactions increases. Although discussed only for spin states with m = 0, the same effective Hamiltonian is also able to describe those with m = 1 and m = −1, respectively, as the excitation energy of those states due to the hopping of electrons is 0 both in the Hubbard Hamiltonian and in the effective spin Hamiltonian. b. 3 sites, spin S = 1/2 Since the fourth-order per- turbation term in (11) involves four successive hopping events of electrons, the interaction can involve spins or orbitals, respectively, beyond two sites up to four sites and thus can go beyond the pair interaction typical for the Heisenberg model. Considering three sites, the per- turbation theory results, however, again in a pair inter- action analogous to the Heisenberg model. The only dif- ference with respect to the system with two sites is a change of the prefactor, i.e. of the Heisenberg exchange parameter, respectively (cf. Tab. I, for simplicity we as- sumed the same t for all hopping events. The effect of different hopping elements tij will be analyzed be- low). Again, this spin Hamiltonian is capable of de- scribing all the subspaces for different m (here, m = −3/2,−1/2, +1/2, +3/2). c. 4 sites, spin S = 1/2 For four sites, the fourth or- der perturbation produces terms, which can be subsumed to the Heisenberg term, but generates also additional † † † † 1,↓c1,↑c2,↑c3,↓c4,↓ +h.c., for 3,↑c 4,↑c ones, for example, c 2,↓c four sites with m = 0. In contrast to the terms above, this term flips four spins instead of two. If we collect all these terms of fourth order and express them in terms of spin operators we obtain H4 sites 4thorder = 10 t4 U 3 (cid:88)(cid:48) ijkl (cid:16)S i · S j − ninj (cid:88)(cid:48) 4 (cid:17)(cid:16)S k · S l − nknl (cid:17) 4 (17) (18) which can be divided into a 4-spin term H4 sites 4 spins = 10 t4 U 3 ijkl (S i · S j) (S k · S l) , plus a Heisenberg term with the prefactor J = 10 t4/U 3, and a constant energy shift of size 15 t4/U 3. The prefac- tor in Eq. (18) will be called −K in this paper. Equation (18) is a simplified version of the more com- plex four-spin interaction [8, 10, 43] introduced in (3), namely for the case when the hopping parameters be- tween all the atoms are the same. In a real system this is rarely the case as the value of the hopping pa- rameter t depends on the distances between the two in- volved atoms, the types of orbitals, but also on the en- vironment, for details see also Section: III B. Carrying out a more explicit calculation of the fourth-order term with pair-dependent hopping parameter tij, the prefactor K ∝ −t4 in (17), (18) changes to ring paths of hopping with Kijkl ∝ −tijtjktkltli and with spin terms as in (3). d. N > 4 sites, spin S = 1/2 Going up to more sites (e.g., 5, 6, and 8) we showed no additional spin in- teraction terms emerge and the previously shown spin Hamiltonians (Heisenberg plus 4-spin) describe fully the energy landscape. The calculated prefactors for the case that the same hopping parameter t exists between all sites are shown in Tab. I. Additional interaction terms will emerge beyond fourth order perturbation calcula- tions, e.g., six-order terms for N ≥ 6, which is beyond the scope of this paper. 2. Spin S ≥ 1 , The extension to systems with larger spins per site, which is made possible by more than one half-filled or- bital per site, is in principle straightforward, but in practice significantly more complex. The Hilbert space becomes much larger and we need to switch from a single-band to a multi-band Hubbard model with quite some additional interaction parameters, which ultimately adds considerable complexity to the prefactors or the ex- change parameters of the spin models, respectively (see Appendix for details). To keep the prefactors simple and transparent, we discuss here results for the simpli- fied case, where hopping interactions between equal and different orbitals are identical and orbital independent, t(cid:48) = t, and the Coulomb repulsion and the exchange in- teraction of electrons at the same site but different or- bitals, U(cid:48) = 0 and J(cid:48) = 0, are neglected (see Tab. II), valid assuming that the Coulomb energy is larger if the electrons are not just at the same site but also in the same orbital, i.e. for U(cid:48) (cid:28) U and J(cid:48) (cid:28) J. However, these sim- plifications do not alter the functional nature of the spin models, just simplify prefactors. The full prefactors can be found in the appendix. † 1,1,↓c1,1,↑c2,2,↑. a. 2 sites, spin S = 1 Starting again with the simplest S = 1 model of two sites with two orbitals per site, we find in second order perturbation terms in which two spins are reversed. For example, for m = 0 † c 2,2,↓c is such a term. As it can be seen, there is always one orbital per site involved in those spin flips. Collecting now all the terms which arise in second order perturbation we again end up with the Heisenberg Hamiltonian , but with a prefactor of J = −2t2/(U +JH). In fourth order perturbation, however, more important differences to the system with only one orbital per site occur. In addition to the term shown above, there appear additional terms as † 2,2,↓c † † † 1,1,↓c1,1,↑c1,2,↓c2,1,↓c2,2,↑ , 1,2,↑c 2,1,↑c c (19) where all the electron spins are reversed and thus all or- bitals are involved in this interaction. For this reason, we can already see that each site is involved twice in this in- teraction and therefore has to occur twice in the effective spin Hamiltonian. And indeed, by using the spin opera- tors, the resulting effective interaction can be written as 4thorder ∝ (cid:88)(cid:48) H2 sites (cid:18) ij (cid:19)2 S i · S j − ni nj 4 , (20) which can be simplified into the biquadratic interaction (2), a Heisenberg term, and a constant energy shift. The prefactors for this system and the systems introduced in the following with the previously named assumptions on the parameters of the multi-band Hamiltonian (5) can be found in Tab. II. The prefactors for systems treated with an unrestricted parameter set are shown in the Appendix. The appearance of the biquadratic interaction for S = 1 dimers is consistent with the spin-algebra, which states that the highest independent powers of pair interactions is given by (S i · S j)2S. For S = 1/2 dimers, the bi- quadratic term can always be expressed as the sum of the Heisenberg term and a constant shift, and thus dis- appears. Similarly, in S = 1 systems higher powers of (S1 · S2)n, with n ≥ 3, can be expressed in a sum of the biquadratic and Heisenberg term as well as a constant shift, and disappear too. b. 3 sites, spin S = 1 Considering a system with three sites and two orbitals at each site, second order per- turbation theory reproduces again the Heisenberg model between different pairs of the three sites. Fourth order perturbation enables the reverse of spins in four different orbitals, which in a system with 3 sites can be facilitated 7 TABLE II. Calculated prefactors of the Heisenberg exchange (J), the biquadratic (B), the three-spin (Y ) and the four-spin interactions (K) for different numbers of sites with S > 1/2 obtained by going up to 4th order in the Lowdin partitioning. We set t(cid:48) = t and U(cid:48) and J(cid:48) were set to zero (see text). sites S J − 2t2 2 3 4 2 3 U +JH (U +JH)3 1 1 − 2t2 1 − 2t2 3/2 − 2t2 3/2 − 2t2 U +JH + 36t4 + 96t4 + 6t4 + 90t4 U +JH (U +JH)3 U +JH (U +JH)3 B −20t4 (U +JH)3 −20t4 (U +JH)3 −20t4 (U +JH)3 −20t4 (U +JH)3 −20t4 Y 0 −40t4 (U +JH)3 −40t4 K 0 0 −10t4 (U +JH)3 (U +JH)3 0 −40t4 0 0 U +JH (U +JH)3 (U +JH)3 (U +JH)3 in two different ways: Either the four orbitals are taken just at two different sites or they are distributed over all three sites, of which one is the site where the electron spin is reversed in both orbitals, while at each of the other two sites only one orbital is involved in the hopping. The for- mer one results again in a biquadratic interaction. The latter one includes terms like † † † 1,1,↓c1,1,↑c2,1,↓c2,2,↓c3,2,↑ 2,1,↑c 2,2,↑c n3,1,↑ n1,2,↓c † 3,2,↓c , (21) where we can clearly see that two orbitals (here, the first orbital at site 3 and the second at site 1) are not affected by this hopping term, while the other four change their spin direction. At the end this can be summarized in terms of an effective Hamiltonian S i · S j − ni nj H3 sites 4 4thorder ∝ (cid:88)(cid:48) S i · S k − ni nk 4 (cid:19)(cid:18) (cid:19) (cid:18) , ijk (22) which again can be structured into three different terms namely a Heisenberg term, a constant shift and (S i · S j) (S i · S k) , (23) ijk an Hamiltonian expression we identify as the three-spin interaction introduced in (4). The exchange constant of the 3-spin interaction is called Y henceforth. As we can see from the collection of prefactors in Tab. II and Ap- pendix, the 3-spin constant, Y , is in the same order of magnitude and even by a factor of 2 larger than the bi- quadratic constant, B. Therefore, we suppose that this 3-spin interaction can play an important role in systems in which other higher order interactions such as the bi- quadratic or 4-spin interaction are comparable in size to the Heisenberg one. Iron based thin-film systems are can- didates for such a behavior, because the local magnetic moments and spins, respectively, of Fe are large in these environments. We will demonstrate this below for the exemplary systems Fe/Rh(111) and Fe/Ru(0001). c. 4 sites, spin S = 1 The behavior within the sec- ond order perturbation is the same as before. However, 4 spins ∝ (cid:88)(cid:48) H3 sites within fourth order perturbation calculations and in com- parison to the derivation of the interaction across three sites additional interaction terms are expected since the four orbitals which are involved in the interactions of the fourth order perturbation can now either be divided-up over 2, 3 or 4 sites resulting in the biquadratic, 3-spin and 4-spin interaction, respectively. The prefactors can be found in Tab. II. So we have shown that the 4-spin interaction is not just a result that occurs in S = 1/2 systems, but also in those with S = 1. d. spin S > 1 To clarify whether the previously shown results apply only to S = 1-systems or can also be applied to systems with larger spins, we have also in- vestigated systems with S = 3/2 that represent systems having three orbitals per site exhibiting local magnetic moments of 3 µB. As we can see in Tab. II the consid- ered systems can all be explained by the interplay of the exchange, biquadratic, 3-spin and 4-spin interaction. Additional magnetic interaction terms making use of the nature of at least three orbitals per site would re- quire the concerted hopping of six electrons, which is be- yond the fourth order perturbation theory to which we restrict ourselves in this paper. Candidate interactions of six-order perturbation treatments are six-spin interac- tions involving 2 to 6 sites. One obvious candidate of a six order perturbation treatment is a possible bicubic interaction (S i · S j)3 (24) H6 ∝ (cid:88)(cid:48) ij In order to check whether this bicubic interaction occurs within higher orders, we have decided to study the sys- tem of 2 sites with 3 orbitals up to sixth order in the perturbation. Indeed additional terms occur within the sixth order, which can be explained by the bicubic inter- action with a prefactor of (U +2JH)5 . In general, however, it can be assumed that this bicubic interaction as well as the other possible six-order terms are small compared to the previously studied second- and fourth-order inter- actions because it occurs in an even higher order of the perturbation. 336t6 B. Spin-models at surfaces due to hopping of electrons beyond nearest neighbor Up to now, we presented the results assuming that the hopping properties of the electrons between all the atoms are the same. In a real system this is not the case as the value of the hopping parameter t depends mainly on the distance between the two involved atoms, but also on the type of orbitals, the symmetry, the geometry or the en- vironment. In the case of model Hamiltonians describing strongly localized electron systems, the nearest neighbor (NN) approximation is often sufficient (all hopping pa- rameters t = 0 between all atom pairs except NN-pairs) and the spin models derived above can be applied prac- tically directly. On the other hand, assuming, the same 8 Investigated geometries: (a) a square arrangement FIG. 2. of atoms as it occurs e.g. at the (001)-surface of a bcc or fcc crystal and (b) a hexagonal arrangement as it occurs at the (111)-surface of an fcc crystal or the (0001)-surface of an hexagonal lattice. Sketched are the positions of the atoms and two different neighbor distances. t1 (t2) represents the hopping between nearest (next-nearest) neighbors. hopping parameter t is used between all atom pairs, for example, the interaction between four atoms corresponds to the description of the interactions on a regular tetra- hedron. In general, there is a lot of interest in film, inter- face or surface geometries of periodic lattices with atom coordinations for which the NN or constant-pair approx- imation is unrealistic. We want to take this into account and evaluate above spin models for two common types of surfaces, the (001) or (111) oriented surface of fcc crys- tals using a model with nearest and next-nearest neigh- bor (NNN) hopping. We focus on a periodic S = 1/2 system of one atom type with electron interactions in- volving maximal four lattice sites as indicated in Fig. 2, where the (001) and (111) geometry are sketched. Ob- viously, the former represents a square arrangement of the surface atoms, the latter is a triangular or diamond arrangement of an hexagonal lattice. J∆ S i · S i+δ∆ Within both geometries there are two different dis- tances between atoms, the NN- and NNN-distance. Thus, the respective electron hopping is described by two distinct hopping constants, t1 and t2, summarized to t∆, with ∆ ∈ 1, 2. Going up to second-order perturbation we find as expected the Heisenberg exchange which reads in- dependent of surface geometry H1 = −(cid:88)(cid:48) change constant between NN- (NNN-) pairs. (cid:80) with the respective prefactors J1 (J2) being the ex- de- notes the summation over the NN- (∆ = 1) and NNN- pairs (∆ = 2). δ denotes the number of NN- (NNN-) pairs. While there are two NNN-pairs for the square lat- tice, there is only one pair on the hexagonal one, as one of the diagonals (see diagonal connecting atom 1 and 3 in Fig. 2) is also a NN-pair. J∆ = − 2 t2 U ∆ , δ∆ i,δ∆ with While the second-order expression hold independent on the surface geometry, this is different for fourth-order cor- rections to the Heisenberg exchange and for the higher- (a) (b) order interactions, where we have to differentiate between the square and the hexagonal lattice. 1. (100) surface 2 + 4t2 2 − 2t4 1t2 At fourth-order perturbation we expect additional terms to the Heisenberg model proportional to t4. For the square lattice, fourth-order perturbation results in a correction of J1 by 10t4 1/U 3 and a correction of J2 by (8t4 1)/U 3. Surprisingly, the correction to J2 does not only contain the naively expected cor- rection proportional to t4 2, but includes also correction terms involving NN-hopping proportional to t2 2 and t4 1. This has its origin in (17) where contributions to a ring- hopping involving sites i and j have contributions to the pair-exchange between the spins at sites i and j. 1 t2 For a S = 1/2 system and fourth-order perturbation in electron hopping we also obtain contributions to the 4- spin interaction, here expressed on a cluster of four sites: H4 = − K1[ (S 1 · S 2) · (S 3 · S 4) + (S 1 · S 4) · (S 2 · S 3) − (S 1 · S 3) · (S 2 · S 4) ] (S 1 · S 3) · (S 2 · S 4) − K2 The first term follows the functional form given in (17) and includes all permutations of exclusive NN- interaction. The related prefactor becomes K1 = −80 t4 1/U 3 More precisely, the variation of the hopping amplitudes between different sites result in preferred paths for a ring- hopping. Thus, we expect additional contributions from the NNN-terms to the 4-spin interaction. We have found that these modifications do not affect the previously dis- cussed NN-ring-hopping but add additional permutations of coupling strength K2 ∝ t2 2 of the involved sites to the Hamiltonian and can be written in terms of the sec- ond term in Eq. (25). Taking a geometrical picture, this corresponds to a bow-tie-shaped loop which contains two NNN-hopping events and thus hopping terms over the di- agonals of the square. We therefore call this term bow-tie 4-spin term in the following. The prefactor of this term was determined to be K2 = −160 t2 2/U 3. Thus, the ra- tio between the prefactors for the two mentioned 4-spin terms is 1 t2 1 t2 (cid:18) t2 (cid:19)2 t1 K2 K1 = 2 . (26) Depending on the ratio of t1 and t2, the diagonal term can be of the same order of magnitude as the conventional 4- spin term or it might even dominate and should therefore not be neglected in applications of the spin-model, e.g., Monte-Carlo or spin-dynamic simulations. H4 = − (cid:88) spin interactions can in general be be written as Transferring our findings to an infinite lattice, the 4- (cid:0)K1[(S i · S j)(S k · S l) + (S i · S l)(S j · S k)] (cid:104)ijkl(cid:105)(cid:3) 9 + (K2 − K1)(S i · S k) · (S j · S l)(cid:1) . (27) Here, the notation (cid:104)ijkl(cid:105)(cid:3) denotes sums over unique non- crossing quatruplets of sites of closed loops i → j → k → l → i. In the case of site-independent hopping amplitude, i.e. t1 = t2 and thus 2K1 = K2, Eq. (27) simplifies to (23) while is corresponds to Eq. (3) in case of pure NN- hopping, i.e. t2 = K2 = 0. 2. (111) surface The diamond geometry of the hexagonal (111) lattice offers a different ratio between NNN- and NN-bonds com- pared to the square lattice (see Fig. 2), which is at the end reflected in different contributions to the spin Hamil- tonian. Collecting all Heisenberg-like terms up to fourth-order perturbation results in the following prefactors: J1 = −2 J2 = −2 t2 1 U t2 2 U + 8 + 8 t4 1 U 3 + 4 t4 2 U 3 + 4 (25) t3 1t2 U 3 − 2 U 3 − 2 t3 1t2 t2 1t2 2 U 3 t4 1 U 3 (28) (29) The contributions to the 4-spin interactions are equiva- lent to those for the square lattice (Eq. (25)) with the exception of the prefactor of the diagonal 4-spin term which changes to K2 = −160t3 1t2/U 3. Therefore, the ra- tio (cid:18) t2 (cid:19) t1 K2 K1 = 2 (30) makes it even more likely that this term is comparable in size compared to the conventional 4-spin term. C. Importance of three-spin interaction in iron based magnetic thin-film systems Now we turn to the description of real magnetic atomic monolayer thick transition-metal films. Mag- netic beyond-Heisenberg behavior has been theoretically predicted and experimentally observed for several sys- tems [10, 24, 39 -- 43]. The materials specific theoretical modelling of magnetic interactions is generally carried out by means of density functional theory (DFT) and can be pursued along two different paths: (i) The param- eters, t, U , JH, etc., entering the Hubbard model (5) are determined directly from DFT and expressions derived above are executed to obtain the exchange parameters of the different magnetic interaction terms. Although it is a possible route, the determination of the Coulomb U and the Hund JH parameters have some uncertainties due the screening that should be properly included as a re- sult of those electrons not treated in the Hubbard model explicitly, uncertainties that are sometimes too large to determine the exchange parameters of the spin-model to the level that it is predictable. (ii) The second approach, which we will follow, is to take the classical limit of the spin models above, i.e. work with classical vector spin, S, instead of vector operators, S, and calculate the to- tal energies for a large spectrum of magnetic states in momentum or real space using DFT. The spin model pa- rameters are then obtained by comparing the total energy landscape calculated by DFT and the spin model. 1. Fe/Rh(111) Al-Zubi et al. [44] systematically investigated the mag- netism of Fe monolayers on hexagonal surfaces of dif- ferent 4d transition-metal substrates using DFT. They calculated total energies of a large spectrum of mag- netic structures. This included both spin-spiral states for wave vectors q along the high-symmetry lines of the two- dimensional Brillouin zone and so-called multi-q states of particular q-vectors that allow superpositions of spin spirals of symmetry-equivalent q-vectors. The most re- markable finding was the prediction of a previously un- known up-up-down-down (uudd ) state as ground state in Fe/Rh(111), recently confirmed by spin-polarized scan- ning tunneling microscopy (SP-STM) measurements [45]. The uudd state can be interpreted as interference of 2 spin spirals with wave vectors of opposing directions (2Q- state). In order to understand the origin of this unknown uudd state, they mapped the DFT results onto a spin Hamilto- nian, which included the Heisenberg interaction extended by the biquadratic and the four-spin interaction, the two latter within the nearest-neighbor approximation, and determined the exchange parameters. The choice of the spin Hamiltonian was taken ad hoc, but motivated by previous successes of similar systems [24, 42, 43]. How- ever, they made some puzzling observations. While the energy difference of two unrelated uudd states (see Fig.2 of Ref. 44) characterized by two different wave vectors q should be the same in comparison to the spin-spiral state (1Q-state) with the corresponding q-vector, i.e. , E2Q − E1Q = 4 (2K − B) , (31) not only the absolute value, but also the sign varied for both. Several attempts were made to resolve this discrep- ancy, but only the extension of the spin Hamiltonian by the three-spin interaction, which we systematically de- rived in this paper on grounds of the Hubbard model as an ignored interaction being on the same level as the pre- viously applied biquadratic and 4-spin terms, was able to resolve this issue. In fact, depending on the sign of the exchange parameter, the 3-spin interaction selects one of the two uudd states to become ground state and indeed it was shown that this explains the magnetic ground state of Fe/Rh(111) [45]. 10 2. Fe/Ru(0001) We show here that a monolayer of Fe deposited on Ru(0001) is a further materials system with beyond- Heisenberg behavior and a system which requires the contribution of the three-spin interaction in addition to the biquadratic and four-spin interaction for a proper de- scription of the magnetic properties by a spin model. In difference to Fe/Rh(111), DFT calculations of Al-Zubi et al. [44] revealed the 120◦ N´eel-state (from atom to atom the direction of the magnetic moment changes by 120◦) as the energetically most favorable of all investigated states, and thus the higher-order interactions do not directly de- termine the ground state, but the DFT calculations show that this system exhibits a similarly puzzling energy spec- trum as Fe/Rh(111) and a proper spin model is required for the description of spin-dynamics, spin excitation and the determination of thermodynamic properties. In the following we determine the exchange parame- ter B, Y , and K of the three beyond-Heisenberg inter- actions, biquadratic, three-spin, and four-spin, respec- tively, in the NN-approximation analysing the ab initio data of Al-Zubi et al. [44]. While the single-wavevector spin spiral (1Q-state) is an eigensolution of the clas- sical Heisenberg model for a periodic lattice, beyond- Heisenberg interactions couple modes of different 1Q- states to multi-Q states with different energy and show that this can result in a more accurate description of Fe/Rh(0001). Therefore, we focus in the following on those single-Q vectors in the two-dimensional hexagonal Brillouin zone of q-vectors defined in reciprocal space √ as q = (q1, q2) in units of the inplane reciprocal lat- tice vectors b1(2) = (2π/a)(1/ (−)1), where a is the hexagonal in-plane lattice constant, that can form multi- Q states out of symmetry-equivalent 1Q-states. This includes the high-symmetry point M = (1/2, 1/2) rep- resenting the row-wise antiferromagnetic state, that can form a 3Q-state and the two states (ΓM)/2 = ±(1/4, 1/4) and 3/4(ΓK) = (±1/4,∓1/4) on the high-symmetry lines of the Brillouin zone whose superposition of propa- gating and counterpropagating waves, e.g., (ΓM)/2 and −(ΓM)/2, form 2Q- or uudd -states, respectively. 3, + Inserting now the spin structure expressed as a spin- spiral wave, Si = S(cos(qRi), sin(qRi), 0), where Ri de- notes the position vector to site i, for wave vector q, or linear combination of those into the respective expres- sions for the Heisenberg, biquadratic, three- and four- spin interactions we obtain the following expressions E3Q − EM = E2Q, ΓM E2Q, 3ΓK 2 4 4 16 3 (2K + B − Y ) = 4.6 meV (32) = 4 (2K − B − Y ) = −30.3 meV (33) = 4 (2K − B + Y ) = 7.5 meV (34) − E ΓM − E 3ΓK 2 which we compared with the energy differences (in meV) obtained from DFT. As one can see, the previously iden- tical energy differences for the two uudd states are now separated by 8Y due to the 3-spin interaction. For the prefactors of the three interactions we obtain: B = 4.22 meV, Y = 4.73 meV, K = 0.68 meV (35) The value of the three-spin exchange parameter, Y , is in the same order of magnitude as the biquadratic interac- tion, but is also significantly large compared to the NN- Heisenberg exchange constant J1 (J1 = −6.4 meV) [63] and should therefore not be neglected. Based on our investigation we would argue that the previously puzzling results for Fe/Ru(0001) are the result of the interplay between the biquadratic and a strong 3- spin interaction, which favors one of the magnetic uudd textures over the other, an energy difference that could not be resolved before when the three-spin interaction had been neglected. A final comment on the evaluation of the three-spin interaction. Analogously to the discussion of (25) and (27) the expression (23) can be simplified to H3 = −2Y [(Sj · Si)(Si · Sk) + (Si · Sj)(Sj · Sk) (cid:88) (cid:104)ijk(cid:105)∆ +(Si · Sk)(Sk · Sj)] (36) summing over triangles of NN-sites. IV. SUMMARY AND CONCLUSIONS In this paper we derived consistently and systemati- cally the spin Hamiltonian due to interacting electrons up to fourth order perturbation theory in the Lowdin parti- tioning algorithm. Starting point was the rotationally invariant multi-orbital Hubbard model that described the interacting electrons on a lattice. We showed that Lowdin's downfolding technique is an efficient approach to map the effect of the interacting electrons onto an ef- fective spin model. As a result we obtain the spin Hamil- tonian H = (H1 + H4){for S ≥ 1/2} + (H2 + H3){for S ≥ 1} , (37) which consists of the Heisenberg Hamiltonian H1 (1), the biquadratic (four-spin-two-site) H2 (2), the three-spin (four-spin-three-site) H3 (4), and the four-spin Hamil- tonian (four-spin-four-site) H4 (3). The Heisenberg term emerges already in second order perturbation, but the fourth-order perturbation term adds to the exchange cou- pling parameter. Characteristic of the fourth order terms is the hopping of electrons between 4 orbitals that con- nect maximally four sites. This form remains correct also for higher spins S treated up the fourth order per- turbation theory. On the other hand S = 3/2 has also 6-order contributions and S = 2, would have 6- and 8- order contributions, which we have not calculated. Since the dimension of the matrices H0 and H1 in the Lowdin 11 as (cid:0) n (cid:1)2 n/2 algorithm grows binomially with the number of orbitals , the algorithm becomes quickly involved and at the same time the exchange coupling parameters are becoming increasingly smaller and the terms less impor- tant. The exchange coupling parameters of the different Hamiltonians Hi, with i = 1, . . . , 4, are summarized in detail in the Appendix. The spin-orbit interaction was neglected. Subject to the spin-orbit interaction, Sz does not commute any- more with the Hamiltonian, thus the Hamiltonian does not block-diagonalizes anymore for different m, and the Lowdin partitioning becomes more involved. We showed that our technique is capable of verify- ing the commonly applied Heisenberg model, as well as the four-spin and biquadratic interaction, but unraveled in addition the occurrence of the three-spin interaction. The importance of the three-spin interaction was ver- ified for the systems of one monolayer Fe on Rh(111) and Ru(0001), where ab initio calculations [44] predicted puzzling results on the magnetic states that now could be consistently explained. The unusual up-up-down- down ground state stabilized by three-spin interaction in Fe/Rh(111) could recently be confirmed experimen- tally [45]. ACKNOWLEDGMENTS We would like to thank Nicholas Ohs for fruitful discussions regarding the development of a versatile Lowdin partitioning algorithm. We acknowledge funding from the DARPA TEE program through grant MIPR# HR0011831554 from DOI. Appendix: Prefactors for the complete model In the main text we focused on presenting the prefac- tors, or exchange parameter, respectively, of the different spin-models for the simplified case of orbital independent hopping interactions (ti,α,j,α = t(cid:48) i,α,j,α(cid:48) = t) and for the limits J(cid:48) = 0 and U(cid:48) = 0. Here, we show the extension of the results for which the hopping interaction between the same (ti,α,j,α = t) and between different orbitals (t(cid:48) i,α,j,α(cid:48) = t(cid:48)) are distinct. Analogously the distinction between intra- and inter-orbital onsite Coulomb repul- sion U , and U(cid:48), respectively, and exchange interaction, J and J(cid:48), respectively, is taken into account. Otherwise, all interaction parameters are kept orbital independent for simplicity and remain site independent assuming a periodic lattice of one atom type. In the following, we will denote exchange parameters as Xs×o with X ∈ (J, B, K, Y ) and s and o denoting the number of sites and orbitals, respectively. The pref- actors are calculated up to fourth order in the Lowdin partitioning. 12 J2×2 = − t2 + t(cid:48)2 B2×2 = − 2(t2 + t(cid:48)2)2 (U + JH)3 + U + JH + 4(t2 + t(cid:48)2)2 (U + JH)3 − (t2 − t(cid:48)2)2 2(U + JH)2JH (t2 − t(cid:48)2)2 (U + JH − U(cid:48) + J(cid:48) + H)(U + JH)2 16t2t(cid:48)2 (U + JH − U(cid:48) − J(cid:48) H)(U + JH)2 4(t2 − t(cid:48)2)2 + (2U + U(cid:48))(U + JH)2 + (U + JH − U(cid:48) − J(cid:48) H)(U + JH)2 t4 − 14t2t(cid:48)2 + t(cid:48)2 J3×2 = − t2 + t(cid:48)2 U + JH + − (t2 − t(cid:48)2)2 · 12(t2 + t(cid:48)2)2 (cid:18) (U + JH)3 − − 27 4J 2 H(U + JH) + 3(t4 + 6t2t(cid:48)2 + t(cid:48)4) (2U + 2JH − U(cid:48) − J(cid:48) H)(U + JH)2 12 J 2 H(2U + JH) 3 + 3 4J 2 H(U + 3JH) − 2JH(U + JH)2 3 (cid:19) 3 (2U + 2JH − U(cid:48) + J(cid:48))(U + JH)2 + (2U + JH + U(cid:48))(U + JH)2 B3×2 = − (t2 + t(cid:48)2)2 ·(cid:16) (cid:18) − (t2 − t(cid:48)2)2 · + + − − (cid:17) (U + JH − U(cid:48) − J(cid:48) 3 2 1 (U + JH)3 + 2JH(U + JH)2 − (2U + U(cid:48))(U + JH)2 − 4 H)(U + JH)2 1 (U + JH − U(cid:48) + J(cid:48) H)(U + JH)2 4 (U + JH − U(cid:48) − J(cid:48) H)(U + JH)2 (cid:19) Y3×2 = − 16t2t(cid:48)2 (U + JH − U(cid:48) − J(cid:48) (cid:18) − (t2 − t(cid:48)2)2 · 2t4 + 12t2t(cid:48)2 + 2t(cid:48)4 (2U + 2JH − U(cid:48) − J(cid:48) H)(U + JH)2 − 6t4 + 20t2t(cid:48)2 + 6t(cid:48)4 (U + JH)3 H)(U + JH)2 + + 1 2J 2 H(U + 3JH) 2 1 JH(U + JH)2 + 9 − 2J 2 H(U + JH) 2 J 2 H(2U + JH) (2U + JH + U(cid:48))(U + JH)2 − (2U + 2JH − U(cid:48) + J(cid:48) H)(U + JH)2 J4×2 = − t2 + t(cid:48)2 U + JH − (t2 − t(cid:48)2)2 · (cid:18) − 27 + 3 J 2 H(2U + JH) (2U + 2JH − U(cid:48) + J(cid:48) 2J 2 6 H(U + JH) H)(U + JH)2 + 6t4 + 36t2t(cid:48)2 + 6t(cid:48)4 24 − (2U + 2JH − U(cid:48) − J(cid:48) H)(U + JH)2 + − 6 2J 2 H(3JH + U ) JH(U + JH)2 (2U + JH + U(cid:48))(U + JH)2 16t2t(cid:48)2 (U + JH − U(cid:48) − J(cid:48) 23t4 + 58t2t(cid:48)2 + 23t(cid:48)4 + (U + JH)3 B4×2 = − (t2 − t(cid:48)2)2 · (cid:18) Y4×2 = − (t2 − t(cid:48)2)2 · − 1 2JH(U + JH)2 − − − 2(t4 + 2t2t(cid:48)2 + t(cid:48)4) (U + JH)3 4 (2U + U(cid:48))(U + JH)2 − (U + JH − U(cid:48) − J(cid:48) t4 − 14t2t(cid:48)2 + t(cid:48)4 + H)(U + JH)2 1 (U + JH − U(cid:48) + J(cid:48) H)(U + JH)2 (2U + 2JH − U(cid:48) + J(cid:48) 2 H)(U + JH)2 − (2U + JH + U(cid:48))(U + JH)2 − 2 8 (2U + JH)J 2 H − − + + (cid:18) 8 3 (cid:19) (cid:19) H)(U + JH)2 (cid:19) + 9 − 2(U + JH)J 2 H 2J 2 2t4 + 12t2t(cid:48)2 + 2t(cid:48)4 + (2U + 2JH − U(cid:48) − J(cid:48) H)(U + JH)2 − 1 + H(U + 3JH) 16t2t(cid:48)2 (U + JH − U(cid:48) − J(cid:48) (U + JH)2JH H)(U + JH)2 − 6t4 + 20t2t(cid:48)2 + 6t(cid:48)4 (U + JH)3 (cid:19) 1 13 K4×2 = − 5(t4 + 6t2(t(cid:48)2) + t(cid:48)4) 4(U + JH)3 J2×3 = − 2 3 (t2 + 2t(cid:48)2) (U + 2JH) − (t2 + 2tt(cid:48) + 3t(cid:48)2)(t − t(cid:48))2 2(U + 2JH − U(cid:48) + J(cid:48) 8t(cid:48)2(t − t(cid:48))2 3(U + 5JH − U(cid:48) − J(cid:48) H)(U + 2JH)2 − (t2 + 2tt(cid:48) + 3t(cid:48)2)(t − t(cid:48))2 (U + JH + U(cid:48))(U + 2JH)2 + 6(U + 2JH − U(cid:48) − J(cid:48) H)(U + 2JH)2 − (t2 − t(cid:48)2)2 H)(U + 2JH)2 − 3t4 + 76t2t(cid:48)2 + 76tt(cid:48)3 + 25t(cid:48)4 3t4 − 52t2t(cid:48)2 − 52tt(cid:48)3 − 7t(cid:48)4 9(U + 2JH)2JH 4(t2 + 2t(cid:48)2)2 (U + 2JH)3 (t2 + 2tt(cid:48) + 3t(cid:48)2)(t − t(cid:48))2 3(U + 2JH − U(cid:48) + J(cid:48) 16t(cid:48)2(t − t(cid:48))2 9(U + 5JH − U(cid:48) − J(cid:48) 2(t − t(cid:48))2(t + t(cid:48))2 27(U + 2JH)2JH + H)(U + 2JH)2 + H)(U + 2JH)2 + 8(t2 + 2t(cid:48)2)2 9(U + 2JH)3 9(U + 2JH − U(cid:48) − J(cid:48) 2(t2 + 2tt(cid:48) + 3t(cid:48)2)(t − t(cid:48))2 3(U + JH + U(cid:48))(U + 2JH)2 H)(U + 2JH)2 − B2×3 = + + + J3×3 = − 2 3 (t2 + 2t(cid:48)2) (U + 2JH) − −3t4 + 20t2t(cid:48)2 + 20tt(cid:48)3 − t(cid:48)4 H)(U + 2JH)2 − −110t4 − 456t2t(cid:48)2 − 16tt(cid:48)3 − 444t(cid:48)4 9(U + 2JH)3 6(U + 2JH − U(cid:48) − J(cid:48) (cid:20) − 1 (U + 2JH)2 · 243(U + 2JH)J 2 H − 8t4 + 64t2t(cid:48)2 + 32tt(cid:48)3 + 40t(cid:48)4 9(2U + 4JH − U(cid:48) − J(cid:48) H) 64t4 + 384t2t(cid:48)2 + 128tt(cid:48)3 + 288t(cid:48)4 + (cid:21) 243(5JH + U )J 2 H 16t(cid:48)2(t(cid:48) + 2t)2 9(U + 2JH − U(cid:48) − J(cid:48) H) + 27(U + 2JH) − −10t4 − 80t2t(cid:48)2 − 40tt(cid:48)3 − 50t(cid:48)4 (cid:18) 3(2U + 4JH − U(cid:48) − J(cid:48) H)(U + 2JH)2 − (t − t(cid:48))2 · t2 + 2tt(cid:48) + 3t(cid:48)2 (U + 2JH)2 (cid:20) + · 23t2 + 46tt(cid:48) + 63t(cid:48)2 27(U + 2JH)2JH + (t − t(cid:48))2 (U + 2JH)2 · 9(U + 2JH − U(cid:48) − J(cid:48) 3(U + JH + U(cid:48)) 3t4 − 52t2t(cid:48)2 − 52tt(cid:48)3 − 7t(cid:48)4 (cid:18) 2t2 + 4tt(cid:48) + 6t(cid:48)2 (cid:20) H)(U + 2JH)2 + B3×3 = + + Y3×3 = − (t − t(cid:48))2 · (t2 + 2tt(cid:48) + 3t(cid:48)2) · − 20(U + 2JH) 27JH(5JH + U ) + 1 U + JH + U(cid:48) + 20 3(2U + 3JH + U(cid:48)) 1 + − 2(U + 2JH − U(cid:48) + J(cid:48) H) 3(U + 5JH − U(cid:48) − J(cid:48) 8t(cid:48)2 H)(U + 2JH)2 + 3(2U + 4JH − U(cid:48) + J(cid:48) H) 10 (cid:21) (cid:19) (cid:19) 16t(cid:48)2 + 9(U + 5JH − U(cid:48) − J(cid:48) H) t2 + 2tt(cid:48) + 3t(cid:48)2 3(U + 2JH − U(cid:48) + J(cid:48) H) + + 2(t + t(cid:48))2 27JH 8(t2 + 2t(cid:48)2)2 9(U + 2JH)3 1 (cid:18) − (U + 2JH)2 · + 16 16 9(2U + 3JH + U(cid:48)) − 16 8 + 9(2U + 4JH − U(cid:48) + J(cid:48) H) (cid:21) (cid:19) + 16 81JH 14 ∗ m.hoffmann@fz-juelich.de 1 P. W. Anderson, Exchange in Insulators: Superexchange, Direct Exchange, and Double Exchange, edited by G. Rado and H. Suhl, Vol. 1 (Magnetism Academic, 1963). 2 H. J. Zeiger and G. W. Pratt, Magnetic Interactions in Solids, Monographs on Physics (Oxford University Press, 1973). 3 C. de Graaf and R. Broer, Magnetic Interactions in Solids and Molecules, Theoretical Chemistry and Computational Modelling (Springer International Publishing, 2015). 4 A. Fert, N. Reyen, and V. Cros, Nat. Rev. Mat. 2, 17031 28 A. Aharony, J. Phys. A 10, 389 (1977). 29 J. J. Sudano and P. M. Levy, Phys. Rev. B 18, 5078 (1978). 30 N. Uryu and S. A. Friedberg, Phys. Rev. 140, A1803 (1965). 31 D. S. Rodbell, I. S. Jacobs, J. Owen, and E. A. Harris, Phys. Rev. Lett. 11, 10 (1963). 32 B. D. Gaulin and M. F. Collins, Phys. Rev. B 33, 6287 (1986). 33 S. Bhattacharjee, V. B. Shenoy, and T. Senthil, Phys. Rev. B 74, 092406 (2006). 34 Y. A. Fridman and D. V. Spirin, Low Temp. Phys. 26, 273 (2017). (2000). 5 R. Shindou and N. Nagaosa, Phys. Rev. Lett. 87, 116801 35 J. Lou, T. Xiang, and Z. Su, Phys. Rev. Lett. 85, 2380 (2001). 6 T. Schulz, R. Ritz, A. Bauer, M. Halder, M. Wagner, C. Franz, C. Pfleiderer, K. Everschor, G. M., and R. A., Nat. Phys. 8, 301 (2012). 7 M. dos Santos Dias, J. Bouaziz, M. Bouhassoune, S. Blugel, (2000). 36 H. A. Brown, Phys. Rev. B 11, 4725 (1975). 37 M. Takahashi, J. Phys. C Solid State Phys. 10, 1289 (1977). 38 A. H. MacDonald, S. M. Girvin, and D. Yoshioka, Phys. and L. Samir, Nat. Commun. 7, 13613 (2016). Rev. B 37, 9753 (1988). 8 B. Dup´e, M. Hoffmann, C. Paillard, and S. Heinze, Nat. 39 U. Kobler, R. Mueller, L. Smardz, D. Maier, K. Fischer, Commun. 5, 4030 (2014). 9 A. De´ak, L. Szunyogh, and B. Ujfalussy, Phys. Rev. B 84, 224413 (2011). 10 M. Hoffmann, J. Weischenberg, B. Dup´e, F. Freimuth, P. Ferriani, Y. Mokrousov, and S. Heinze, Phys. Rev. B 92, 020401(R) (2015). 11 B. Gyorffy, A. Pindor, J. Staunton, G. Stocks, and H. Win- ter, J. Phys. F 15, 1337 (1985). 12 S. V. Halilov, H. Eschrig, A. Y. Perlov, and P. M. Oppe- neer, Phys. Rev. B 58, 293 (1998). 13 M. Pajda, J. Kudrnovsk´y, I. Turek, V. Drchal, and P. Bruno, Phys. Rev. B 64, 174402 (2001). 14 M. Fahnle and D. Steiauf, Dissipative magnetization dy- namics close to the adiabatic regime (Wiley Online Library, 2007). 15 R. Singer, F. Dietermann, and M. Fahnle, Phys. Rev. Lett. 107, 017204 (2011). 16 R. Singer, F. Dietermann, and M. Fahnle, Phys. Rev. Lett. 107, 119901(E) (2011). 17 W. Heisenberg, Zeitschrift fur Physik 49, 619 (1928). 18 M. Bode, M. Heide, K. von Bergmann, P. Ferriani, S. Heinze, G. Bihlmayer, A. Kubetzka, O. Pietzsch, S. Blugel, and R. Wiesendanger, Nature 447, 190 (2007). 19 I. Turek, J. Kudrnovsk´y, V. Drchal, and P. Bruno, Philos. Mag. 86, 1713 (2006). 20 E. S¸a¸sıoglu, L. M. Sandratskii, P. Bruno, and I. Galanakis, Phys. Rev. B 72, 184415 (2005). 21 In this paper the spin-orbit interaction is neglected and thus the Dzyaloshinskii-Moriya interaction or the magnetic anisotropies are not part of the discussion. 22 T. Iwashita and N. Uryu, Phys. Rev. B 14, 3090 (1976). 23 P. Ferriani, K. von Bergmann, E. Y. Vedmedenko, S. Heinze, M. Bode, M. Heide, G. Bihlmayer, S. Blugel, and R. Wiesendanger, Phys. Rev. Lett. 101, 027201 (2008). 24 P. Kurz, G. Bihlmayer, K. Hirai, and S. Blugel, Phys. Rev. Lett. 86, 1106 (2001). 25 C. Kittel, Phys. Rev. 120, 335 (1960). 26 M. Lines, Solid State Commun. 11, 1615 (1972). 27 P. Bruno, J. Magn. Magn. Mater. 121, 248 (1993), pro- ceedings of the International Symposium on Magnetic Ul- trathin Films, Multilayers and Surfaces. B. Olefs, and W. Zinn, Z. Phys. B 100, 497 (1996). 40 U. Kobler, A. Hoser, J. Englich, A. Snezhko, M. Kawakami, M. Beyss, and K. Fischer, J. Phys. Soc. Jpn. 70, 3089 (2001). 41 O. N. Mryasov, A. J. Freeman, and A. I. Liechtenstein, J. Appl. Phys. 79, 4805 (1996). 42 Y. Yoshida, S. Schroder, P. Ferriani, D. Serrate, A. Kubet- zka, K. von Bergmann, S. Heinze, and R. Wiesendanger, Phys. Rev. Lett. 108, 087205 (2012). 43 S. Heinze, K. von Bergmann, M. Menzel, J. Brede, A. Ku- betzka, R. Wiesendanger, G. Bihlmayer, and S. Blugel, Nat. Phys. 7, 713 (2011). 44 A. Al-Zubi, G. Bihlmayer, and S. Blugel, Phys. Status So- lidi (b) 248, 2242 (2011). 45 A. Kronlein, M. Schmitt, M. Hoffmann, J. Kemmer, N. Seubert, M. Vogt, J. Kuspert, M. Bohme, B. Alonazi, J. Kugel, et al., Phys. Rev. Lett. 120, 207202 (2018). 46 N. Romming, H. Pralow, A. Kubetzka, M. Hoffmann, S. von Malottki, S. Meyer, B. Dup´e, R. Wiesendanger, K. von Bergmann, and S. Heinze, Phys. Rev. Lett. 120, 207201 (2018). 47 We would like to note that the nomenclature three-spin interactions has recently drawn criticism as the interaction indicates an interaction of three spins, which would lack time-inversion symmetry. In the context of this paper we use it synonymously to four-spin-three-site interaction. 48 T. Iwashita and N. Uryu, J. Phys. Soc. Jpn. 36, 48 (1974). 49 T. Iwashita and N. Uryu, J. Phys. Soc. Jpn. 47, 786 (1979). 50 T. Iwashita and N. Uryu, Phys. Lett. A 76, 64 (1980). 51 H. Kobayashi, I. Tsujikawa, and I. Kimura, J. Phys. Soc. Jpn. 24, 1169 (1968). 52 R. Bastardis, N. Guih´ery, and C. de Graaf, Phys. Rev. B 76, 132412 (2007). 53 E. Muller-Hartmann, U. Kobler, and L. Smardz, J. Magn. Magn. Mater. 173, 133 (1997). 54 P.-O. Lowdin, J. Chem. Phys. 19, 1396 (1951). 55 R. Winkler, Spin-orbit coupling effects in two-dimensional electron and hole systems, Vol. 191 (Springer, 2003). 56 J. R. Schrieffer and P. A. Wolff, Phys. Rev. 149, 491 (1966). 57 S. Bravyi, D. P. DiVincenzo, and D. Loss, Ann. Phys. (N. Y.) 326, 2793 (2011). 58 J. Hubbard, Proc. Royal Soc. A 276, 238 (1963). 59 J. Hubbard, Proc. Royal Soc. A 277, 237 (1964). 60 J. Hubbard, Proc. Royal Soc. A 281, 401 (1964). 61 We use t(cid:48) i,α,j,α(cid:48) instead of only ti,α,j,α(cid:48) to emphasize the two different types of hoppings. This notation will also be used later on in the paper. 62 This ensures that a ferromagnetic alignment (either all spins up or all spins down) gains no energy in higher-order. This agrees well with the underlying Hubbard model where none of the hopping terms have any effect on those ferro- magnetic states as no hopping is possible. 63 B. Hardrat, A. Al-Zubi, P. Ferriani, S. Blugel, G. Bihlmayer, and S. Heinze, Phys. Rev. B 79, 094411 (2009). 15 

1306.1001

2

1306

Evidence for a Magnetic Seebeck effect

The irreversible thermodynamics of a continuous medium with magnetic dipoles predicts that a temperature gradient in the presence of magnetisation waves induces a magnetic induction field, which is the magnetic analog of the Seebeck effect. This thermal gradient modulates the precession and relaxation. The Magnetic Seebeck effect implies that magnetisation waves propagating in the direction of the temperature gradient and the external magnetic induction field are less attenuated, while magnetisation waves propagating in the opposite direction are more attenuated.

cond-mat.mes-hall

cond-mat

Evidence for a Magnetic Seebeck effect Sylvain D. Brechet,1, ∗ Francesco A. Vetro,1 Elisa Papa,1 Stewart E. Barnes,2 and Jean-Philippe Ansermet1 1Institute of Condensed Matter Physics, Station 3, Ecole Polytechnique F´ed´erale de Lausanne - EPFL, CH-1015 Lausanne, Switzerland 2James L. Knight Physics Building, 1320 Campo Sano Ave., University of Miami, Coral Gables, FL 33124, USA The irreversible thermodynamics of a continuous medium with magnetic dipoles predicts that a temperature gradient in the presence of magnetisation waves induces a magnetic induction field, which is the magnetic analog of the Seebeck effect. This thermal gradient modulates the precession and relaxation. The Magnetic Seebeck effect implies that magnetisation waves propagating in the direction of the temperature gradient and the external magnetic induction field are less attenuated, while magnetisation waves propagating in the opposite direction are more attenuated. 3 1 0 2 g u A 4 ] l l a h - s e m . t a m - d n o c [ 2 v 1 0 0 1 . 6 0 3 1 : v i X r a PACS numbers: 75.76.+j, 76.50.+g The discovery of the spin Seebeck effects in ferromag- netic metals [1], in semiconductors [2], and in insula- tors [3], has generated much research for spin transport in ferromagnetic samples of macroscopic dimensions sub- jected to temperature gradients. The interplay of spin, charge and heat transport defines the rich field known as spin caloritronics [4]. Prompted by these recent devel- opments, we established a formalism describing the irre- versible thermodynamics of a continuous medium with magnetisation [5]. In this letter, we test a particular experimental predic- tion of this formalism on a YIG slab. We argue that the thermodynamics of irreversible processes implies the ex- istence of a magnetic counter-part to the well-known See- beck effect. We show how a thermally induced magnetic field modifies the Landau-Lifshitz equation and provide experimental evidence for the Magnetic Seebeck effect by the propagation of magnetisation waves in thin crys- tals of YIG. The effect of a temperature gradient on the dynamics of the magnetisation on a YIG slab with and without Pt stripes was investigated recently by Obry et al. [6], Cunha et al. [7], Silva et al. [8], Padron-Hernandez et al. [9, 10], Jungfleisch et al. [11] and Lu et al. [12]. In general, irreversible thermodynamics predicts cou- plings between current and force densities. In equation (86) of reference [5], we identified the magnetisation force term m∇ B. For an insulator like YIG, there is no charge current. As explained in detail in reference [5], the transport equation (94) of [5] implies that the mag- netisation force density M∇ B ind induced by a thermal force density − n kB ∇ T is proportional and opposite to this force density, i.e. M∇ B ind = λ n kB ∇ T , (1) which corresponds to equation (155) of reference [5], where λ > 0 is a phenomenological dimensionless param- eter, kB is Boltzmann's constant and n = 1.1·1028 m−3 is the Bohr magneton number density of YIG. The thermo- dynamic formalism does not allow for a direct estimation of λ. The numerical value of this parameter needs to be evaluated directly from the experimental data, as shown below. In the bulk of the sample, as shown in reference [5], the magnetisation force density has the structure of a Lorentz force density [13] expressed in terms of the mag- netic bound current density jM = ∇ × M [14], M∇ B ind = jM × B ind . Thus, using vectorial identities, the phenomenological re- lations (1) and (2) imply that in the bulk of the system the magnetic induction field B ind, induced by a uniform temperature gradient ∇ T in the presence of a magnetic bound current density ∇ × M, is given by, i.e. B ind = εM × ∇ T , (2) (3) (4) where the phenomenological vector εM is given by, εM = − λ n kB (∇ × M) −1 . By analogy with the Seebeck effect, we shall refer to this phenomenon as the Magnetic Seebeck effect. The time evolution of the magnetisation M is given by the Landau-Lifschitz-Gilbert equation, i.e. M = γ M × B eff − α MS M × M , (5) where γ is the gyromagnetic ratio, α (cid:39) 10−4 is the Gilbert damping parameter of YIG [15], MS = 1.4 · 105 A m−1 is the magnitude of the effective saturation magnetisation of YIG at room temperature [16]. The ef- fective magnetic induction field B eff includes the external field B ext, the demagnetising field B dem, the anisotropy field B ani, which behaves as an effective saturation mag- netisation in the linear response [17] and finally a ther- mally induced field B ind given by the relation (3). The exchange field B int [18] is negligible in the following, as we consider magnetostatic modes [19]. The demagnetis- ing field B dem breaks the spatial symmetry and gener- ates an elliptic precession cone. After performing the linear response of the magnetisation in the presence of a thermally induced field B ind, we shall describe how the demagnetising field B dem affects the magnetic suscepti- bility. We found evidence for the Magnetic Seebeck effect by exciting locally, at angular frequency ω (cid:39) 2.74 · 1010 s−1, the ferromagnetic resonance of a YIG slab of length Lz = 10−2 m, width Ly = 2 · 10−3 m and thickness Lx = 2.5 · 10−5 m, subjected to a temperature gradient as small as ∇ T (cid:39) 2 · 103 K m−1 generated by Peltier elements. The excitation field is applied on the slab using a local antenna as detailed in reference [20]. For signal transmission experiments, two antennae are used, set ap- proximatively 8 mm apart, as shown on Fig. 1. Note that a similar setup for a gradient orthogonal to the YIG slab was investigated recently [7]. For reasons explained be- low, these two setups can be expected to probe different mechanisms. FIG. 1. Time-resolved transmission measurement of magneti- sation waves The external magnetic induction field B ext applied on the YIG film consists of a uniform and constant field B0 and a small excitation field b = bx x + by y locally oscillating in a plane orthogonal to B0 = B0 z. In the limit of a small excitation field, i.e. in the linear limit, the magnetisation field M consists of a uniform and constant field MS = MS z and a response field m = mx x + my y locally oscillating in a plane orthogonal to MS such that m (cid:28) MS . The linear response of the magnetisation to the excitation field, according to the time evolution equation (5) is given by, m = γ (m × B0 + MS × B1) − α MS MS × m , (6) where the first-order magnetic induction field B1 yields, B1 = b − µ0 (cid:0)kT · ∇ −1(cid:1) m , 2 −1 × m(cid:1) × ∇ T (cid:0) −1(cid:1) m − 1 1 M 2 S To obtain the expressions (7) and (8), we used the linear vectorial identity, ∇ −1 × ∇ T = ∇ T · ∇ (∇ × M) (cid:0) −1·∇ = 1 and the last term on the RHS vanishes where ∇ since it averages out on a precession cycle. (∇ T ) ∇ −1 m , 1 M 2 S M 2 S = The vectorial time evolution equation (6) is written explicitly in Cartesian coordinates as, mx =(cid:0)ω0 + ωM kT · ∇ my = −(cid:0)ω0 + ωM kT · ∇ −1(cid:1) my + α my − ωM µ−1 −1(cid:1) mx − α mx + ωM µ−1 0 by , 0 bx , (9) where the angular frequencies ω0 and ωM are defined respectively as, ω0 = γ B0 , ωM = γ µ0 MS . (10) In a stationary regime, The magnetic excitation field b and the magnetisation response m are oscillating at an angular frequency ω, which is expressed in Fourier series as, bk ei(k·x− ωt+ π 2 ) , by = mk ei(k·x− ωt+ π 2 ) , my = k k bk ei(k·x− ωt) , mk ei(k·x− ωt) , (11) (cid:88) (cid:88) k bx = mx = (cid:88) (cid:88) k where the eigenstates bk and mk are complex-valued and dephased. The Cartesian components of the eigenmodes kx,y,z satisfy the boundary conditions of null m at the surface of the sample, kx,y,z = nx,y,z π Lx,y,z , (12) where nx,y,z ∈ N [20]. The eigenstates of the excitation field bk and the re- sponse field mk are related through the magnetic suscep- tibility χk, i.e. mk = µ−1 0 χk bk . (13) The time evolution equations (9), the definition (10), and the Fourier series (11) in the stationary regime imply that the magnetic susceptibility χk is given by, (7) χk = − 1 Ω − Ω0 + i (α Ω + kT · k−1) , (14) µ0 is the magnetic permeability of vacuum and the ther- mal wave vector, where the dimensionless parameter Ω and Ω0 are respec- tively defined as, kT = λ n kB µ0 M 2 S ∇ T . (8) Ω = ω ωM , Ω0 = ω0 ωM . (15) CuAntennaePeltierelementOscilloscopeRF pulse generatorAmplifierCrystal detectorB0B1YIG3 The demagnetising field B dem = − µ0 mx x causes the damping and the magnetic susceptibility χkx along the x-axis to differ respectively from the damping and the magnetic susceptibility χky along the y-axis. The reso- nance frequency(cid:112)ω0 (ω0 + ωM ) is given by Kittel's for- Ω − (cid:112)Ω0 (Ω0 + 1) + i rx,y (α Ω + kT · k−1) mula [21] to first-order in α and kT . Thus, the magnetic susceptibilities χkx,y yield, χkx,y = − 1 , (16) where rx,y > 0 are phenomenological damping scale fac- tors accounting for symmetry breaking. As shown by Cunha et al. on Fig.1(a) of reference [7], the propagating modes of the magnetisation waves in the bulk of YIG are magnetostatic backward volume modes (MSBVM) propagating in the direction − k−1. The ex- pressions (16) and (8) for the magnetic susceptibilities and the thermal wave vector kT , imply that the mag- netisation waves propagating from the cold to the hot side, i.e. kT · k−1 < 0, are less attenuated by the temper- ature gradient and the magnetisation waves propagating from the hot to the cold side, i.e. kT ·k−1 > 0 are further attenuated. Thus, the opening angle of the precession cone of the magnetisation m for a magnetisation wave propagating in the direction of the temperature gradient decreases less than the opening angle for a magnetisation wave propa- gating in the opposite direction, as shown on Figure 2. This is confirmed experimentally by detecting induc- tively at one end of the sample the signal that results from an excitation pulse of 15 ns duration at the other end. The signals obtained by sweeping the magnetic in- duction field B0 for the propagation of magnetisation waves from the cold end to the hot end or from the hot end to the cold end are given on Fig. 3. Clearly, the waves propagating from the cold to the hot side appear to decay less rapidly than the waves propagating from the hot to the cold side. The time evolution of the signals for the waves prop- agating in the direction of the gradient or opposite to it are obtained by averaging the signals over the range of the magnetic induction field B0 and displayed on Fig. 4. The signal is a convolution of kz modes that have dif- ferent group velocities and decay exponentially due to the damping. The peaks were identified in reference [22] as the result of the propagation of odd modes. Since the peaks of the transmitted signals are detected at the same time, the temperature gradient does not affect sig- nificantly the kz mode group velocities. Moreover, from the logarithmic scale for the signal on Fig. 4, a larger difference in attenuation between the signals for small kz modes is inferred. This is in line with the theoretical prediction, made by equation (16), for the Magnetic See- beck effect to be proportional to k−1 z . Moreover, since the FIG. 2. Propagation of magnetisation waves from the cold to the hot side (top) and vis versa (bottom). The cones describe the precession of the magnetisation at excitation m(0) and at detection m(τ ). The amount of damping depends on the rel- ative orientation kT of the temperature gradient with respect to the magnetisation wave propagation direction − k−1. FIG. 3. Transmitted signals from the cold to the hot side and from the hot to the cold side as a function of the magnetic field B0 and of the detection time after a 15 ns pulsed excitation at 4.36 GHz. The lighter areas correspond to a larger signal. T∇m(0)B0YIGHottoColdm(τ)kT∝Tzyxm(0)B0YIGColdtoHotkT∝m(τ)k-1k-1k-1k-1∇86848280787674B0(mT)86848280787674Hot to ColdHot to ColdB0(mT)-20020406080100120140160Time(ns)Cold to HotCold to HotFIG. 4. Transmitted signal as a function of time after a 15 ns pulsed excitation at 4.36 GHz. relative difference between the signals is due to the tem- perature gradient, we can estimate the relative difference between the damping terms α Ω and kT · k−1 appearing in the expression (16) for the magnetic susceptibilities. Comparing the signals at t = 40 ns, we find that the di- mensionless parameter λ (cid:39) 6·10−7, which corresponds to a thermal damping ratio kT ·k−1/α Ω (cid:39) 0.3 less that an order of magnitude below the self-oscillation threshold. The difference in attenuation between the signals is also shown on the FMR spectrum detected 70 ns after the pulse and displayed on Fig. (5). The spectral linewidth ∼ 0.2 mT corresponds to inhomogeneous broadening, since it is much larger than the homogeneous linewidth ∼ α B eff [23]. 4 to the temperature dependence of the saturation mag- netisation. Moreover, in contrast to the claim made in reference [6], Fig. 4 shows that magnetisation waves can propagate with and against the temperature gradient and that the effect of the temperature is proportional to k−1 z . For a temperature gradient orthogonal to the YIG plane, Cunha et al. [7] showed that the temperature gra- dient affects the propagation of magnetisation waves only when Pt is deposited on the YIG slab. The effect is ac- counted for by a model of spin injection and spin pump- ing, detailed by Ando et al. [24], at the interface between Pt and YIG. The quantitative analysis of the data is pre- sented in reference [8]. In reference [7], it is stated clearly that the effect does not occur in the absence of Pt on the surface. When Pt is removed in such a setup where kT · k−1 = 0, the mechanism invoked by Cunha et al. is not operative and our mechanism is not effective either. In summary, we point out that thermodynamics of irreversible processes implies a coupling between heat current and magnetisation precession in a temperature gradient. This effect can be expressed by an induced magnetic field B ind proportional to the applied temper- ature gradient. Thus, we suggest to refer to it as a Magnetic Seebeck effect, since it is the magnetic ana- log of the regular Seebeck effect. It is distinct from the magneto-Seebeck effect, which refers a change in the See- beck coefficient due to the magnetic response of nanos- tructures [25]. We analyse how the Landau-Lifshitz equa- tion is modified, and find a contribution to the dissipa- tion that is linear in ∇ T . Hence, this effect can in- crease or decrease the damping, depending on the ori- entation of the wave vector of the excited magnetostatic mode with respect to the temperature gradient. If the temperature gradient could be made strong enough, i.e. kT · k−1 > α Ω, then the damping would be negative and the magnetisation would undergo self-oscillation. This would be analogous to the magnetisation self-oscillation described in chapter 7 of reference [26] and the heat- equivalent of Berger's SWASER predicted for charge- driven spin polarised currents [27]. for We thank Fran¸cois A. Reuse, Klaus Maschke and Joseph Heremans insightful comments and ac- knowledge the following funding agencies : Polish- Swiss Research Program NANOSPIN PSRP-045/2010; Deutsche Forschungsgemeinschaft SS1538 SPINCAT, no. AN762/1. FIG. 5. FMR signal of a 15 ns pulsed excitation at 4.36 GHz detected after 70 ns, after baseline correction. As rightly pointed out in reference [6], the tempera- ture dependence of the saturation magnetisation affects the amplitude of the magnetisation waves. However, since our experimental setup is sufficiently close to the self-oscillation threshold for a temperature gradient that is small enough, we expect the dynamic contribution kT · k−1 to be larger than the static contribution due ∗ sylvain.brechet@epfl.ch [1] K. Uchida, S. Takahashi, K. Harii, J. Ieda, W. Koshibae, K. Ando, S. Maekawa, and E. Saitoh, Nature 455, 778 (2008). [2] C. M. Jaworski, J. Yang, S. Mack, D. D. Awschalom, J. P. Heremans, and R. C. Myers, Nat Mater 9, 898 (2010). kT·k-1<0kT·k-1>0Time(ns)0.46.00.60.81.02.04.0-20020406080100120140160kT·k-1<0kT·k-1>00.51.01.52.074767880828486B0 (mT)5 [3] K. Uchida, J. Xiao, H. Adachi, J. Ohe, S. Takahashi, J. Ieda, T. Ota, Y. Kajiwara, H. Umezawa, H. Kawai, G. E. W. Bauer, S. Maekawa, and E. Saitoh, Nat Mater 9, 894 (2010). [4] G. E. W. Bauer, E. Saitoh, and B. J. van Wees, Nature Materials 11, 391 (2012). [5] S. D. Brechet and J.-P. Ansermet, Eur. Phys. J. B 86, 318 (2013). [6] B. Obry, V. I. Vasyuchka, A. V. Chumak, A. A. Serga, and B. Hillebrands, Applied Physics Letters 101, 192406 (2012). [16] F. Boukchiche, T. Zhou, M. L. Berre, D. Vincent, B. Payet-Gervy, and F. Calmon, PIERS 2010 Cambridge 1, 700 (2010). [17] J. A. Duncan, B. E. Storey, A. O. Tooke, and A. P. Cracknell, Journal of Physics C Solid State Physics 13, 2079 (1980). [18] C. Kittel, Reviews of Modern Physics 21, 541 (1949). [19] A. A. Serga, A. V. Chumak, and B. Hillebrands, Journal of Physics D Applied Physics 43, 264002 (2010). [20] E. Papa, S. E. Barnes, and J.-P. Ansermet, IEEE Trans- actions on Magnetics 49, 1055 (2013). [7] R. O. Cunha, E. Padr´on-Hern´andez, A. Azevedo, and [21] C. Kittel, Introduction to Solid State Physics, 8th ed. S. M. Rezende, Phys. Rev. B 87, 184401 (2013). (Wiley, New York, 2004). [8] G. L. da Silva, L. H. Vilela-Leano, S. M. Rezende, and A. Azevedo, Applied Physics Letters 102, 012401 (2013). [9] E. Padr´on-Hern´andez, A. Azevedo, and S. M. Rezende, [22] E. Padr´on-Hern´andez, A. Azevedo, and S. M. Rezende, Applied Physics Letters 99, 192511 (2011). [23] S. V. Vonsovskii, Ferromagnetic Resonance (Pergamon: Journal of Applied Physics 111, 070000 (2012). Oxford, 1966). [10] E. Padr´on-Hern´andez, A. Azevedo, and S. M. Rezende, Physical Review Letters 107, 197203 (2011). [11] M. B. Jungfleisch, T. An, K. Ando, Y. Kajiwara, K. Uchida, V. I. Vasyuchka, A. V. Chumak, A. A. Serga, E. Saitoh, and B. Hillebrands, Applied Physics Letters 102, 062417 (2013). [12] L. Lu, Y. Sun, M. Jantz, and M. Wu, Physical Review Letters 108, 257202 (2012). [13] F. A. Reuse, Electrodynamique (PPUR: Lausanne, 2012). [14] D. J. Griffiths, Introduction to Electrodynamics, 3rd ed. (Prentice-Hall, Upper Saddle River, 1999). [15] H. Kurebayashi, O. Dzyapko, V. E. Demidov, D. Fang, A. J. Ferguson, and S. O. Demokritov, Nat Mater 10, 660 (2011). [24] K. Ando, S. Takahashi, K. Harii, K. Sasage, J. Ieda, and E. Saitoh, Physical Review Letters S. Maekawa, 101, 036601 (2008). [25] M. Walter, J. Walowski, V. Zbarsky, M. Munzenberg, M. Schafers, D. Ebke, G. Reiss, A. Thomas, P. Peretzki, M. Seibt, J. S. Moodera, M. Czerner, M. Bachmann, and C. Heiliger, Nat Mater 10, 742 (2011). [26] S. E. Barnes, Spin Current, edited by S. Maekawa and S. O. Valenzuela and E. Saitoh and T. Kimura (Oxford University Press, 2012). [27] L. Berger, IEEE Transactions on Magnetics 34, 3837 (1998). 

1704.04941

1

1704

Topological Semimetals carrying Arbitrary Hopf Numbers: Hopf-Link, Solomon's-Knot, Trefoil-Knot and Other Semimetals

We propose a new type of Hopf semimetals indexed by a pair of numbers $(p,q)$, where the Hopf number is given by $pq$. The Fermi surface is given by the preimage of the Hopf map, which is nontrivially linked for a nonzero Hopf number. The Fermi surface forms a torus link, whose examples are the Hopf link indexed by $(1,1)$, the Solomon's knot $(2,1)$, the double Hopf-link $(2,2)$ and the double trefoil-knot $(3,2)$. We may choose $p$ or $q$ as a half integer, where torus-knot Fermi surfaces such as the trefoil knot $(3/2,1)$ are realized. It is even possible to make the Hopf number an arbitrary rational number, where a semimetal whose Fermi surface forms open strings is generated.

cond-mat.mes-hall

cond-mat

Topological Semimetals carrying Arbitrary Hopf Numbers: Hopf-Link, Solomon's-Knot, Trefoil-Knot and Other Semimetals Department of Applied Physics, University of Tokyo, Hongo 7-3-1, 113-8656, Japan Motohiko Ezawa We propose a new type of Hopf semimetals indexed by a pair of numbers (p, q), where the Hopf number is given by pq. The Fermi surface is given by the preimage of the Hopf map, which consists of loops nontrivially linked for a nonzero Hopf number. The Fermi surface forms a torus link, whose examples are the Hopf link indexed by (1, 1), the Solomon's knot (2, 1), the double Hopf-link (2, 2) and the double trefoil-knot (3, 2). We may choose p or q to be a half integer, where the Fermi surface is a torus knot such as the trefoil knot (3/2, 1). It is even possible to make the Hopf number an arbitrary rational number, where a semimetal whose Fermi surface forms open strings is generated. It (3D) space1,2. Introduction: Weyl semimetals are described by the two- band model equipped with a point node in the three- is characterized by the dimensional monopole charge in the momentum space3. Line-nodal semimetals or loop-nodal semimetals are also possible in the 3D space, where the zero-energy Fermi surface is given by a line or a closed loop5–15. Recently, a nodal-chain semimetal is proposed, where loop nodes touch with each other4. A natu- ral question is whether nontrivial Fermi-surfaces made of loop nodes such as links and knots are possible. The two-band Hamiltonian H = S (k) · σ is a prototype of Hopf insulators16–21, where σ = (σx, σy, σz) are the Pauli matrices and S (k) is the normalized pseudospin S (k) = 1 spanning the sphere surface S2. On the other hand, the 3D Brillouin zone is identical to the torus T3. A homotopy from T3 to S2 is characterized by the Hopf number. It has been argued that Hopf insulators with arbitrary Hopf numbers are possible17. Nontrivial Hopf textures are also discussed in cold atoms22,23, light fields24 and liquid crystal25. In this Letter we investigate topological semimetals, where Fermi surfaces consist of nontrivial loops with arbitrary Hopf numbers. We explore the Hamiltonian H = Sxσx + Szσz, where the zero-energy condition reads Sy = ±1. The Fermi surface is the preimage of the points S± = (0,±1, 0) in the mapping T3 → S2, and it consists of two loops. They are linked for a nonzero Hopf number. We construct Fermi sur- faces comprised of the Hopf link, the Solomon's knot and oth- ers: See Fig.1. They are "torus links" lying on the surface of a torus. Furthermore, we construct Fermi surfaces comprised of torus knots by choosing half integer Hopf numbers, among which there arises in particular a trefoil-node Fermi surface. We can even choose any one rational number as a Hopf num- ber, where the Fermi surface forms open strings though it de- scribes no longer a topological semimetal. Torus-link semimetals: The Hamiltonian of topological Hopf insulators is given by16–21 H (k) = S (k) · σ, (1) where S (k) is the normalized pseudospin field, S (k) = 1. It is written as S = z†σz in terms of the CP1 field z. We define it by (cid:19) (cid:18) z↑ z↓ z ≡ (cid:112)η↑2p + η↓2q 1 = (cid:19) (cid:18) ηp↑ ηq↓ , (2) FIG. 1: Bird's eye's view of the almost zero-energy surface of the Hamiltonian Hxz. (a) The Hopf-link semimetal with (p, q) = (1, 1), (b) the Solomon's-knot semimetal with (2, 1), (c) the double Hopf- link semimetal with (2, 2), which consists of two Hopf links. (d) the double trefoil-knot semimetal with (3, 2). We have chosen m = 2 in Eq.(3) to draw figures. The preimage of Sy = 1 is colored in magenta, while that of Sy = −1 is colored in cyan. where η↑ and η↓ are complex numbers given by16–20 η↑ (k) = sin kx + i sin ky, η↓ (k) = sin kz + i(cos kx + cos ky + cos kz − m), (3) while p and q are integers. We note that p and q are originally introduced as a set of coprime integer17 but here we do not impose it. We will see that p and q can be generalized even to rational numbers though a cut is introduced. The normalized pseudospin is explicitly represented as Sx + iSy = 2ηp↑ ¯ηq↓ η↑2p + η↓2q , Sz = η↑2p − η↓2q η↑2p + η↓2q . (4) We consider a class of pseudospin textures indexed by a pair of numbers (p, q) in the Hamiltonian (1). The CP1 field takes valued on the 3D sphere S3 since it contains four real numbers, N1 = Re z↑ (k), N2 = Im z↑ (k), N3 = Re z↓ (k) and N4 = Im z↓ (k), together with the It gives a mapping T3 → S3, for the Brillouin zone is a 3D torus. On the other normalization condition (cid:80) i = 1. i N 2 7 1 0 2 r p A 7 1 ] l l a h - s e m . t a m - d n o c [ 1 v 1 4 9 4 0 . 4 0 7 1 : v i X r a hand, the normalized pseudospin expressed as S = z†σz de- fines a mapping S3 → S2, for it takes values on the sphere S2. Consequently the underlying structure of the Hamiltonian (1) is a mapping T3 → S3 → S2 from the Brillouin zone to the pseudospin space. The combined mapping T3 → S2 is indexed by the Hopf number16–21 Γ(p, q) = − 1 2π2 (cid:90) BZ By inserting (3) to this formula we obtain dk εµνρτ Nµ∂kx Nν∂ky Nρ∂kz Nτ . (5)  0, pq, −2pq, Γ (p, q) = 1 < m < 3 m > 3 m < 1 . (6) We note that the definition of the Hopf number (5) is different from the previous literature17, in which it is defined in terms of (η↑, η↓) rather than (z↑, z↓). The formula (6) is understood intuitively as follows: When p = q = 1, it is a well-known formula for the Hopf number17, which indicates that there ex- ist one circle in the interior of the torus and one circle around its axis of rotational symmetry. Now, z↑ ∝ ηp↑ and z↓ ∝ ηp↓ imply that there exist p and q of these circles. The energy spectrum of the Hamiltonian (1) reads E (k) = z (k). The Fermi surface of the topological Hopf insulator is constructed by the intersection of the three curved surfaces, Sx = Sy = Sz = 0. In general, the intersection of three surfaces is null, which results in an insulating state. ±(cid:113) S2 x (k) + S2 y (k) + S2 In order to construct a model having a Fermi surface, it is necessary to reduce the number of the conditions on the zero- energy states. There exist three trivial ways to do so. We may employ any one of the conditions, Sx = Sz = 0, Sy = Sz = 0, or Sx = Sy = 0. Obviously, they are the zero-energy conditions of the following three Hamiltonians, Hxz (k) = Sx (k) σx + Sz (k) σz, Hyz (k) = Sy (k) σy + Sz (k) σz, Hxy (k) = Sx (k) σx + Sy (k) σy, (7) (8) (9) respectively. The Fermi surface constructed by the intersec- tion of the two curved surfaces is a line node in general. In all these models we use the same CP1 field (2) to define Si (k). Only the Hamiltonian Hxz preserves the combination symme- try P T of the time reversal T and inversion symmetry P . The Hxz model: We first investigate the Hamiltonian Hxz. From the normalization condition on S (k), the zero-energy condition is expressed as Sy (k) = ±1. Namely, the Fermi surface is the preimage of the points S± = (0,±1, 0) in the combined Hopf map T3 → S2, which is a circle S1 in the 3D Brillouin zone. Consequently there are at least two loops corresponding to the preimage of Sy (k) = 1 and Sy (k) = −1 in the 3D Brillouin zone. These two loops form a link when the Hopf number is nonzero. Various links indexed by a pair (p, q) are realized in the 3D Brillouin zone. We show an almost zero-energy sur- face E = δ with δ (cid:28) 1 for (p, q) in Figs.1 and 2. The 2 FIG. 2: Top view of almost zero-energy surface of the Hamiltonian Hxz in the 3D Brillouin zone. See the caption of Fig.1 for (a)∼(d). preimage of Sy (k) = 1 is colored in magenta and that of Sy (k) = −1 is colored in cyan. For example, the Hopf link and the Solomon's knot are realized by taking pairs (1, 1) and (2, 1), respectively. The Fermi surface for (3, 2) is given by the combination of two trefoils, which we call a double trefoil. We recall the terminology in the link theory. A link lying on the surface of a torus is called a torus link. The torus link T (p, q) winds q times around a circle in the interior of the torus, and p times around its axis of the rotational symmetry. In the present context, the surface determined by the condition Sz = 0 gives a torus. Thus the node indexed by (p, q) is the torus link T (2p, 2q), where the factor 2 appears because it is the preimage of the two points S± = (0,±1, 0). According to the link theory, T (2p, 2q) is identical to T (2q, 2p). Further- more, T (−2p, 2q) link and T (2p,−2q) are mirror images of T (2p, 2q). If p and q are not relatively prime, we have a torus link with more than one component. The number of the loops is given by gcd(2p, 2q) = 2gcd(p, q), where gcd represents the great- est common divisor. For example, the Fermi surface consists of four loop nodes for p = 2 and q = 2. This looks a bit odd since we have discussed that two loops arise as the preim- age of Sy = ±1. It is an interesting problem how such a Fermi surface consisting more than two loops is realized for gcd(p, q) (cid:54)= 1. We assume gcd(p, q) = s. Then we can write p = sp(cid:48) and q = sq(cid:48), where p(cid:48) and q(cid:48) are coprime integers. The solutions are given by η1/s↑ , where η↑ and η↓ satisfy the Sy = ±1 for the model with p(cid:48) and q(cid:48). The roots of η1/s↑ have s solutions, which results in the s Fermi loops. and η1/s↓ and η1/s↓ In the following, we derive the equations to determine the link. The zero-energy states must satisfy the condition Sz = z†σzz = z↑2 − z↓2 = 0. By combining it with the normalization condition z↑2 + z↓2 = 1, the CP1 field is 3 FIG. 3: Bird's eye's view of the almost zero-energy surface of the Hamiltonian Hxy for (a) the pair (p, q) = (1, 1) and (b) (2, 1). parametrized as FIG. 4: Trefoil-knot semimetal with p = 3/2 and q = 1. (a) Bird's eye's view and (b) the top view of the almost zero-energy surface. There is a cut on the ky = 0 plane for kx < 0, where the magenta and cyan curves touch. 1√ 2 1√ 2 z↑ = (10) The zero-energy states correspond to Sy = ±1, which reads z↓ = eiθ↓ . eiθ↑ , Sy = z†σyz = − sin (θ↑ − θ↓) = ±1. The solution is given by θ↑ − θ↓ = ∓ π z↓ = ±iz↑, or 2 . We find the relation ηq↓ = ±iηp↑. (11) These two equations determine the Fermi surface made of links. The Hyz model: The Fermi surface of Hyz is topologically equivalent to that of Hxz model. The only difference is that the Fermi surface is rotated 90 degree between the Hxz and Hyz models. The Hxy model: The Fermi surface of Hxy looks very dif- ferent, where some of the Fermi surfaces form lines penetrat- ing the whole Brillouin zone: See Fig.3. The model Hxy is instructive since we obtain various analytical expressions. The zero-energy condition in the model Hxy is given by the preimage of Sz = z↑2 − z↓2 = ±1. By combining it with the normalization condition z↑2 + z↓2 = 1, the preimage of Sz = 1 is given by z↑2 = 1 and z↓2 = 0, which is equivalent to the condition z↓2 = η↓2q = 0, where η↓2 = sin2 kz + (cos kx + cos ky + cos kz − m)2 = 0. (12) The solution is We may also analyze this Hamiltonian system from the view point of the Berry curvature. We introduce a continuum model defined by η↑ (k) = kx + iky, η↓ (k) = kz + i(3 − m −(cid:0)k2 (cid:1) /2). x + k2 y + k2 z (14) The Fermi surface is topologically equivalent to the original model (3). The only difference is the approximation of the y = 2(3 − m), which does not loop (13) by the circle k2 change the linking structure. By using them we may derive explicitly the eigenstate ψ(cid:105) of the Hamiltonian (9). The Berry connection Aki = −i(cid:104)ψ ∂ki ψ(cid:105) is given by x + k2 k4 z + k2 p , 2 Ak = Aθ = Az = 2kqkz z (k2 + 2m − 4) + (k2 + 2m − 6)2 , z − k2 − 2m + 6(cid:1) q(cid:0)k2 k4 z + k2 z (k2 + 2m − 4) + (k2 + 2m − 6)2 , (15) (16) (17) where we have introduced the polar coordinate kx = k cos θ and ky = k sin θ. We show the Berry curvature in Fig.5. We find a vortex structure along a line node (cyan line in Fig.3) described by kx = 0, ky = 0, kz with −π ≤ kz ≤ π, and a y = 6 − 2m, kz = 0. Actually this circle described by k2 line node has a p-fold degeneracy. Indeed, we calculate the Berry phase along a loop encircling the line node to find that x + k2 cos kx + cos ky = m − 1, kz = 0, (13) ΓB = Aidki = Aθdθ = pπ, (18) which represents q-fold degenerate loop nodes. On the other hand, the preimage of Sz = −1 is given by z↑2 = 0 and z↓2 = 1, which is equivalent to the condition z↑2 = η↑2p = 0, where which is quantized and represents the degeneracy. In the same way, the Berry phase along a small circle around a loop node (magenta loop in Fig.3) is obtained as η↑2 = sin2 kx + sin2 ky = 0. ΓB = Aidki = Aθdθ = qπ, (19) The solution represents p-fold degenerate line nodes along the kz direction described by kx = nxπ, ky = nyπ, kz with −π ≤ kz ≤ π, where nx and ny are integers. representing the degeneracy of the loop node. Thus the in- dices p and q are detected separately, while only the product pq appears in the Hopf number. (cid:73) (cid:73) (cid:90) 2π 0 0 (cid:90) 2π 4 Torus-knot semimetals: We have seen that the Fermi sur- faces consist of at least two loops, which form a torus link. It is an interesting problem whether we can construct a Fermi surface consist of one nontrivial loop, which is a torus knot. Since the number of loops is given by 2gcd(p, q), we must take one of p and q non-integer. We find that a torus-knot Fermi surface is realized by taking a half-integer p. For exam- ple, we can realize a trefoil Fermi surface by taking p = 3/2 and q = 1, which is shown in Fig.4. There is a cut in the momentum space due to the square root contribution in the Hamiltonian. Namely, only the magenta or cyan curve does not form a closed loop but form a loop with the combination of them. Open string semimetals: It is possible to choose even any rational numbers for p or q, which creates an open string Fermi surface as shown in Fig.6. This is understood as fol- lows. The model with p and q gives the torus link T (2p, 2q). If one of the p and q is not a half integer, it cannot describe a closed loop, generating to an open string. The ends of the open string are on the cut plane. The author is very much grateful to N. Nagaosa for many helpful discussions on the subject. This work is supported by the Grants-in-Aid for Scientific Research from MEXT KAK- ENHI (Grant Nos.JP17K05490 and 15H05854). Note added: During the preparation of this manuscript, we became aware of closely related works26–28, where various linked nodal semimetals are proposed. Especially, this work has turned out to be a generalization of the work27, where only the case with p = q = 1 is studied. FIG. 5: (a) Stream plot of the Berry curvature (Ax, Ay) along the kx-ky plane with kz = 0 and (b) Stream plot of the Berry curvature (Ax, Az) on the kx-kz plane at ky = 0. FIG. 6: Open-string semimetal with p = 1/3 and q = 1. (a) Bird's eye's view and (b) top view of almost zero-energy surface. There is a cut along the ky = 0 plane for kx < 0, where the ends of the open string exist. 1 P. Hosur, X.L. Qi, C. R. Physique 14, 857 (2013). 2 S. Jia, S.-Y. Xu, M. Z. Hasan, Nature Materials 15, 1140 (2016). 3 S. Murakami, New J. Phys. 9, 356 (2007). 4 T. Bzduek, Q.-S. Wu, A. Ruegg, M. Sigrist and A. A. Soluyanov, Nature 538, 75 (2016). 5 S. Mandal and N. Surendran Phys. Rev. B 79, 024426 (2009). 6 A. A. Burkov, M. D. Hook, and L. Balents, Phys. Rev. B 84, 7 M. Phillips and V. Aji, Phys. Rev. B 90, 115111 (2014). 8 C. Fang, Y. Chen, H.-Y. Kee and L. Fu, Phys. Rev. B 92, 081201 235126 (2011). (2015). Rev. B 93, 205132 (2016). 16 J. E. Moore, Y. Ran and X.-G. Wen, Phys. Rev. Lett. 101, 186805 17 D.-L. Deng, S.-T. Wang, C. Shen, and L.-M. Duan, Phys. Rev. B 18 D.-L. Deng, S.-T. Wang, and L.-M. Duan, Phys. Rev. B 89, (2008). 88, 201105(R) (2013). 075126 (2014). mat/arXiv:1612.01518. 19 D.-L. Deng, S.-T. Wang, K. Sun, and L.-M. Duan, cond- 20 R. Kennedy, Phys. Rev. B 94, 035137 (2016). 21 C. Liu, F. Vafa and C. Xu cond-mat/arXiv:1612.04905. 22 Y. Kawaguchi, M. Nitta, and M. Ueda, Phys. Rev. Lett. 100, 9 L. S. Xie, L. M. Schoop, E. M. Seibel, Q. D. Gibson, W. Xie, and R. J. Cava, APL Materials 3, 083602 (2015). 180403 (2008). 10 R. Yu, H. Weng,Z. Fang, X. Dai and X. Hu, Phys. Rev. Lett. 115, 23 D. S. Hall, M. W. Ray, K. Tiurev, E. Ruokokoski, A. H. Gheorghe 11 Y. Kim, B. J. Wieder, C. L. Kane, and A. M. Rappe, Phys. Rev. 24 H. Kedia, I. Bialynicki-Birula, D. Peralta-Salas, and W. T. M. 036807 (2015). Lett. 115, 036806 (2015). Soc. Jpn. 85, 013708 (2016). 12 A. Yamakage, Y. Yamakawa, Y. Tanaka, Y. Okamoto, J. Phys. 13 M. Ezawa, Phys. Rev. Lett. Phys. Rev. Lett. 116, 127202 (2016). 14 M. Hirayama, R. Okugawa, T. Miyake, and S. Murakami, Nat. Com. 8, 14022 (2017). 15 Y.-H. Chan, C.-K. Chiu, M. Y. Chou, and A. P. Schnyder, Phys. and M. Mottonen, Nat. Phys. 12, 478 (2016). Irvine, Phys. Rev. Lett. 111, 150404 (2013). 25 P. J. Ackerman and I. I. Smalyukh, Phys. Rev. X 7, 011006 (2017). 26 W. Chen, H.-Z. Lu, and J.-M. Hou, cond-mat/arXiv:1703.10886. 27 Z. Yan, R. Bi, H. Shen, L. Lu, S.-C. Zhang, and Z. Wang, cond- mat/arXiv:1704.00655. 28 P.-Y. Chang and C.-H. Yee, cond-mat/arXiv:1704.01948. 

1604.03125

1

1604

Covalent functionalization by cycloaddition reactions of pristine, defect-free graphene

Based on a low temperature scanning tunneling microscopy study, we present a direct visualization of a cycloaddition reaction performed for some specific fluorinated maleimide molecules deposited on graphene. These studies showed that the cycloaddition reactions can be carried out on the basal plane of graphene, even when there are no pre-existing defects. In the course of covalently grafting the molecules to graphene, the sp2 conjugation of carbon atoms was broken and local sp3 bonds were created. The grafted molecules perturbed the graphene lattice, generating a standing-wave pattern with an anisotropy which was attributed to a (1,2) cycloaddition, as revealed by T-matrix approximation calculations. DFT calculations showed that while both (1,4) and (1,2) cycloaddition were possible on free standing graphene, only the (1,2) cycloaddition could be obtained for graphene on SiC(0001). Globally averaging spectroscopic techniques, XPS and ARPES, were used to determine the modification in the elemental composition of the samples induced by the reaction, indicating an opening of an electronic gap in graphene.

cond-mat.mes-hall

cond-mat

Daukiya et al. Covalent functionalization by cycloaddition reactions of pristine, defect-free graphene L. Daukiya 1, C. Mattioli2, D. Aubel1, S. Hajjar-Garreau1, F. Vonau1, E. Denys1, G. Reiter3, J. Fransson4, E. Perrin5, M-L. Bocquet5, C. Bena6−7, A. Gourdon2 and L. Simon1∗ 1Institut des Sciences des Mat´eriaux de Mulhouse, CNRS-UMR 7361, Universit´e de Haute Alsace, 3Bis, rue Alfred Werner, 68093 Mulhouse, France. 2 Nanosciences group, CEMES CNRS-UPR 8011, 29 Rue Jeanne Marvig, BP 94347, 31055 Toulouse, France 3Physikalisches Institut, Universitat Freiburg, Hermann-Herder-Strasse 3, 79104 Freiburg, Germany 4Department of Physics and Astronomy, Uppsala University, Box 516, SE-751 21 UPPSALA, Sweden 5 Dpt of Chemistry, UMR ENS-CNRS-UPMC 8640, Ecole Normale Superieure, F-75005 Paris, France 6Institut de Physique Th´eorique, CEA/Saclay, Orme des Merisiers, 91190 Gif-sur-Yvette Cedex, France and 7Laboratoire de Physique des Solides, CNRS, UMR-8502. Paris Sud, 91405 Orsay CEDEX, France (Dated: September 20, 2018) ∗ corresponding author Email address: laurent.simon@uha.fr 6 1 0 2 r p A 1 1 ] l l a h - s e m . t a m - d n o c [ 1 v 5 2 1 3 0 . 4 0 6 1 : v i X r a Typeset by REVTEX 1 Abstract Based on a low temperature scanning tunneling microscopy study, we present a direct visual- ization of a cycloaddition reaction performed for some specific fluorinated maleimide molecules deposited on graphene. These studies showed that the cycloaddition reactions can be carried out on the basal plane of graphene, even when there are no pre-existing defects. In the course of covalently grafting the molecules to graphene, the sp2 conjugation of carbon atoms was broken and local sp3 bonds were created. The grafted molecules perturbed the graphene lattice, gener- ating a standing-wave pattern with an anisotropy which was attributed to a (1,2) cycloaddition, as revealed by T-matrix approximation calculations. DFT calculations showed that while both (1,4) and (1,2) cycloaddition were possible on free standing graphene, only the (1,2) cycloaddition could be obtained for graphene on SiC(0001). Globally averaging spectroscopic techniques, XPS and ARPES, were used to determine the modification in the elemental composition of the samples induced by the reaction, indicating an opening of an electronic gap in graphene. PACS numbers: 68.65.-k, 81.16.Fg, 81.07.-b, 81.16.Rf, 82.30.RS, 82.65.+r 2 Because of its fascinating properties, graphene is presently considered for a large number of potential applications. However, functionalizing this gapless highly non-reactive semi- conductor with the appropriate molecules remains an important challenge. Combined with supramolecular chemistry, chemical functionalization, i.e., creating covalent bonds by con- verting sp2 into sp3 orbitals, represents a promising path, because it allows to selectively modify the surface of graphene with a high spatial control. Such a surface modification (hybridization) can be realized by Diels-Alder or other cycloaddition reactions. Graphene undergoes cycloaddition or Diels Alder (D-A) reactions mainly because of the degeneracy of the electronic states at the Dirac point. The states close to the Fermi level may give rise to either an anti-symmetric or symmetric graphene orbital. These orbitals al- low graphene to function as both donor and acceptor within the Frontier Molecular Orbital (FMO) theory [1 -- 4]. For exfoliated graphene, a D-A reaction and its reversibility have been demonstrated for the first time by the team of Robert C. Haddon [1]. The progress of the D-A reaction was followed by Raman spectroscopy and the ratio between the G and D bands which ascertained that graphene was modified. However, such global measurements did not allow to identify the locations where graphene was actually functionalized. This seminal work by Haddon et al. raises the question if pre-existing defect (step-edges, holes,...) are required to allow for the D-A reaction. Subsequently, two theoretical studies [5, 6] con- cluded that only the edges of graphene layers, or holes therein, may be functionalized by a cycloaddition reaction. However, no such reaction was predicted to be possible within the bulk of a clean graphene layer. Indeed, calculation results for a D-A reaction in the bulk predict a highly endothermic value of up to 2.6 eV. Recently, by combining scanning tun- neling microscopy (STM) and density functional theory (DFT) calculations, a cycloaddition reaction has been evidenced for a graphene layer on iridium, involving a non-expected 1,3 cycloaddition of graphene in the low electron density regions of the Moir´e pattern [7]. There, the allyl reactivity of epitaxial graphene was due to the presence of the substrate Ir. The endothermic character of the cycloaddition reaction was preserved, but was characterized by a more modest value of + 0.4 eV, accessible by choosing an appropriately high temperature. This observation demonstrates that a D-A reaction is possible when using a graphene layer 'activated' by an underlying metal substrate. Here, we combine STM, angle resolve photoemission spectroscopy (ARPES) and X-ray photoelectron spectroscopy (XPS) to study the Diels-Alder reaction on a monolayer (ML) 3 and a bilayer (BL) epitaxial graphene on SiC(0001), using recently synthesized fluorinated maleimide molecules (see Figure1). The maleimide group of these molecules represents the reactive part, and the fluorine atoms were introduced to polarize the double bonds of the maleimide group and thereby strongly increased the dienophilic character of the molecule and thus its reactivity [8]. To deposit these molecules onto graphene, the substrates were simply immersed for several tens of hours in a toluene solution containing the molecules. After retraction from the solution, the samples were intensively rinsed. Our results provide strong evidence of the formation of chemical bonds between these molecules and the graphene layer through the cycloaddition reaction . The formation of the chemical bonds was ascertained by the increase of the sp3 component of the C1s peak, and a decrease of the Fermi velocity. In addition, a tendency for the opening of a gap was evidenced by ARPES. Moreover, the STM images provide also a strong evidence for the formation of covalent bonds by a cyclo-addition reaction in the middle of clean graphene terraces. Indeed, chemi- cally grafting these molecules to the graphene layer generates standing waves pattern which indicates that graphene is locally highly perturbed. The grafted molecules are associated to bright features on the STM images that are dispersed throughout the bulk of graphene terraces where no defect is expected in pristine graphene, at least not with such a high density. The smallest of these features are surrounded by standing waves which take the form of parallel straight lines, similar in anisotropy to those observed close to armchair graphene step edges. We present an experimental comparison of these standing-wave pat- terns with patterns generated by other possible defects, as well as with patterns generated by armchair and zigzag step edges. We also discuss how identifying the type of dominant quasiparticle scattering may reveal the nature of the cycloaddition reaction. Thus we use T-matrix approximation calculations to compare quasiparticle interference patterns and the local density of states when assuming various possible configurations of atoms affected by the cycloaddition reaction, in particular a one atom (1,1) configuration, and some two-atom configurations [(1,2), (1,3), or (1,4)] . We find that the shape of the standing-wave patterns depends not only on the defect geometry, but also on the type of sublattice combinations (AA or AB) of the atoms affected by the reaction. The highly anisotropic perturbation pattern of the density of states was reproduced for the (1,2) and (1,4) configurations of the cycloaddition reaction. Moreover, we show via large-scale DFT calculations on realistic graphene layers on a SiC substrate that only the (1,2) cycloaddition is stable, but not the 4 (1,4) or (1,3) ones. On bilayer graphene, the STM revealed similar patterns; moreover using XPS we measured an increase of the sp3 component in the peak corresponding to the C1s core level, and the band structure modifications were significantly more pronounced. In this study, we used maleimide derivative molecules. As depicted in figure 1, when in- teracting with these molecules in a D-A reaction, graphene can be considered as a diene reacting with a dienophile molecule. However, at this stage it was not possible to predict which type of cycloaddition reaction will occur. This is particularly difficult for graphene on SiC(0001) which is n-doped (thus the chemical reaction does not occur at the Dirac point). Specifically, we used 3,5- bis(trifluoromethyl)phenyl substituted maleimide derivatives (for which we use the abbreviated name: fluorinated maleimide (FMAL)). Several maleimide- type molecules (M1 to M3) have been tested. The reactivity of the molecules has been studied for graphene on SiO2 and graphene on SiC(0001) using Raman spectroscopy[9]. It was found that only M3-type molecules reacted with epitaxial graphene. We demonstrate here that we can chemically functionalize graphene, a fascinating possibility which opens up the road for future research in the functionalization of graphene, as well for more systematic studies in particular using higher spatial resolution investigations. EXPERIMENTAL METHOD The scheme in figure 2 illustrates the growth of graphene on SiC (0001) by thermal evap- oration of Si atoms. A SiC(0001) substrate was annealed at temperatures above 1250◦C [10, 11]. This led to the exodiffusion of silicon atoms and the formation of a buffer layer which is a graphene monolayer partially covalently bonded to the silicon atoms of the SiC(0001) substrate exhibiting a SiC-6x6 superstructure. This buffer layer is semiconducting. The graphene monolayer (ML graphene) or graphene bilayer (BL graphene) showed the char- acteristic properties of graphene and interacted via van der Waals forces with the buffer layer. We controlled the number of graphene layers via the annealing time and anneal- ing temperature. To obtain homogeneous surfaces, we assured a low pressure (less than 1.10−10mbar) during the annealing. The resulting graphene layers were characterized by XPS and ARPES measurements. We used an hemispherical electron analyzer (Scienta R3000), and a monochromatic X-ray source (AlKα). For ARPES we used the UV source HeII (40.6eV). Our STM experiments were performed at 77 K at a base pressure in the 5 10−11 mbar range in ultra high vacuum (UHV) with a LT-STM from Omicron. Following the procedure represented in fig .2, the deposition of molecules was carried out on monolayer as well as bilayer graphene. THEORETICAL METHOD The DFT calculations were carried out using the Vienna ab initio simulation package (VASP)[12, 13], with the generalized-gradient approximation of PBE type [14] as the XC functional. The electron-ion interaction was described by the projector augmented wave method [15, 16]. Plane waves were used as the basis set, and the energy and augmentation charge cutoffs were set to 300 eV and 645 eV. In order to get a detailed and realistic understanding of the graphene/SiC interfacial structure on the Si-face of our SiC substrates, and to derive its influence to the electronic structure of graphene, we adopted an adequately √ large area of commensurability, i.e., a 13 × 13 graphene domain on a 6 substrate (referred as to 6R3) [17 -- 19]. The SiC substrate was modeled by two SiC bilayers, √ 3 × 6 3 SiC the top face being Si and the bottom being C, saturated by hydrogen atoms, in total including 432 atoms. All structures were relaxed until the total forces were smaller than 0.05 eV/A. On this substrate, different types of cycloaddition reactions with the fluorinated maleimide molecule have been calculated. The stabilized cycloadducts have been compared with respect to structure and energy with the ones formed on free-standing graphene. RESULTS Figure 3 A) shows a typical XPS spectrum of a pristine ML graphene. We can identify four components. The C1s component of SiC corresponds to the carbon atoms of the SiC substrate. The C − sp2 component is attributed to the graphene top layer and the S1 − sp3 and S2 − sp2 components correspond to the buffer layer carbon atoms covalently bonded to silicon atoms and carbon atoms in the graphene-like structure of the buffer layer, respectively [20, 21]. Figure 3 B) shows the corresponding peaks for the same sample after having been immersed in a FMAL toluene solution for 80 hours, followed by the rinsing procedure described in figure 2. The presence of FMAL molecules on the surface of graphene 6 resulted in a marked C-sp2 (G) component and emergence of a new C-sp3 component shown as dotted line in figure3B). The attachment of molecules via their aromatic ring contributed to the increase of the component G. C-sp3 is associated to the graffting of the molecules, this component emerges in the deconvolution by keeping constant the ratio of S1-sp3 and S2-sp2 with the SiC component before and after the immersion which leads to consider that the buffer layer is unchanged in the process. The formation of sp3 bonds, expressed by the increase of the component S1, was associated with the grafting of the molecules. The other contributions of carbon configurations were less important in intensity but still could be identified. Typically, as we will demonstrate for a cycloaddition for the (1,2) configuration, grafting of a single molecule changed the sp2 orbitals of 5 bonds into sp3 orbitals. Even for longer immersion times, due to the low sensitivity factor of the organic component, the contributions of grafted molecules in the C1s spectra were much less intense and contributed mainly to the tail of the C1s peak, i.e. for binding energies above 288 eV (see figure D) in supporting information). As provided in the supplementary information, the O1s, N1s and F1s spectra and their respective positions showed unambiguously that the FMAL molecules were deposited in the intact form on the graphene surface. In Figure 3E), we present a plot of the intensity of the S1 component, together with the intensity of F1s, for immersion times ranging from 20 to 80 hours. Both intensities exhibited the same evolution, assuring that the increase in sp3 bonds could be attributed to an increase in number of grafted molecules. Figures 3 C) and D) show the dispersion bands measured by ARPES around the K-point. The band structure was measured before and after deposition of molecules on monolayer graphene (Figure 3 C) and bilayer graphene (Figure 3 D), respectively. We observe that the band structure was preserved and could clearly be measured by ARPES, even after immersion and without any further treatment of the sample in UHV. A tendency for the opening of a small gap was marked by the decrease of the intensity around the Dirac point which is much more pronounced than for pristine graphene. This is clearly visible on the profiles taken as a function of energy at constant momentum at the K point (see figure 3). In addition, the Fermi velocity, measured via the slope of the linear dispersion, decreased with immersion time. This indicates that the number of covalently bonded molecules in- creased. A similar behavior has been observed, for example, for induced defects caused by ion bombardment [22]. We notice that the opening of the gap was more pronounced for the 7 case of bilayer graphene shown in Figure 3 D). The dispersion of the lower energy band, which corresponds to the upper graphene layer, and that initially was linearly crossing the Dirac point became rather parabolic. This change revealed an increase of the effective mass, as expected, and is consistent with the opening of a gap. In figure 4, we present STM images of a typical sample. In A), a large image (400x400 nm2) of a pristine and defect-free graphene layer, as loaded in the UHV system, is shown together with a sample after 80 hours of immersion leading to the deposition of molecules on the graphene surface. We observe dark regions which correspond to the buffer layer. After immersion, we additionally observe many bright features, corresponding to the fluorinated maleimide molecules and potentially to some products introduced via the rinsing process. Some of these deposited species were found to be highly mobile on the graphene surface, even at 77K, generating difficulties for obtaining proper STM images. To verify that the images contained the signature of grafted molecules, we first scanned at low bias (at a few meV and large currents up to 1A). This allowed to remove the potentially remaining free species. After performing several scans and cleaning steps of the tip (achieved through high voltage pulses performed in areas outside the area of interest), we observed more stable bright features accompanied by standing-wave patterns, as shown in figure 4 B). These standing- wave patterns revealed a strong local modification of the underlying graphene lattice. The 3 × √ √ corresponding 3 modulation of the density of states is commonly observed around defects, generated, for example, by ion bombardment [22], or by defects in the graphene layer (missing atoms), often decorated with impurities. These defects break the conjugation of the sp2 bonds of graphene. This seems to be a necessary condition for the observation of strong standing-wave patterns. Indeed in a recent detailed STM study of nitrogen doping of graphene by substitution [23], standing-wave patterns were observed around N atoms substituting the C atoms only if vacancies were also present, such as for the pyridinic-N defect. In the latter case the conjugation is broken. Quasi-particle interferences having a momentum-vector which connects two different iso-energy contours around the K-point (K- K'), as described in the scheme shown in figure 4 D) are also called intervalley scattering [24, 25]. In our case, for the smaller objects (see encircled object in the right lower part of figure 4 B)), the standing-wave patterns were anisotropic with parallel straight lines exhibiting a large contrast variation. Such standing-wave patterns with large amplitudes are usually observed 8 at armchair step edges, as shown in figure 4 F) [27]. On the other hand patterns resulting from point defects or large defects (holes) typically show a six-fold symmetry [24, 26]. An example of such pattern is observed around the large object surrounded by arrows in figure 4 B) and exhibits a more isotropic standing wave pattern, which we attribute to a pile containing several molecules. Here, we discuss the possible standing-wave patterns created by a single molecule grafted onto the defect-free surface of graphene in analogy with the step-edge geometry. Figure 5 A) shows different configurations (n,p) for the cycloaddition reaction, where n and p represent the positions of the carbon atoms. There, grafting of a molecule changes a sp2 bond into sp3 bond. A (1,1) configuration corresponds to a single- site modification such as for a point defect, whereas a (1,p) configuration corresponds to a molecule grafted at two sites as expected for the D-A reaction of maleimide molecules. For the (1,1) configuration, the electronic states of only one atom are perturbed (the molecule creates only one bond with the graphene surface). In the case of the formation of a sp3 bond, the conjugation with the three first nearest neighbor bonds is broken (these affected bonds are represented by dotted lines). The resulting defect in the graphene lattice acts like a missing atom (a point defect for the (1,1) configuration) for which the three nearest neighbor atoms (represented as blue dots in the figure), which belong to the sub-lattice A, act as scatterers for the surrounding graphene quasiparticles. For the cycloaddition configuration (1,2), corresponding to grafting at the first and second carbon atom of a hexagonal cycle, the remaining scatterer atoms belong to both sub-lattices, i.e., A and B. If we consider the (1,2) sp3 bonded atoms as missing atoms in the sp2 network of graphene, the resulting defect is anisotropic and has a structure similar to that of an armchair-like edge. For a cycloaddition configuration (1,3), we achieve an isotropic defect with scatters atoms belonging only to the sub-lattice A. For the (1,4) cycloaddition configuration, the generated pattern becomes more isotropic, however, scatterers belong, as for the (1,2) configuration of the cycloaddition reaction, to both sub-lattices, i.e., A and B. For the analysis of our experimental results, we thus focused on identifying the type of sub-lattice affected by the reaction, because this allowed us to discriminate between patterns arising from zigzag-like and armchair-like step edges. Consequently, we are able to decide which type of defect symmetry (armchair-step- edge or zigzag-step-edge) was characteristic to the cycloaddition reaction and thus generated a characteristic pattern via the scattering and interference process of quasiparticles. In figures 6 B), we describe the modified electronic structure due to impurity scattering 9 for the impurity configurations described in (figure 6 A)), respectively. We use T-matrix approach to obtain the real space electron Green function (GF) G(r, r(cid:48), ω) . The local density of state N (r) in real space is related to the GF through N (r) = −ImG(r, r(cid:48), ω)/π. Only elastic scattering is considered [32 -- 35]. We show in figure 6 C) the Fourier transforms (FT) of the local density of state calculated using the Born approximation, which is quasi-equivalent in this limit, to the k space T-matrix approximation [24, 36]. For a (1,1) point defect, the modification of the the density of states exhibits a clear six-fold symmetry. For the (1,2) configuration, the calculation shows a rectangular-like anisotropic pattern generated by the defect. For the (1,3) configuration, we recover a hexagonal symmetry, and for the (1,4) configuration, an anisotropic pattern similar to, but less pronounced than the pattern corresponding to the (1,2) configuration, the difference consisting in that the (1,4) patterns recovers a hexagonal symmetry at larger scale. These calculations show that, while the shape of the defect is important, the type of sub-lattice atoms affected by the cycloaddition reaction is even more important. This can be understood by analyzing the patterns of the Fourier transforms (FT) of the direct space images (figure 6 C)) in comparison with the experimental FT shown in the insert figure 4C). For the (1,1) configuration, representing a simple point defect, the FT revealed strong features around the Γ points, and less intense features around the K points with isotropic intensity (the intensity is the same for the six K-point features). A strong anisotropy is however observed for the (1,2) configuration of the cycloaddition reaction. The modified local density of states (LDOS) around the perturbed atoms shows a square- like anisotropic structure, and in the FT pattern, this translates into the Γ-point intensity being less intense than for the (1,1) case and into a strong anisotropy between the six features around the K points. The reduction in the Γ-point intensity corresponds to the fact that in this case the intervalley scattering processes are dominant, similar to the standing-wave patterns observed for armchair step edges. For the (1,3) configuration of the cycloaddition reaction, the FT pattern showed pro- nounced structures around the Γ points and less intense features around the K points, as for the point defect (1,1). Here only scatterers form the A sub-lattice contributed which favors intervalley scattering, similar to the standing-wave patterns observed in the vicinity of zigzag step edges. For the last case, the (1,4) configuration of the cycloaddition reaction, same as for the (1,2) configuration of the cycloaddition reaction, we observe a decrease of 10 intensity for the features around the Γ points, and an anisotropy in the features around the K points. We conclude that the (1,2) and (1,4) configurations of the cycloaddition reaction reveal more anisotropic and more pronounced standing-wave patterns than the point defect (1,1) and the (1,3) configuration, and are thus the configurations consistent with the ob- served experimental features. Finally, we have performed DFT calculations of interfacial properties (see THEORETICAL METHOD) in order to reveal the role of the substrate with respect to forming a stable chemical bond between graphene and the molecules by a cycloaddition reaction. Our DFT calculations showed that both the (1,2) and (1,4) configuration would allow a stable bond between the molecule and a free standing graphene layer. However, such a bond could not form for the (1,3) configuration. For graphene on SiC(0001), only the (1,2) configura- tion resulted in a stable bond, as depicted in Figure 6. The corresponding total energy of the resulting bond was by ca. 0.4 eV lower as compared to the same bond on free stand- ing graphene. Using a SiC(0001) substrate is therefore favoring the cycloaddition reaction. However, the (1,2) cycloaddition for graphene on SiC exhibits an energy level which is by ca. + 2 eV higher than that of the reactant. This high energy value would not normally allow for the observed cycloaddition reaction, which, as we show, was however possible at room temperature. Nevertheless we want to point out that we have performed this reaction not in vacuum (as assumed in the calculations) but by immersing the graphene surface for about 80 hours in a toluene solution of molecules. This difference in the deposition conditions hints at a lowering of the vacuum endothermicity due to the dissolution of the molecules by toluene. CONCLUSION The results of our study provide a strong evidence that cycloaddtion reaction between graphene and fluorinated maleimide molecules is possible at room temperature. We observed that this reaction happened on a region of pristine graphene layer, which did not contain any pre-existing defect. The formation of covalent bonds was ascertained by the increase of the sp3 component of the XPS spectrum. Furthermore, using STM we have visualized marked standing-wave patterns on graphene, which were not observed for physisorbed molecules. These patterns, in particular their geometry and symmetry, indicated that the 11 cycloaddition reaction occurred in the (1,2) or (1,4) configuration, as confirmed by the T-matrix approximation. DFT calculations clearly show that (1,2) configuration is stable on G/SiC(0001). We have interpreted the standing-wave patterns observed in direct space in analogy to standing-wave patterns observed in the vicinity of armchair and zig-zag step edges on graphene. ARPES measurements revealed a tendency for the opening of a gap and a slight decrease of the Fermi velocity. For bilayer graphene, the gap opening was more pronounced. Furthermore, we were able to distinguish which sub-lattice of graphene was perturbed by covalently grafting fluorinated maleimides molecules to graphene. In the future, in order to generate periodic molecular patterns chemically attached to graphene, we will incorporate the reactive group used here in larger molecules which, in addition, allow for supramolecular self-assembly. We expect that, by creating covalent bonds, we will be able to form ribbons with a precise control of the nature (armchair or zigzag) of their edges. This work was supported by the R´egion Alsace, the ANR ChimigraphN, the Universit´e Franco-Allemande and the ERC Starting Independent Researcher Grant NANOGRAPHENE 256965. [1] S. Sarkar, E. Bekyarova, S. Niyogi, R. C. Haddon, J. Am. Chem. Soc., 2011, 133, 3324. [2] S. Sarkar, E. Bekyarova, and R. C. Haddon, Accounts of Chemical Research, 2012, 45(4), 673. [3] K. Fukui, Theory of Orientation and Stereoselection, Top. Curr. Chem., 1970, 15(12), 1. [4] R. E. Townshend, G. Ramunni, G. Segal, G. W. J. Hehre, L. Salem, J. Am. Chem. Soc., 1976, 98, 2190. [5] Y. Cao, S. Osuna, Y. Liang, R. C. Haddon and K. N. Houk, J. Am. Chem. Soc, 2013, 135 (46),17643. [6] P.A. Pablo, Chemistry A European Journal, 2013, 19, 15719. [7] S.J. Altenburg, M. Lattelais, B. Wang and M.-L. Bocquet and R. Berndt, J. Am. Chem. Soc., 2015, 137, 9452. [8] Cristina MATTIOLI Thesis: "Design, synthesis and study of molecules for graphene func- 12 tionalization". Chapter 6: "Maleimide derivatives for covalent functionalization of graphene". University of Toulouse. [9] M. Rubio-Roy, C. Mattioli, O. Couturaud, J.R. Huntzinger, A. Gourdon, E. Dujardin, sub- mitted. [10] L. Simon, J. L. Bischoff, and L. Kubler, Phys. Rev. B, 1999, 60, 11653. [11] C. Berger, Z. Song, T. Li, X. Li, A. Y. Ogbazghi, R. Feng, Z. Dai, A. N. Marchenkov, E. H. Conrad, P. N. First, and W. A. de Heer, J. Phys. Chem. B, 2004, 108, 19912. [12] G. Kresse and J. Furthmuller, Computational Materials Science, 1996, 6, 15. [13] G. Kresse and J. Furthmuller, Phys. Rev. B,1996, 54, 11169. [14] J.P. Perdew, K. Burke and M. Ernzerhof, Phys. Rev. L, 1996, 77, 3865. [15] P.E. Blochl, Physical Review B, 1994, 50, 17953. [16] G. Kresse and D. Joubert, Phys. Rev. B, 1999, 59, 1758. [17] C. Riedl, U. Starke, J. Bernhardt, M. Franke, and K. Heinz, Phys. Rev. B, 2007, 76, 245406. [18] A. Charrier, A. Coati, T. Argunova, F. Thibaudau, Y.Garreau, R. Pinchaux, I. Forbeaux, J.-M. Debever, M. Sauvage-Simkin and J.-M. Themlin, Journal of Applied Physics, 2002, 92, 2479. [19] V.W. Brar, Y. Zhang, Y. Yayon, A. Bostwick, T. Ohta, J.L. McChesney, K. Horn, E. Roten- berg, and M. F. Crommie, Applied Physics Letters, 2007, 91, 122102. [20] K. V. Emtsev, F. Speck, Th. Seyller, and L. Ley, Phys. Rev. B, 2008, 77, 155303. [21] C. Riedl C. Coletti, U. Starke, J. Phys. D: Appl. Phys., 2010, 43, 374009. [22] L. Tapaszt`o, G. Dobrik, P. Nemes-Incze, G. Vertesy, Ph. Lambin, and L. P. Bir`o, Phys. Rev. B,2008, 78, 233407. [23] Y. Tison, J. Lagoute, V. Repain, C; Chacon, Y. Girard, S. Rousset, F. Jouckert, D. Sharma, L. Henrard, H. Amara, A. Ghedjatti, and F. Ducastelle, ACSNano, 2015, 9, 670. [24] L. Simon, C. Bena, F. Vonau, M. Cranney, and D. Aubel, J. Phys. D,2011, 44, 464010. [25] M. Cranney, F. Vonau, P. B. Pillai, E. Denys, D. Aubel, M. M. De Souza, C. Bena, and L. Simon, Europhys. Lett., 2010, 91, 66004. [26] L. Simon, C. Bena, F. Vonau, D. Aubel, H. Nasrallah, M. Habar and J.C. Peruchetti, Eur. Phys. J. B, 2009, 69, 351. [27] Step edges play an important role in determining transport properties [28], particularly in the case of ribbons of graphene, and are widely studied. Interpretation of the resulting edge 13 electronic patterns are subject of intense debates, notably concerning the scattering processes from so-called zigzag or armchair edges [29 -- 31] [28] Ken-ichi Sasaki, Katsunori Wakabayashi, Phys. Rev. B,2010, 82, 035421. [29] H. Yang, A; J. Mayne, M. Boucherit, G. Comtet, G. Dujardin and Y. Kuk, Nano Lett., 2010, 10, 943. [30] C. Parka, H. Yangb, Andrew J. Mayne, G. Dujardin, S. Seod, Y. Kuka, J. Ihma, and G. Kimd, PNAS, 2011, 10846, 18622. [31] A. Mahmood, P. Mallet and J-Y Veuillen, Nanotechnology,2012, 23 055706. [32] G. A. Fiete, E. J. Heller, Rev. Mod. Phys., 2003, 75, 933. [33] J. Fransson and A. V. Balatsky, Phys. Rev. B,2012, 85 161401R. [34] H. Hammar, P. Berggren and J. Fransson, Phys. Rev. B, 2013, 88, 254418. [35] J. Fransson, J-H. She, L. Pietronero, and A. V. Balatsky, Phys. Rev. B, 2013, 87, 245404. [36] C. Bena and L. Simon Phys. Rev. B 83 (2011)115404. 14 Figures Figure 1: Description of a Diels-Alder reaction as an example of cycloaddition on pristine graphene. Here, graphene is considered as a diene which reacts with a dienophile molecule. We used 3,5- bis(trifluoromethyl)phenyl substituted maleimide derivatives (named fluorinated maleimide (FMAL)). Several molecules of the maleimide-type M1 to M3 have been tested. Only M3-type molecules reacted with epitaxial graphene. The fluorinated atoms are introduced to polarise the double bond of the maleimide group, and so strongly increased the dienophilic character of the molecule [8]. 15 Figure 2: . (A). Description of the growth of Graphene on SiC. Graphene was generated by exodiffusion and evaporation of silicon atoms from the SIC substrate by annealing at appropriate temperatures. This process lead to formation of covalently bonded layer of carbon atoms called buffer layer, and the subsequent formation of one or more graphene layers. (B) Monolayer or Bilayer graphene samples were dipped in the solution of FMAL molecules in toluene for different duration. (C) shows the rinsing procedure for removal of physisorbed molecules 16 Figure 3: Global measurements (XPS and ARPES) of graphene before and after the reaction with FMAL molecules (dipping duration 80 hours). A) and B) shows the deconvoluted C1s spectra before and after dipping respectively. The successful reaction of FMAL molecules is represented by the change in sp2 component which is associated with graphene and sp3 (dotted lines). In the supplementary information, we present the F1s, N1s and O1s components which assure that the molecule is intact. E) shows the evolution of sp3 component of C1s spectra and F1s spectra as a function of dipping time. C) and D) shows the band structure measured by ARPES for monolayer and bilayer graphene before and after the DA reactions, respectively. The constant momentum dispersion curve shows a dip in the intensity as an a tendency for opening of a gap. The change in the Fermi velocity for ML graphene as a function of the immersion time is given in F). . 17 18 Figure 4: A) Large area (400x400 nm2) STM scan (-1.5 V, 500 pA) on a pristine, defect-free graphene layer, as loaded in UHV, together with the corresponding image (-1.5V, 518 pA) for this sample after immersion for 80 hours in a toluene solution of FMAL molecules. (B) 20x20 nm2 area image (-888 mV and 500pA) showing bright features surrounded by standing-wave patterns, suggesting a strong modification of graphene layer as a result of covalently grafted molecules. The bright features which show standing waves pattern are attributed to grafted molecules are highlighted in the insert in B). C) Zoomed image on small features for a sample immersed for 21 hours (-1.1V and 128 pA ). The insert in C) shows the Fourier Transform (FT) of the image. D) Schematic representation of the reciprocal lattice and possible scattering momentum vectors, representing zig-zag or arm-chair step edges, respectively. We represent the intravalley (momentum which connects two isocontours of same nature (K-K or K'-K') and intervalley scattering (for two different isocontour K-K') observed for zig-zag and armchair step edges, respectively. Interference patterns observed for (E) a zig-zag step edge (-66.7mV, 121pA, 8.4x3.9nm2) and for (F) an armchair step edge (-27.3mV, 500pA, 8.4x3.9nm2). 19 Figure 5: A) Schematic representation of the possible defects created by the formation of sp3 bonds. For a point defect configuration (1,1) and cycloaddition configurations (1,2), (1,3) and (1,4), respectively. T-matrix approximation calculations of the modified local density of states in direct space (B) and in K-momentum space (power spectrum) (C), respectively. Figure 6: DFT optimized structure of the 1,2 cycloadduct for graphene on SiC. a) Side view of atomistic details of the supercell employed in the DFT calculations. b) Zoomed top view of a) 20 

1301.7635

1

1301

Current-field diagram of magnetic states of a surface spin valve in a point contact with a single ferromagnetic film

We present a study of the influence of an external magnetic field H and an electric current I on the spin-valve (SV) effect between a ferromagnetic thin film (F) and a sharp tip of a nonmagnetic metal (N). To explain our observations, we propose a model of a local surface SV which is formed in such a N/F contact. In this model, a ferromagnetic cluster at the N/F interface plays the role of the free layer in this SV. This cluster exhibits a larger coercive field than the bulk of the ferromagnetic film, presumably due to its nanoscale nature. Finally, we construct a magnetic state diagram of the surface SV as a function of I and H.

cond-mat.mes-hall

cond-mat

Low Temperature Physics/Fizika Nizkikh Temperatur, 2013, v. 39, No. 3, pp. Current-field diagram of magnetic states of a surface spin valve in a point contact with a single ferromagnetic film I.K. Yanson1, O.P. Balkashin1, V.V. Fisun1, Yu.I. Yanson2, and Yu.G. Naidyuk1 1B. Verkin Institute for Low Temperature Physics and Engineering of the National Academy of Sciences of Ukraine 47 Lenin Ave., Kharkov 61103, Ukraine E-mail: naidyuk@ilt.kharkov.ua 2Kamerlingh Onnes Laboratory, Leiden University, Niels Bohrweg 2, 2333 CA Leiden, The Netherlands Received December 21, 2012 We present a study of the influence of an external magnetic field H and an electric current I on the spin-valve (SV) effect between a ferromagnetic thin film (F) and a sharp tip of a nonmagnetic metal (N). To explain our ob- servations, we propose a model of a local surface SV which is formed in such a N/F contact. In this model, a fer- romagnetic cluster at the N/F interface plays the role of the free layer in this SV. This cluster exhibits a larger coercive field than the bulk of the ferromagnetic film, presumably due to its nanoscale nature. Finally, we con- struct a magnetic state diagram of the surface SV as a function of I and H. PACS: 72.25.–b Spin polarized transport; 73.40.Jn Metal-to-metal contacts; 75.75.+a Magnetic properties of nanostructures; 85.75.–d Magnetoelectronics; spintronics. Keywords: point contacts, spin valve, spin transfer torque, magnetic phase diagram. 1. Introduction Spin-valve (SV) effect on the electric conductance of F1/N/F2-type nanopillars, where F1,2 represent ferromag- netic layers and N is a nonmagnetic spacer layer, generated great interest both from the fundamental point of view and from its application perspective in spintronic devices [1]. The fundamental interest lies in the spin-transfer torque (STT) in a SV by a spin-polarized current that passes through it [2]. This effect could be used to control the magnetic state of a SV [3]: the passage of a spin-polarized current can change the orientation of the magnetization М of the layers F1 and F2 relative to each other, depending on the direction of the current flow. Hence, if one would fix the magnetization of the layer F1 (fixed layer), the magnet- ization of the other layer F2 (free layer) could be set either parallel (P) or antiparallel (AP) to F1 depending on the direction of an electric current that flows through the sys- tem. These two states, i.e., the P and the AP states, lead to a different conductivity of a SV due to the giant magnetoresistance effect [4]. This results in a hysteresis in the resistance of a SV with a transition between the low resistance state (P orientation) and a high resistance state (AP orientation) during a bipolar current sweep. In addi- tion, the direction of M1 and M2 can also be controlled by an external magnetic field. If the layers F1 and F2 have different coercivities, the P and AP orientations of M1,2 can be realized, forming two states in the magnetoresis- tance R(H) of a SV. These states produce hysteresis loops (meanders) in the R(H) dependence within the range of the coercive fields of the fixed and the free layers. In particular, one is interested in the behavior of a SV structure both in an external magnetic field and under elec- tric current. This situation has been studied in pillar-type SV structures based on Co and permalloy (Ni80Fe20), for which current-field (I–H) diagrams of the magnetic state of a SV were obtained [5–7]. The effects of a flow of electrons that creates a spin transfer torque, and a magnetic field H can mutually enhance or suppress each other, leading to an asymmetric I–H phase diagram. Additionally, as proposed in Ref. 6, other factors, such as the self-Oersted-field of the current, generation of non-equilibrium magnons, and ther- mal effects due to heating of the structure by a high current density, may influence the shape of the I–H phase diagram. Spin-valve effects that are similar to those found in the li- thographically-made conducting F1/N/F2 nanopillars were © I.K. Yanson, O.P. Balkashin, V.V. Fisun, Yu.I. Yanson, and Yu.G. Naidyuk, 2013 I.K. Yanson, O.P. Balkashin, V.V. Fisun, Yu.I. Yanson, and Yu.G. Naidyuk also observed in point contacts between a single ferromag- netic thin film and a nonmagnetic metallic tip [8–10]. Ac- cording to Refs. 8, 9, the role of the free layer F2 in such a system is taken by a ferromagnetic domain that is formed at the point contact. However, a study of this effect in Co thin films of varying thickness (3–100 nm) and point con- tacts with diameters from several tens down to few na- nometers raised doubts about the formation of “conven- tional” ferromagnetic domains as an explanation of this effect [10]. Instead, a model of a surface SV (SSV) has been proposed, where a ferromagnetic layer of a few atoms thick at the surface, which has a weakened magnetic bond to the bulk of the film, plays the role of the free layer F2. Recently it was shown that other effects that are charac- teristic of the F1/N/F2 pillar structures can also be observed on a single ferromagnetic film [11–13]. These effects in- clude the dynamic SV effect [11], formation of spin vortex states [12], and exchange bias effect [13]. The size of the point contacts in these studies was at least an order of magnitude smaller than the existing lithographically- prepared F1/N/F2 pillars, thus showing that the SV effect is preserved on the nanoscale. In this paper we present our study on the influence of an electric current and an external magnetic field applied sim- ultaneously to a N/F point contact. Along with gaining a deeper understanding of the spin-dependent processes in such structures, we also constructed the I–H phase diagram of the magnetic states of a SV formed in a point contact. We expect this study to contribute to a more adequate model of SV effects observed in point contacts. 2. Experimental details and a model of a point contact Figure 1 shows a schematic representation of a multi- layer thin-film structure that we used in our measurements. A Cu film with a thickness of 100 nm was deposited on a Si substrate by sputtering. This buffer layer of Cu provides a geometry in which the electric current flows nearly per- pendicular to the plane of the thin film layers (CPP geomet- ry). A ferromagnetic Co layer with a thickness of ≤ 100 nm was deposited onto the Cu layer. The Co layer was capped with a layer of either Cu or Au of several nanometers to protect it from oxidation. The surface of this thin film structure was contacted by a sharpened Cu tip using a me- chanical manipulator. This tip acted as a nonmagnetic elec- trode, forming a point contact between the tip and the thin film structure. The formation mechanism of the second (free) layer F2 in this system is not yet clear. According to Ref. 10, appar- ently a very thin ferromagnetic layer F2 is formed at the interface between an F-film and a nonmagnetic metallic tip N. The formation of this interface layer may be due to the inter-diffusion between the F-film and the N-tip. The mag- netic properties of this interface layer F2 differ from those of the bulk F-film due to a large density of nonmagnetic Fig. 1. Schematic representation of the sample layout with a buff- er Cu layer, a single Co film, and a Cu or Au capping layer that is contacted by a Cu needle. Current and voltage leads that are at- tached to the system as shown. Small oval at the surface of the Co film represents an interfacial ferromagnetic cluster. lattice defects and magnetic spin-lattice defects in the in- ter-diffusion region. Surprisingly, this layer consists of a single ferromagnetic domain, as indicated by the measured dV dI H nearly-rectangular magnetoresistance hysteresis / ) ( loops [10]. At the boundary between the interfacial F2- layer and the bulk F-film a spin-glass layer can be formed. This spin-glass layer plays the role of a spacer, which ena- bles the magnetization of the F2-layer to rotate freely with respect to the bulk F-layer. Please note that the spacer layer in this case is not a nonmagnetic metal film with relatively weak magnetic scattering, such as in conventional nano- pillar structures, but contains strongly interacting chaoti- cally directed spins of the F-metal. Thus, to observe the SV effect in this system the thickness of this spin-glass layer must be very small, namely, of the atomic scale. We would like to note that it is relatively unlikely that upon contact the inter-diffusion between the N-tip and the F-film would lead to the formation of a “free” ferromag- netic cluster surrounded by an atomically-thick spin-glass domain wall. Indeed, we observe the SV effects only in about 10% of the contacts formed between the N-tip and the F-film at random locations on the surface. As men- tioned in Ref. 14, formation of magnetic clusters is possi- ble in the interfacial region of alternately deposited Co and Cu films due to the immiscibility of the two components. A similar structure with implanted 5 nm Co clusters in a nonmagnetic layer on top of a Co film was investigated in [15]. The behavior of this special SV structure, in which the Co clusters play the role of F2, is similar to what we obtain using our system. It is likely that similar clusters can be created during the formation of a mechanical contact between a tip and a thin film even at low temperatures due 2 Low Temperature Physics/Fizika Nizkikh Temperatur, 2013, v. 39, No. 3 Current-field diagram of magnetic states of a surface spin valve in a point contact with a single ferromagnetic film to metal yielding at high mechanical stress. Furthermore, since the formation of an electric contact is monitored by the onset on an electric current under an applied voltage of about 5 V, an electric discharge may accompany the con- tact formation. This discharge may cause “instant” melting and recrystallization of the metal in the vicinity of the con- tact, thus promoting the cluster formation. This cluster should not be larger than the size of the contact in order to observe the STT effect. If the cluster is too large, then the STT effect would be too weak to change its magnetization and no hysteresis in the electric re- sistance would be observed. This cluster is most probably formed at the surface, since we observed the SV effects in ferromagnetic layers of only 3 nm thick and in contacts with the diameter of only 2 nm [10]. The state of such a nanostructured magnetic system differs significantly from the homogeneous state in the bulk of a ferromagnetic film. Hence, the domain wall around the cluster that is pinned at the structural inhomogeneity can be very thin, according to both the theoretical estimations [18] and experimental measurements [19,20]. The main difference of this cluster model from the “domain” model in Ref. 9 is that the size of the cluster may not exceed the size of the contact, which is of the order of several nanometers to several tens of na- nometers. Since the size of the ferromagnetic cluster should be comparable to the size of the point contact (~10 nm), the following sequential configuration of the layers is possible: bulk polycrystalline ferromagnetic film F1, transitional atomically-thick spin-glass layer, and a single-domain cluster F2. This ferromagnetic cluster has to be located precisely at the point contact, which explains the low prob- ability (≤ 10%) of obtaining a F1/N/F2 structure by this method. Yet the occasional observation of the SV effect in dV dI V R V= the curves and the absence of hysteresis ( / ) ) ( dV dI H R H= loops in the magnetoresistance curves / ( ) ( ) and vice versa can be explained within this model. The absence of the latter, occurring much more often than the former, can be explained if during the acquisition of the R(H) curves the external magnetic field aligns the magnet- ization vectors of many different clusters (if these are pre- sent) up to saturation. However, the total spin torque trans- fer of the conductance electrons can be insufficient for the change of its magnetization direction if the size of the clus- ter is much larger than the diameter of the point contact. Hence, there would be no hysteresis in the R(V) depend- ence. In a less common case, if the coercivities of the clus- ter and the bulk ferromagnetic layer are identical or the magnetic bond between the two is strong, the change of the magnetization direction of the cluster and the bulk layer due to an external magnetic field would proceed simulta- neously. On the other hand, due to the highest current den- sity at the point contact and, accordingly, at the cluster, the STT can lead to the change of the cluster magnetization without influencing the magnetization of the bulk layer. The STT effect of the current on the bulk layer would be insignificant due to a substantial difference in the geomet- rical extension of the point contact and the layer, which results in a rapid decrease of the current density due to the spread of the current flow in the bulk of the film. In this case only a hysteresis in the R(V) dependence would be observed. 3. Experimental results and discussion Figure 2 shows a dependence of the differential re- dV dI V R V sistance of a point contact as a func- / ( ) ( ) = tion of the applied dc voltage V. Negative voltage (current) corresponds to a flow of electrons from the tip to the film. One can see a hysteresis of the resistance, which is caused by a change of the orientation of magnetization of the in- terfacial cluster (microdomain) relative to the bulk Co lay- er due to the STT by the electron current. The behavior of the resistance at large negative (positive) bias voltages of the order of 50–60 mV can be explained by the excitation of the magnetization vector precession in either the interfa- cial domain or the bulk of the film, see [21,22]. The change of the differential resistance ΔR in Fig. 2 constitutes 0.8%, which correlates well with the 1% height of the measured magnetoresistance curve, see the inset in Fig. 2. The slightly smaller value of ΔR in the dV dI V / ( ) dependence can be attributed to an inhomogeneously dis- tributed current density in the contact due to the lateral spread of the current. Thus, the STT effect is not sufficient for the switch of the magnetization direction in the periph- eral area of the contact. The size of the contact can be estimated by using the well-known Wexler’s formula [23]: dV dI V of a Fig. 2. Dependence of the differential resistance ) ( / Co–Cu point contact R0 = 6.4 Ω in zero external magnetic field acquired during a bipolar current sweep. The inset shows the magnetoresistance of the same contact as a function of the ap- plied magnetic field at zero bias voltage. Low Temperature Physics/Fizika Nizkikh Temperatur, 2013, v. 39, No. 3 3 I.K. Yanson, O.P. Balkashin, V.V. Fisun, Yu.I. Yanson, and Yu.G. Naidyuk R W (cid:17) + , (1) ρ ρ l 16 π 2 d 2 3 d where d is the contact diameter, ρ is the resistivity of the film material, and l is the electronic mean free path. For this parti- cular contact with the resistance of R = 6.4 Ω we obtain d = = 20 nm. This leads to a current density of more than 2·109 A·cm–2 at a bias of –60 mV, at which the transition to dV dI V curve occurs. In this esti- the upper branch of the / ( ) mation we used the lower limit of the resistivity ρ = 10 μΩ·cm for Co thin films and a value of ρl = 8.5·10–12 μΩ·cm2 (calcu- lated for a charge carrier density of 5.8·1022 cm–3 [24]). In the calculation we neglected the resistance of the Cu electrode, which is much smaller than that of the Co film. 4. Magnetization reversal cycle of a surface spin valve In the following we consider the magnetization reversal cycle of a point contact that exhibits the SV effect at dif- ferent values of the electric current at positive bias voltag- es, see Fig. 3. A decrease of the magnetic field from its maximum value of ~ 4 kOe does not result in the change of the mutual orientation of magnetizations of the bulk layer and the surface cluster, i.e., M1 and M2, respectively. The- se remain parallel to the external field. The resistance of the point contact remains at its minimal value. Even at Н < 0 coercive force keeps the positive orientation of M1 and M2 until the magnetic field reaches a value of about –0.2 kOe. At this field the magnetization of one of the layers flips and becomes parallel to the external field. Thus, the spin valve becomes “closed”, i.e., its resistance becomes maximal, since M1 and M2 are aligned in opposite direc- tions. Based on Refs. 13, 14, we suggest that the magneti- zation of the bulk film M1 is flipped at lower absolute val- ues of the external magnetic field than the magnetization of the surface cluster M2 due to a larger coercivity of the lat- ter. This suggestion is also supported by the absence of the influence of the bias current on the position of the slopes around H = 0 of the R(H) curves, see Figs. 3 and 4. If the magnetic field is swept further along the negative - ≈−3 kOe the magnetization of the surface clus- axis, at H2 Fig. 3. Magnetization reversal cycles of the same contact as in Fig. 2 at different positive bias voltages (currents). At a positive bias the electrons flow from the Co film into the Cu tip. At I = 0 mA several curves are plotted over each other. Additionally, at this bias voltage a lesser hysteresis loop, which corresponds to the magnetization reversal of the bulk film only, was acquired during a magnetic field sweep in the region of –0.5 kOe < Н < +0.5 kOe. The coercive field, which is the half-width of the small hysteresis loop, equals to about 0.17 kOe for the bulk Co layer. At 1.56 mA bias current two magnetization cycles are shown, and at 4.69, 6.25, and 7.8 mA single cycles are presented. The curves at non-zero biases are shif- ted along the vertical axis by 0.1 Ω for clarity. T = 4.2 K. Fig. 4. Magnetization reversal cycles of the same contact as in Fig. 2 at different values of negative bias voltage (current). The R(H) loops are observed at bias currents lower than I = −7.8 mA. Up to I = = – 6.25 mA the shape of the magnetization cycle curve is almost identical to the one acquired at zero bias. However, at I = – 6.25 mA it is considerably different from the similar cycle at a positive bias of I = +6.25 mA in Fig. 3. Jumps in the R(H) dependence are ob- served at a bias of I = −7.8 mA, which are caused by unstable mag- netic states. The arrows next to the curves show the field sweep direction. s represents the starting point and f represents the ending point of the sweep in the lower curve. Т = 4.2 K. 4 Low Temperature Physics/Fizika Nizkikh Temperatur, 2013, v. 39, No. 3 Current-field diagram of magnetic states of a surface spin valve in a point contact with a single ferromagnetic film ter is flipped. Thus, the magnetization vectors M1 and M2 become parallel to each other and the resistance is reduced by ΔR. Please note that the resistance decreases by the same value of ΔR as after the magnetization reversal of the bulk layer F1 around H = 0. Usually, the resistance change ΔR, which is defined as ΔR = (RAP – RP)/RP, equals to several percent, unless a multidomain structure inside the contact leads to the reduction of this value. Here RAP equals to the resistance of a point contact with the anti- parallel magnetizations M1 and M2 and RP is the contact resistance at the parallel alignment of the magnetizations. The magnetization reversal of the system during a magnet- ic field sweep in the opposite direction occurs in a similar fashion. Thus, a complete magnetization reversal cycle of the R(H) dependence consists of two symmetrically- located loops of magnetoresistance. Magnetization cycle of the bulk layer F1 is not sensitive to the changes of the polarity and the amplitude of the bias current, whereas the behavior of the surface cluster during consecutive magnetization cycles is much more susceptible to such changes, see Fig. 3. To verify that, Fig. 4 shows magnetization reversal cycles of a contact with the SV ef- fect at negative values of the bias current. 5. Simultaneous influence of electric current and external magnetic field on the magnetoresistance of a point contact with the SV effect In the following we analyze the simultaneous influence of the electric current and the external magnetic field on the magnetization direction of the interfacial single-domain cluster and, correspondingly, on the electric resistance of the whole N/F structure. A flow of spin-polarized current through the contact is accompanied by the STT effect, which influences its magnetization in addition to the exter- nal field. By analogy, we will call this influence the “STT field” in the following discussion. One can see from Fig. 3 Н H+ − that leads to the magneti- that the magnetic field 2( ) 2 zation flipping of the cluster at zero bias is slightly less than ±3 kOe. An increase in the bias current is accompa- nied by a shift of the switching field H2 to lower absolute values. This is the result of the cumulative action of the external and the STT fields. The latter reduces the value of 2Н ± required to flip M2, thus narrowing the external field the R(H)-loop. At a bias current of +7.8 mA, the R(H)-loop disappears completely in this contact, see Fig. 3. Please note that the reversal of the magnetic field and the corre- sponding change of the magnetization direction of the bulk film lead to a change of the direction of the STT field. Hence, the R(H) curves are virtually symmetrical relative 2H − and 2H + are to H = 0, i.e., the absolute values of equal. Similar influence of an electric current on the width of the R(H)-loops has been also observed in F1/N/F2 pillar structures based on Co [5]. We conclude that a positive current leads to a reduction 2Н ± . The electron flow that passes through the bulk Co of layer becomes spin-polarized. Upon the entry of this flow into the surface cluster, an additional STT field is created, which results to a reduction of the width of the R(H)-loops. On the other hand, if the current is negative, it helps to maintain the antiparallel alignment of M1 and M2, since the effective STT field now acts against the external mag- netic field. As a result, one should observe an increase of 2Н ± of the surface cluster and a corre- the switching field sponding widening of the R(H)-loops. Such an effect was observed in F1/N/F2 structures based on permalloy (Ni84Fe16) [6]. However, our results show that at negative 2Н ± does not increase. Instead, bias (current) the value of it decreases with an increasing negative bias. Nevertheless, if we compare the R(H)-loops acquired at the same positive and negative currents (see e.g., curves at 6.25 mA in Figs. 3 and 4), we see that at negative currents the R(H)-loops are broader. Thus, we can conclude that there the STT field counteracts the external magnetic field. Figure 5 shows the resulting current-field phase dia- gram of stable magnetic states of a surface SV. This dia- 2H − at 2H + and gram is based on the measured values of different values of bias voltage. Bias dependence of the coercive field Hc of the surface cluster, calculated as H+ − cH H is also shown. ( ) / 2 = − 2 2 2Н ± in the I–H diagram The absence of an increase of (Fig. 5) at negative currents can be linked to a spontaneous generation of magnons during the energetic relaxation of the conductance electrons, which is described theoretically Fig. 5. Current-field phase diagram of magnetic states of a sur- face SV. The diagram is based on the measured values of H+ (triangles pointing upwards) and H– (triangles pointing down- wards) from the magnetoresistance curves in Figs. 3 and 4. Long arrows represent directions in the phase space during the acquisi- tion of an R(H) dependence. Stars represent the behavior of the H+ − cH H ( ) / 2 of the surface domain (clus- coercive field = − 2 2 ter). Filled symbols correspond to the switching fields of the bulk layer and hollow symbols represent the switching fields of the surface cluster. Low Temperature Physics/Fizika Nizkikh Temperatur, 2013, v. 39, No. 3 5 I.K. Yanson, O.P. Balkashin, V.V. Fisun, Yu.I. Yanson, and Yu.G. Naidyuk in [7]. In that paper it was shown that an increase of nega- 2Н ± in the tive bias should lead to a monotonic increase of absence of this relaxation mechanism. A slight decrease of 2Н ± observed at negative bias currents can be attributed to the influence of the magnetic (Oersted) field of the electric current and a possible heating of the contact at high DC bias. Similar effect was observed in spin valves with an interlayer of a CuPt alloy, which resulted in a current flow without spin-polarization [6]. Finally, we would like to discuss the possible dynamic states of a SV. It is known that, in addition to the P–AP switching of the magnetic states of a SV, the STT effect can lead to an excitation of various vibration modes of the magnetization precession. These modes include a coherent rotation of a local magnetic moment as well as a non- coherent generation of magnons. At large currents the sta- tionary precession of the magnetization leads to the for- mation of singularities, such a peaks and valleys, on the dV dI V dependence (see Fig. 2) due to such a dynamic / ) ( behavior of the magnetization. This results in the formation of a dynamic states region on the I–H phase diagram. In dV dI H curves in our measurements we have acquired / ( ) the bias region of V < ±50 mV (I < ±7 mA, Fig. 2), i.e., we have limited ourselves to the hysteretic region of the dV dI V dependence where no such singularities occur. / ( ) 6. Conclusions The main results of this work can be summarized as fol- lows. 1. We have proposed a model of a local surface spin valve that is formed in a N/F point contact. In this model, the bulk ferromagnetic layer acts as a spin polarizer of the electric current, which then passes through the magnetic cluster at the surface of the layer. Magnetization switching in the bulk ferromagnetic layer occurs at weaker magnetic fields than in the surface cluster. Possibly, the anisotropy of the surface magnetic cluster is larger than that of the bulk layer due to the atomic size of the cluster and its clus- ter-like nature. This large anisotropy results in a higher coercive field of the cluster. 2. We have constructed an I–H phase diagram of stable magnetic states of a surface spin valve based on a Co–Cu microcontact. The magnetization switching field of the surface cluster decreases until its complete disappearance with an increase of the positive bias current (the spin- polarized electrons flow from the bulk film to the surface) in an in-plane external magnetic field. This effect is due to the parallel alignment of the external and the STT fields to each other leading to the enhancement of the total field. The width of the corresponding hysteresis loops in the magnetoresistance curves decreases symmetrically with an increasing negative bias. In contrast, the width of the R(H)- loops at negative biases is larger than their width at posi- tive bias voltages. This effect is due to a counteraction of the STT field to the external magnetic field. 3. There is a general tendency of the R(H)-loops to be- come narrower during an increase of the absolute value of the bias independently of its polarity. This effect could be due to the reduction of the coercivity of the surface cluster by a magnetic (Oersted) field of the bias current, which can reach ~1.5 kOe for a current of I = 7.8 mA that flows through a contact of 20 nm diameter at V = 50 mV. Addi- tionally, generation of non-equilibrium magnons and ther- mal effects due to a high current density of 109 A/cm2 could lead to the reduction of the R(H)-loop width. 1. S.A. Wolf, D.D. Awschalom, R.A. Buhrman, J.M. Daugh- ton, S. von Molnar, M.L. Roukes, A.Y. Chtchelkanova, and D.M. Treger, Science 294, 1488 (2001). 2. B. Dieny, V.S. Speriosu, S. Metin, S.S.P. Parkin, B.A. Gur- ney, P. Baumgart, and D.R. Wilhoit, J. Appl. Phys. 69, 4774 (1991). 3. D.C. Ralph and M.D. Stiles, J. Magn. Magn. Mater. 320, 1190 (2008). 4. M. Baibich, J.M. Broto, A. Fert, F. Nguyen Van Dau, F. Pet- roff, P. Etienne, G. Creuzet, A. Friederich, and J. Chazelas, Phys. Rev. Lett. 61, 2472 (1988). 5. S. Urazhdin, H. Kurt, W.P. Pratt, Jr, and J. Bass, Appl. Phys. Lett. 83, 114 (2003). 6. S. Urazhdin, W.P. Pratt, and J. Bass, J. Magn. Magn. Mater. 282, 264 (2004). 7. S. Urazhdin, Phys. Rev. B 69, 134430 (2004). 8. Y. Ji, C.L. Chien, and M.D. Stiles, Phys. Rev. Lett. 90, 106601 (2003). 9. T.Y. Chen, Y. Ji, C.L. Chien, and M.D. Stiles, Phys. Rev. Lett. 93, 026601 (2004). 10. I.K. Yanson, Y.G. Naidyuk, V.V. Fisun, A. Konovalenko, O.P. Balkashin, L.Y. Triputen, and V. Korenivski, Nano Lett. 7, 927 (2007). 11. O.P. Balkashin, V.V. Fisun, I.K. Yanson, L.Yu. Triputen, А. Konovalenko, and V. Korenivski, Phys. Rev. B 79, 092419 (2009). 12. I.K. Yanson, V.V. Fisun, Yu.G. Naidyuk, O.P. Balkashin, L.Yu. Triputen, A. Konovalenko, and V. Korenivski, J. Phys.: Condens. Matter. 21, 355004 (2009). 13. I.K. Yanson, Yu.G. Naidyuk, O.P. Balkashin, V.V. Fisun, L.Yu. Triputen, S. Andersson, V. Korenivski, Y.I. Yanson, and H. Zabel, IEEE Transactions on Magnetics 46, 2094 (2010). 14. M.M.H. Willekens, Th.G.S.M. Rijks, H.J.M. Swagten, and W.J.M. de Jonge, J. Appl. Phys. 78, 7202 (1995). 15. X.J. Wang, H. Zou, and Y. Ji, Appl. Phys. Lett. 93, 162501 (2008). 16. S.J. Steinmuller, C.A.F. Vaz, V. Ström, C. Moutafis, C.M. Gürtler, M. Kläui, J.A.C. Bland, and Z. Cui, J. Appl. Phys. 101, 09D113 (2007). 17. R.A. Khan and A.S. Bhatti, J. Magn. Magn. Mater. 323, 340 (2011). 6 Low Temperature Physics/Fizika Nizkikh Temperatur, 2013, v. 39, No. 3 Current-field diagram of magnetic states of a surface spin valve in a point contact with a single ferromagnetic film 18. P. Bruno, Phys. Rev. Lett. 83, 2425 (1999). 19. O. Pietzsch, A. Kubetzka, M. Bode, and R. Wiesendanger, Phys. Rev. Lett. 84, 5213 (2000). 20. M. Kläui, C.A.F. Vaz, J. Rothman, J.A.C. Bland, W. Werns- dorfer, G. Faini, and E. Cambril, Phys. Rev. Lett. 90, 097202-2 (2003). 21. S. Urazhdin, W.L. Lim, and A. Higgins, Phys. Rev. B 80, 144411 (2009). 22. S.I. Kiselev, J.C. Sankey, I.N. Krivorotov, N.C. Emley, R.J. Schoelkopf, R.A. Buhrman, and D.C. Ralph, Nature 425, 380 (2003). 23. G. Wexler, Proc. Phys. Soc. 89, 927 (1966). 24. W. Gil, D. Görlitz, M. Horisberger, and J. Kötzler, Phys. Rev. B 72, 134401 (2005). Low Temperature Physics/Fizika Nizkikh Temperatur, 2013, v. 39, No. 3 7 

1508.05693

1

1508

Spin Dynamics Simulation of the Magneto-Electric Effect in a Composite Multiferroic Chain

A composite multiferroic chain with an interfacial linear magneto-electric coupling is used to study the magnetic and electric responses to an external magnetic or electric field. The simulation uses continuous spin dynamics through the Landau-Lifshitz-Gilbert equations of the magnetic spin and the electric pseudo-spin. The results demonstrate an accurate description of the distribution of the magnetisation and polarisation are induced by applied electric and magnetic field, respectively.

cond-mat.mes-hall

cond-mat

39th ANNUAL CONDENSED MATTER AND MATERIALS MEETING, 3 - 6 Feb. 2015. Wagga Wagga, NSW, Australia Spin Dynamics Simulation of the Magneto-Electric Effect in a Composite Multiferroic Chain Zidong Wang (王子东) and Malcolm J. Grimson Department of Physics, University of Auckland, Auckland 1010, New Zealand. A composite multiferroic chain with an interfacial linear magneto-electric coupling is used to study the magnetic and electric responses to an external magnetic or electric field. The simulation uses continuous spin dynamics through the Landau-Lifshitz-Gilbert equations of the magnetic spin and the electric pseudo-spin. The results demonstrate an accurate description of the distribution of the magnetisation and polarisation are induced by applied electric and magnetic field, respectively. 1. Introduction Multiferroic materials, i.e., materials exhibit more than one ferroic (magnetic, electric, or elastic) state [1], particularly the ferroelectric (FE) and the ferromagnetic (FM) composited order is currently received intensive investigation [1-3]. In this present work, we developed the theoretical spin dynamic simulation of a one-dimensional multiferroic composite chain, which coupled by FM and FE orders. The external field driven dynamics of the magnetisation and the electric polarisation of a FM/FE system that shows a magneto-electric (ME) coupling at the interface [2,3]. The ME coupling can induce the ME effect, which is the phenomenon of inducing electric polarisation (magnetisation) by applying a magnetic (electric) field. The ME effect in composite multiferroic materials results by the combination of magnetostrictive and piezoelectric effects [4]. This can be written in a simple form, 𝑀𝐸 = 𝑒𝑙𝑒𝑐𝑡𝑟𝑖𝑐 𝑚𝑒𝑐ℎ𝑎𝑛𝑖𝑐𝑎𝑙 × 𝑚𝑒𝑐ℎ𝑎𝑛𝑖𝑐𝑎𝑙 𝑚𝑎𝑔𝑛𝑒𝑡𝑖𝑐 𝑜𝑟 𝑀𝐸 = 𝑚𝑎𝑔𝑛𝑒𝑡𝑖𝑐 𝑚𝑒𝑐ℎ𝑎𝑛𝑖𝑐𝑎𝑙 × 𝑚𝑒𝑐ℎ𝑎𝑛𝑖𝑐𝑎𝑙 𝑒𝑙𝑒𝑐𝑡𝑟𝑖𝑐 (1) For this purpose we considered a two-component multiferroic chain consisting of 𝑁𝑆 localized magnetic moments and 𝑁𝑃 polarisation sites. The schematic view is in Fig. 1. Ferromagnet Ferroelectric Fig. 1. (colour online) A schematic of the composite multiferroic sample built of FM and FE chains. The red and blue arrows indicate the magnetic spins and the electric pseudo-spins [5,6], respectively. The interface between FM/FE chains is indicated by a yellow line. The FM part of the chain is a normal metal (e.g. iron, cobalt or nickel), whereas the FE part is, for example, BaTiO3 or PbTiO3 [3]. The total energy of the composite multiferroic system in general one-dimensional consists of three parts, (2) where 𝐻𝑆is the conventional Heisenberg Hamiltonian describes the FM part of the multiferroic chain with 𝑁𝑆 magnetic spins, 𝐻 = 𝐻𝑆 + 𝐻𝑃 + 𝐻𝑆𝑃 (3) where 𝐽𝑆 is the nearest neighbour exchange coupling and 𝛫𝑆 is the z-directional uniaxial anisotropy constant. The magnetisation vector 𝑆𝑖⃗⃗⃗ = (𝑆𝑖 𝑧) at site 𝑖 = 1, … , 𝑁𝑆, with the 𝑦, 𝑆𝑖 𝑥, 𝑆𝑖 (−𝐽𝑆𝑆𝑖⃗⃗⃗ ∙ 𝑆𝑗⃗⃗⃗ − 𝛫𝑆(𝑆𝑖 𝑧)2) 𝑧 − 𝐵(𝑡) ∑ 𝑆𝑖 𝑁𝑆 𝑖=1 𝐻𝑆 = ∑ 𝑁𝑆 <𝑖,𝑗> 39th ANNUAL CONDENSED MATTER AND MATERIALS MEETING, 3 - 6 Feb. 2015. Wagga Wagga, NSW, Australia normalization 𝑆𝑖⃗⃗⃗ = 1, and 𝑺𝒋 = 𝑺𝒊−𝟏 + 𝑺𝒊+𝟏 denotes the sum of neighbour spins. The last term in (3) shows the Zeeman energy induced by the magnetic spins and an external magnetic field, 𝐵(𝑡), this field is applied alone the z-axis and has the time dependent form. The classical Heisenberg Hamiltonian describes the FE part of the multiferroic system with 𝑁𝑃 pseudo-spins that represents the interacting dipoles, 𝑁𝑃 <𝑘,𝑗> 𝑧)2) 𝑥, 𝑃𝑘 𝑦, 𝑃𝑘 − 𝐸(𝑡) ∑ (−𝐽𝑃𝑃𝑘⃗⃗⃗⃗ ∙ 𝑃𝑗⃗⃗ − Κ𝑃(𝑃𝑘 𝐻𝑃 = ∑ (4) where 𝑃𝑘⃗⃗⃗⃗ = (𝑃𝑘 𝑧) is a component of a pseudo-spin vector at site 𝑘 = 1, … , 𝑁𝑃, denotes as the directional electric dipole moment [5,6], the amplitude of the pseudo-spin vector is set to be𝑃𝑘⃗⃗⃗⃗ = 1. 𝐽𝑃 is the nearest exchange interaction coupling and 𝛫𝑃 is the z-directional uniaxial anisotropy constant in the FE part. The system is subject to an external electric driving field, 𝐸(𝑡), that couples to the pseudo-spins in the system. The interface effects between the magnetic spin and the electric dipole systems are described by the dipole-spin interaction Hamiltonian (5), with a linear ME coupling, 𝑔 [3]. (5) 𝑁𝑃 𝑘=1 𝑧 𝑃𝑘 𝐻𝑆𝑃 = −𝑔(𝑺𝑁𝑆 ∙ 𝑷1) 2. Spin Dynamics Method To describe the magnetisation dynamics in the FM, a dynamic equation of spins named Landau-Lifshitz-Gilbert (LLG) equation, has been used at the atomic level [2], where 𝛾𝐹𝑀 ′ = 𝛾 1+𝛼𝐹𝑀 𝜕𝑺 𝜕𝑡 = −𝛾𝐹𝑀 ′ [𝑺 × 𝑯𝑆 𝑒𝑓𝑓(𝑡)] − 𝜆𝐹𝑀 [𝑺 × [𝑺 × 𝑯𝑆 𝑒𝑓𝑓(𝑡)]] (6) 2 , 𝛾 is the gyromagnetic ratio and 𝛼𝐹𝑀 is the dimensionless damping factor. 𝜆𝐹𝑀 = 𝛾𝛼𝐹𝑀 1+𝛼𝐹𝑀 2 denotes Gilbert damping term. 𝑯𝑆𝑖 𝑒𝑓𝑓 = 𝜕𝐻𝑆 𝜕𝑺𝑖 is the effective magnetic field, which is a derivative of the system Hamiltonian with respect to the magnetisation, acting on each magnetic spin. In the FE part, we used a simple pseudo-spin model [5,6] to describe the locations of the electric dipole. Since an electric dipole is a separation of positive and negative charges, a measure of this separation gives the magnitude of the electric dipole moment, it is a scalar. In ′ = 0) and the the spin dynamics system, no precession of the pseudo-spins is expected (i.e., 𝛾𝐹𝐸 polarisation dynamics are described [7], ∂𝑷 ∂t = −𝜆𝐹𝐸 [𝑷 × [𝑷 × 𝑯𝑃 𝑒𝑓𝑓(𝑡)]] (7) where the 𝜆𝐹𝐸 is the intrinsic damping parameter for the electric pseudo-spins and the local effective electric field in each atomic plane is defined as 𝑯𝑃𝑘 𝑒𝑓𝑓 = . 𝜕𝐻𝑃 𝜕𝑃𝑘 3. Numerical Results To demonstrate the response of the FM/FE chain to a driving field, we used the composite multiferroic model as shown in Fig. 1, with 50 spins on each side (e.g., 𝑁𝑆 = 𝑁𝑃 = 50), the nearest exchange interaction coupling 𝐽𝑆 = 𝐽𝑃 = 100, the z-directional uniaxial anisotropic ′ = constant 𝛫𝑆 = 𝛫𝑃 = 0.01, the ME coupling 𝑔 = 1, the normalised gyromagnetic ratio 𝛾𝐹𝑀 1, and the damping factor 𝜆𝐹𝑀 = 𝜆𝐹𝐸 = 0.1; the applied driving field is dynamic sinusoidal type, either magnetic 𝐵(𝑡) or electric 𝐸(𝑡) field, with a magnitude of 10. The numerical results obtained by a fourth order Runge-Kutta method and the magnetic and the electric responses were presented in Fig. 2. Top two panels show the FM/FE chain driven by a magnetic field; the mean magnetisations in the x-, y- and z-components are gained in Fig. 2(a), and Fig. 2(b) shows the mean magnitude of polarisation in each component. By using the same method, Fig. 2(c) and (d) show the magnetic and the electric responses under an electric driving field. In general, the electric driving field gives a quicker response than the magnetic driving field. 39th ANNUAL CONDENSED MATTER AND MATERIALS MEETING, 3 - 6 Feb. 2015. Wagga Wagga, NSW, Australia (a) (c) (b) (d) Fig. 2. (colour online) Dynamic magnetic and electric responses in the FM and FE chains, respectively. Panels (a) and (b) show the mean magnetisation (red) in the FM and the mean polarisation (blue) in the FE to an external magnetic field (black dash); (c) and (d) show the similar results as (a) and (b), but driven under an electric field(black dot dash). Green and yellow curves represent the x- and y-components, respectively. (a) (b) (c) (d) Fig. 3. (colour online) The z-component hysteresis loops and the spin waves at seven specific moments. The mean magnetisation and polarisation are indicated by red and blue, respectively. Each type of symbol represents a different time. The yellow line in each panel (c) and (d) shows the interface between FM/FE chains. In order to verify the behaviour of response, a closer inspection of the FM/FE chain at seven specific moments is shown in Fig. 3. In Fig. 3(a) and (c), the magnetic (red) and electric (blue) hysteresis loops present the z-component mean responses contained in Fig. 2. The magnetic spins in the FM part are directly driven by the applied magnetic field is presented in left hand side of Fig. 3(b); on the other side, the electric pseudo-spins catch up slowly, given that they are only driven by the interfacial spins via the ME effect. A similar effect is displayed in the system with an electric driving field in Fig. 3(d). In Fig. 3(b) and (d) further show that magnetic spins with full precession have greater flexibility than the electric pseudo-spins without precession. Free boundary conditions at both end of the composite chain are applied in these simulations. 39th ANNUAL CONDENSED MATTER AND MATERIALS MEETING, 3 - 6 Feb. 2015. Wagga Wagga, NSW, Australia (a) (b) (c) (d) Fig. 4. (colour online) Trajectory of the particular spins in a cycle of the magnetic (top panels)/electric (bottom panels) driving field. The magnetic spins, (a) and (c) are from the FM part and the electric pseudo-spins, (b) and (d) are from the FE part. Multiple colours represent different magnitudes alone the z-axis. The spin dynamics method allows us to follow the trajectories of the spins in the FM/FE chain. In Fig. 4(a) and (c), we follow a representative magnetic spin located in the bulk FM over a cycle of the driving field. Also, we show the similar behaviour for one electric pseudo- spin in the bulk FE in Fig. 4(b) and (d). The precession shown by the magnetic spin is remarkably different from the behaviour shown by the electric pseudo-spin. 4. Conclusion In this paper, the ME effect has been demonstrated by the spin dynamic method in a 1-D composite multiferroic chain. This work used the classical Heisenberg model in both FM and FE sections. As proof of concept, the response of the pseudo-spins shows a kind of flipping behaviour with respect to the electric dipole moments. Additionally, the study of the composite multiferroic system can also be done by a Monte Carlo approach. The modelling results are consistent for both methods [2,8]. Acknowledgments The author gratefully acknowledges Zhao BingJin & Wang YuHua for financial support. References [1] Schmid H 1994 Ferroelectrics 162 317. [2] Sukhov A, Jia C L, Horley P P and Berakdar J 2010 J. Phys.: Condens. Matter 22 352201. [3] Chotorlishvili L, Khomeriki R, Sukhov A, Ruffo S and Berakdar J 2013 Phys. Rev. Lett. 111 117202. [4] Nan C W 1994 Phys. Rev. B 50 6082. [5] de Gennes P G 1963 Solid State Commun. 1 132. [6] Elliott R J and Young A P 1974 Ferroelectrics 7 23. [7] Liao J W, Atxitia U, Evans R F L, Chantrell R W and Lai C H 2014 Phys. Rev. B 90 174415. [8] Wang Z and Grimson M J 2015 Eur. Phys. J. Appl. Phys. 70, 30303. 39th ANNUAL CONDENSED MATTER AND MATERIALS MEETING, 3 - 6 Feb. 2015. Wagga Wagga, NSW, Australia Declaration This paper is published in the Australian Institute of Physics (The AIP) for 39th ANNUAL CONDENSED MATTER AND MATERIALS MEETING, Wagga Wagga, NSW, Australia. Link: meeting http://www.aip.org.au/info/?q=content/39th-annual-condensed-matter-and-materials- 

1203.0331

1

1203

Unitarity of scattering and edge spin accumulation

We consider a 2D ballistic and quasi-ballistic structures with spin-orbit-related splitting of the electron spectrum. The ballistic region is attached to the leads with a voltage applied between them. We calculate the edge spin density which arises in the presence of a charge current through the structure. We solve the problem with the use of the method of scattering states and clarify the important role of the unitarity of scattering. In the case of a straight boundary it leads to exact cancellation of long-wavelength oscillations of the spin density. In general, however, the smooth spin oscillations with the spin precession length may arise, as it happens, e.g., for the wiggly boundary. Moreover, we show that the result crucially depends on the form of the spin-orbit Hamiltonian.

cond-mat.mes-hall

cond-mat

Unitarity of scattering and edge spin accumulation 1 Department of Physics, University at Buffalo, SUNY, Buffalo, NY 14260-1500 and Alexander Khaetskii1 , and Eugene Sukhorukov2 2 Department of Theoretical Physics, University of Geneva, 24 quai Ernest Ansermet, CH-1211, Switzerland (Dated: September 5, 2018) We consider a 2D ballistic and quasi-ballistic structures with spin-orbit-related splitting of the electron spectrum. The ballistic region is attached to the leads with a voltage applied between them. We calculate the edge spin density which arises in the presence of a charge current through the structure. We solve the problem with the use of the method of scattering states and clarify the important role of the unitarity of scattering. In the case of a straight boundary it leads to exact cancellation of long-wavelength oscillations of the spin density. In general, however, the smooth spin oscillations with the spin precession length may arise, as it happens, e.g., for the wiggly boundary. Moreover, we show that the result crucially depends on the form of the spin-orbit Hamiltonian. PACS numbers: 72.25.-b, 73.23.-b, 73.50.Bk Currently, there is a great interest, both experimental and theoretical, to spin currents and spin accumulation in various mesoscopic semiconductor structures [1, 2]. Both phenomena are due to spin-orbit (s-o) coupling and are of great importance for the future of spin electronics. The edge electron spin density accumulation, related to the Mott asymmetry in electron scattering off impurities, has been recently measured [3]. Moreover, the edge spin density in the two-dimensional (2D) hole system, which is due to the intrinsic mechanism [4] of the s-o interaction, has also been observed [5]. It is well known [2], that in the diffusive regime (and when a spin diffusion length is much larger than a mean free path), the spin density ap- pearing near the boundary is entirely determined by the spin flux coming from the bulk. For example, in the dif- fusive regime and in the case of the Rashba Hamiltonian, when the spin current in the bulk is zero, the spin density component perpendicular to the plane is zero everywhere down to the sample boundary [6]. In an opposite case, when the spin precession length is much shorter than the mean free path, the situation is much less investigated. An example of such a system is a mesoscopic structure with s-o-related splitting of the electron spectrum ∆R, in the limit ∆Rτp ≫ 1, where τp is the mean free time. Besides, in the presence of s-o inter- action, the boundary scattering itself is the source of gen- eration of the spin density. It is obvious, that the char- acteristic length near a boundary, where the spin density arises is the spin precession length, Ls = ¯hvF /∆R, with vF being the Fermi velocity. This mechanism of the spin density generation is the subject of our paper. We show that various situations may arise, depending on the form of the s-o Hamiltonians. We start with a 2D system described by the Rashba Hamiltonian in the ballistic limit, where a mean free path is much larger than the sample sizes. The ballistic region is attached to the leads, and a voltage V applied between the leads causes a charge current through the structure, as shown in Fig. 1. Since the electric field is absent inside an ideal ballistic conductor, the edge spin polarization appears not as a result of the acceleration of electrons by an electric field, but rather due to the difference in pop- ulations of left-moving and right-moving electrons. The combined effect of boundary scattering and spin preces- sion leads to oscillations of the edge spin polarization. Note that there is no relation between the spin current in the bulk, which is zero in the considered case, and the edge spin accumulation. The problem of the spin density accumulation in a bal- listic system and for a straight boundary has been con- sidered earlier in Refs. [7,8] with the help of the Green's functions method. Surprisingly, in this case the long- wavelength oscillations of the spin density cancel, and the final result contains only Friedel-like oscillations with the momentum 2kF . This effect may be interpreted as s-o splitting of the Friedel oscillations in the charge den- sity: two charge oscillations corresponding to spin-up and spin-down orientations get shifted with respect to each other in the presence of the s-o interaction. Therefore, strictly speaking, this phenomenon is different from a s-o- related accumulation of the spin density upon boundary scattering. Besides, the method used in Ref. [7] does not allow to understand the reason for the cancellation of long-wavelength oscillations of the spin density. We solve the problem of edge spin accumulation by using scattering theory, with scattering states coming from different leads of the structure and, therefore, hav- ing different occupations. The simplicity of the method allows us to gain an insight into the underlaying physics. We show that it is the unitarity of scattering that leads to the exact cancellation of long-wavelength oscillations of the spin density with the period Ls in the case of a straight boundary. It should be also mentioned that the observed behaviour is closely related to the effective one-dimensional character of scattering, arising from the translational invariance along the boundary. However, (cid:77) 1(cid:77) x + + (cid:32)(cid:72)(cid:80) F (cid:14) eV 2/ (cid:38) j y x (cid:32)(cid:72)(cid:80) F (cid:16) 2/eV FIG. 1: Left: Schematics of the boundary specular scattering in the presence of spin-orbit coupling. Plus and minus modes are shown for the same energy and the same wave vectors along the boundary. Right: Geometry of the system. Voltage V applied to the ideal leads causes a charge current through the ballistic region. the case of a straight boundary appears to be a rather exceptional one. In general, smooth spin oscillations with the spin precession length Ls arise, as it happens, for ex- ample, for the wiggly boundary, or for scattering off a circular impurity in a 2D electron system [9, 10]. This is a consequence of the fact that in higher dimensions the conditions of the unitarity of scattering take a different form, as explained below. In all these situations, the spin 2 density decays as a power law of the distance from the scatterer. Rashba s-o Hamiltonian in the bulk of a ballistic 2D electron system takes the following form H(p) = p2 2m + α 2 ~n[~σ × p], (1) where ~n is the normal to the plane, ~σ are the Pauli ma- trices, and p is the 2D momentum. The solutions of this Hamiltonian corresponding to the helicity values M = ± have the form exp(ipr/¯h)χM (p), where r = x, y. The explicite form of the spinors and their eigenenergies is χ±(ϕ) = 1 √2(cid:18) 1 ∓ieiϕ (cid:19) , ǫM (p) = p2 2m + M 2 αp, with ϕ being the angle between the momentum p and the positive direction of the x-axis. We consider the semi-infinite system and choose the x- axis to be directed perpendicular to the boundary (x = 0) of the 2D system (see Fig. 1). The wave functions, which obey zero boundary conditions at x = 0, are obviously the scattering states, which constitute the complete set of the orthonormal functions. Two scattering states cor- responding to incident plus and minus modes with given wave vector along the boundary and the same energy are + χ+(ϕ)eikx + F −+ χ−(ϕ1)eik1x]; Ψ(0) Ψ(0) + (x, y) = eiky y[χ+(π − ϕ)e−ikx + F + χ−(ϕ1)eik1x]; Ψ(0) (x, y) = eiky y[χ−(π − ϕ1)e−ik1x + F + − − χ+(ϕ)eikx + F − − Ψ(0) − + (0, y) = 0, (0, y) = 0. (2) (3) Here, the wave vectors are defined as follows S acquire the following form: k2 = k2 + − k2 y, k2 1 = k2 − − k2 y, ¯hk± = m(vF ∓ α 2 ), (4) S+ + = F + + , S− − = F − − , S−+ = S+ − = F −+r vx,− vx,+ , (6) where p± = ¯hk± are the momenta at the Fermi energy in the plus and minus modes. The angles ϕ, ϕ1 may be expressed as sin(ϕ) = ky/k+ and sin(ϕ1) = ky/k− (see Fig. 1). From Eqs. (2) and (3), one finds the scattering ampli- tudes F + + and F −+ : F + + = − (eiϕ1 − e−iϕ) (eiϕ1 + eiϕ) ; F −+ = − 2 cos ϕ (eiϕ1 + eiϕ) . (5) where vx,i = ∂ǫi/∂px are the group velocities. For the Rashba model one has vx,−/vx,+ = cos ϕ1/ cos ϕ. The wave functions Eqs. (2) and (3) may now be used to calculate the average z-component of the spin as a function of coordinates: hSz(x)i = Xi=±Z dky (2π)2 dǫ vx,i fF (ǫ, ky) ×h Ψ(0) i (x) Sz Ψ(0) i (x)i, (7) One can check that the amplitudes F − for the − incident minus mode with the same ky and the same en- ergy are obtained from F + + and F −+ by replacing ϕ ↔ ϕ1. Then, the components of the unitary scattering matrix and F + − where fF (ǫ, ky) is the Fermi distribution function, which takes either of two values, fF (ǫ − µ − eV /2) or fF (ǫ − µ + eV /2), depending on the sign of ky. We find, that one may distinguish various contributions to hSz(x)i with Im Im yk yk  k  k   * * 2 2 * * 1 1 * * 2 2 * * 3 3  k  k   ykRe ykRe FIG. 2: The original contour Γ1 along the real axis can be deformed into the part Γ2 going along the imaginary axis, and the part Γ3 going far from the origin. different oscillation periods, which originate from an in- terference of different terms in Eqs. (2) and (3). The smooth part of hSz(x)is, which involves the interference of the outgoing waves [two last terms in Eqs. (2) and (3)], reads: hSz(x)is ∝Z dkydǫfF (ǫ, ky) √vx,−vx,+ ×"Ahχ−(ϕ1) Szχ+(ϕ)iei(k−k1 )x + c.c.#, 1 (8) where A = S+ + · (S−+ )⋆ + S+ − · (S− − )⋆. Here we used the fact that the distribution function fF (ǫ, ky), describing a particular lead, has the same value at given energy for plus and minus mode. Note, that the period of oscillations of the exponential factor ei(k−k1)x in Eq. (8) is of the order of the spin precession length. However, the term (8) vanishes, because the expression A is nothing but a non-diagonal component of the iden- tity matrix S S†. Thus, we obtain an interesting result, that the only reason for the cancellation of the long-wave length oscillations with the period Ls in hSz(x)i is the unitarity of scattering. By taking into account in Eq. (7) the terms responsible for the interference between incoming and the outgoing waves [for example, between first and second terms in Eq. (2)], and adding the contribution from the evanescent It modes [16], we reproduce Eq. (16) of the Ref. [7]. can be written in the form hSz(x)i = (eV /8π2mv2 F )ImI, where I =Z k− 0 dky k+k− + k2 ky y − kk1 (eikx − eik1x)2. Note, that in the interval k+ < ky < k−, the quantity k has purely imaginary value, which corresponds to the evanescent modes. From this form of the presentation, we can immediately see that hSz(x)i contains only 2kF com- ponent, while all the long-wavelength oscillations cancel exactly. Indeed, with the branch cut along the real ky axis between the points −k− and +k− (see Fig. 2), the 3 integrand function in I is an analytical function of the variable ky in the right half plane Reky > 0 (for positive x). Since we need the imaginary part of I, the integra- tion is going alone the upper edge of the branch cut from 0 up to +k− and then back along the lower edge of the branch cut. Because of the analyticity mentioned above, this integral is equal to the one taken along the imaginary axis of ky = iκ. Then, for x ≫ λF the latter integral is determined by small κ ≪ kF : I ≃ −2(eik−x − eik+x)2Z ∞ 0 dκκeixκ2/kF , the spin for gives density hSz(x)i which ≈ (eV /2π2vF x) cos(2mvF x) sin2(αmx/2), coinciding with the result of Ref. [7]. Therefore, the total spin per unit length along the boundary scales as R ∞ 0 dxhSz(x)i ∝ α2. Note, that the main contribution to this integral comes from small distances from the boundary, x ≃ λF . The cancellation of smooth spin density oscillations in case of the Rashba Hamiltonian and straight boundary occurs also in the quasi-ballistic situation: L ≫ l ≫ Ls, where L is the sample size, and l is the mean free path. In this case, the electric field in the bulk of the sample is finite. Therefore, the distribution functions for the plus and minus modes, f++(~k+) and f−−(~k−), are determined by the electric field and by scattering off the impurities in the bulk of a system [11]. The wave vectors ~k+ and ~k−, shown in Fig. 1, correspond to a given energy and a given wave vector along the boundary. In the quasi- ballistic case considered here, these functions are equal, i.e. f++(~k+) = f−−(~k−), similar to a ballistic situa- tion. Under such a condition, the unitarity of scattering, see Eq.(8), leads to the cancellation of smooth edge spin density oscillations, in contrast to what has been stated in the literature [12]. Indeed, when the electric field is parallel to the boundary, the distribution functions in questions are f++(~k+) = f++(k+) sin ϕ, and f−−(~k−) = f−−(k−) sin ϕ1 (see Eq. (9) of Ref. [11]). For the case of the Rashba Hamiltonian, the following relation has been obtained k+f−−(k−) = k−f++(k+) [13]. Then, the ratio is f++(~k+)/f−−(~k−) = k+ sin ϕ/k− sin ϕ1 = ky/ky = 1. Depending on the form of the s-o Hamiltonian, the unitarity may show up in totally different ways, leading, in general, to different patterns of the edge spin den- sity. In particular, boundary conditions play a crucial role. Let us consider 2D holes, described by the cubic (in 2D momentum) s-o Hamiltonian, still keeping in mind the ballistic case and an abrupt straight boundary. The unitarity condition, i.e. the charge flux conservation, ac- quires a different form. This may be seen already for the case of a normal incidence, where the Hamiltonian has the form p2 In the case of the plus incident mode, applying previous zero boundary condi- tions (which sometimes are used for this Hamiltonian in a numerical treatment of the edge spin accumulation prob- lem), one obtains F + + = 0, F −+ = i. In other words, af- x/2m + ασxp3 x. + + __ x y FIG. 3: Schematics of scattering of the plus incident mode by a wiggly boundary, x = W sin(2πy/λ). Apart from the main scattering channels (solid lines), there are additional scattering waves with the wave vectors along the boundary shifted by ±2π/λ (dashed lines). ter elastic backscattering the helicity value changes sign, which is a consequence of the fact that σx is conserved. Then, for the charge flux to be conserved, one needs the equality of the group velocities v+ and v−, corresponding to the plus and minus modes at the same energy. How- ever, in contrast to the case of the Rashba Hamiltonian, where v+ = v− , those velocities are not equal for the cubic Hamiltonian: v+ − v− = α(p+ + p−)2. The formal way to resolve the trouble is to note that the cubic Hamil- tonian has three solutions for a given energy, one of them corresponding to a momentum larger than kF (for small spin-orbit coupling). Thus, in general, the unitarity of scattering in slow plus and minus channels is violated, because some flux is carried away by a fast oscillating mode. Therefore, the corresponding coefficient A in Eq. (8) does not vanish, and smooth spin density oscillations occur. However, for a physically relevant confining po- tential which grows to infinity in a smooth manner, the fast oscillating modes are not excited. Then, the uni- tarity of scattering into slow modes is restored, and the smooth spin density oscillations do not appear. Let us consider scattering off a wiggly boundary, shown in Fig. 3, for the case of the Rashba Hamiltonian. In this case the translational invariance is broken, therefore the condition of the unitarity of scattering takes a different form, as compared to the case of a straight boundary. As a result, the cancellation of the smooth spin density oscillations does not take place, leading to the total spin that is not small in the parameter α [14, 15]. In order to demonstrate this effect, we consider the mathemati- cally simple case of the abrupt impenetrable boundary described by the equation: x = ζ(y) ≡ W sin(2πy/λ). To the lowest order in W , the boundary condition reads: Ψ(0, y) + ζ(y) d Ψ(0, y) dx = 0 (9) We are looking for the solution in the perturbative form Ψ±(x, y) = Ψ(0) (x, y), where the zeroth or- ± der functions are given by Eqs. (2) and (3). The first (x, y) + Ψ(1) ± 4 order correction, proportional to W , is the superposi- tion of scattering waves with the wave vectors along the boundary shifted by ±2π/λ (see Fig. 3), and with the k-vectors in the x-direction given by: > < )2. > k )2, k 2π λ 2π λ + − (ky ± − − (ky ± 1 =rk2 < =rk2 From now on, we assume λF ≪ Ls ≪ λ. the spin density, we consider the case x/√λλF ≪ 1. In ad- dition, in order to obtain an analytical expression for In contrast to the case of the straight boundary, there are oscillations with three different periods: 2kF - os- cillations, and the oscillations with two long periods, ξ and Ls. Here ξ = 1/qk2 + is the new length scale. Under the conditions considered in the paper we obtain the set of inequalities λF ≪ ξ ≪ Ls, where Ls = 1/(k− − k+) = ¯h/mα is the spin precession length. If ky → k+ (i.e., k → 0), then k1 tends to 1/ξ, which clarifies the physical meaning of ξ. For the contribution of the long wavelength oscilla- − − k2 tions, we obtain [17]: hSz(x, y)i = (2π)2¯hvF eV ) cos( 2πy λ )Ilong(x), Ilong(x) = 2 sin( x ξ ) + 2πW ( λ 2 cos( x ξ ) x ξ + 2x ξ ∂ ∂xZ 1 0 dze−(x/ξ)z cos + π 2Ls N1( 1 x x Ls ) − x√1 − z2 ξ , (10) x + 2 cos(x/ξ) where the last term is the contribution of the evanes- cent modes, and N1(x) is the Bessel function of the second kind. At the distances x ≪ ξ, we obtain the following dependence Ilong = −2x2/3ξ3 + (x/2L2 s)[γ + ln(x/2Ls)]. In the opposite limit, x ≫ ξ, we find N1(x/Ls) − 1 Ilong = π . At even larger 2Ls distances, x ≫ Ls, one obtains smooth oscillations with the period of the order of Ls, and the amplitude being proportional to √α. Note, that the total spin per unit length along the boundary is proportional to the integral R dxIlong(x) ≃ (π/2Ls)R ∞ dx/x ≃ − ln(Ls/ξ) − ln(√λλF /ξ) ≃ −(1/2) ln(λ/λF ), i.e., it is ξ dxN1(x/Ls)−R not small in s-o coupling, in contrast to the case of a straight boundary. √λλF ξ x In conclusion, we have considered the problem of the edge spin accumulation in mesoscopic structures with spin-orbit-related splitting of the energy spectrum, when the associated spin precession length is much smaller than the mean free path. In the presence of the charge current, the spin density develops oscillations near an edge in the direction transverse to the boundary. We have used the scattering states method to clarify the physics associated with this effect. The result crucially depends on the form of s-o Hamiltonian and the bound- ary conditions. The unitarity of scattering in the case of a straight boundary and Rashba Hamiltonian leads to the cancellation of long-wave length spin density oscilla- tions. On the contrary, the spin density in the case of wiggly boundary oscillates with a large period of the or- der of the spin precession length, quite similar to the case of electron scattering off antidots in 2D system with the Rashba Hamiltonian [10]. We acknowledge the financial support from the Pro- gram "Spintronics" of RAS, the Russian Foundation for Basic Research (Grant No. 07-02-00164-a), the Swiss Na- tional Science Foundation and SPINMET project (FP7- PEOPLE-2009-IRSES). [1] H.A. Engel, E.I. Rashba, and B.I. Halperin, in Handbook of Magnetism and Advanced Magnetic Materials, ed. by H. Kronmuller and S. Parkin, Vol.5 (John Wiley and Sons, New York, 2007). [2] M.I. Dyakonov, and A.V. Khaetskii, in Spin Physics in Semiconductors, Springer Series in Solid- State Sciences, ed. by M.I. Dyakonov (Springer, Berlin, 2008). [3] Y.K.Kato et al., Science 306, 1910 (2004). [4] S. Murakami et al., Science 301, 1348 (2003); J. Sinova et al., PRL 92, 126603 (2004). [5] J. Wunderlich et al., PRL 94, 047204 (2005). [6] Ya. Tserkovnyak et al., PRB 76, 085319 (2007); O. Bleibaum, PRB 74, 113309 (2006). [7] V.A. Zyuzin, P.G. Silvestrov, and E.G. Mishchenko, PRL 99, 106601 (2007). [8] G. Usaj and C.A. Balseiro, Europhys. Lett. 72, 621 (2005); A. Reynoso et al., PRB 73, 115342 (2006). 5 [9] Only in the effective 1D situation the unitarity of scat- tering leads to the cancellation of long-wavelength os- cillations of the spin density. This is no longer the case for 2D systems. The problem for electrons, described by the Rashba Hamiltonian, scattered off a circular scatterer (an antidot) may be solved exactly in the presence of a charge current. The Sz-component of the spin density, oscillating with period Ls, appears around the antidot [10]. [10] A. Khaetskii, E. Sukhorukov, to be published. [11] A. Khaetskii, PRB 73, 115323 (2006). [12] E.B. Sonin, PRB 81, 113304 (2010). [13] Note the different meaning of quantities k+ and k− used here, and quantities p+ and p− used in [11]. [14] Note, that Ref. [15] considers the case of a straight boundary, which is smooth in the transverse direction, instead of abrupt one. If the width of the boundary is larger than Ls, two neighboring strips of the spin density of the opposite signs arise. We stress, that the physics discussed in Ref. [15] is different from that we consider here. Since the boundary is still straight, the unitarity of scattering leads to the cancellation of smooth oscillations with the period Ls. As a result, the total spin, found in [15], is small and proportional to α2. [15] P.G. Silvestrov, V.A. Zyuzin, and E.G. Mishchenko, PRL 102, 196802 (2009). [16] When k+ < ky < k−, the wave vector in x-direction is purely imaginary for the reflected plus mode. This mode is called evanescent. This situation takes place only for the incident minus mode, since only this mode comes from the leads under the indicated conditions. [17] For details, see A. Khaetskii, E. Sukhorukov, JETP Lett. 92, 244 (2010). 

1510.06854

1

1510

Reliable determination of the Cu/n-Si Schottky barrier height by using in-device hot-electron spectroscopy

We show the operation of a Cu/Al_2O_3/Cu/n-Si hot-electron transistor for the straightforward determination of a metal/semiconductor energy barrier height even at temperatures below carrier-freeze out in the semiconductor. The hot-electron spectroscopy measurements return a fairly temperature independent value for the Cu/n-Si barrier of 0.66 $\pm$ 0.04 eV at temperatures below 180 K, in substantial accordance with mainstream methods based on complex fittings of either current-voltage (I-V) and capacitance-voltage (C-V) measurements. The Cu/n-Si hot-electron transistors exhibit an OFF current of ~2 * 10^-13 A, an ON/OFF ratio of ~10^5 and an equivalent subtreshold swing of ~96 mV/dec at low temperatures, which are suitable values for potential high frequency devices.

cond-mat.mes-hall

cond-mat

Reliable determination of the Cu/n-Si Schottky barrier height by using in-device hot-electron spectroscopy Subir Parui1,*, Ainhoa Atxabal1, Mário Ribeiro1, Amilcar Bedoya-Pinto1, Xiangnan Sun1, Roger Llopis1, Fèlix Casanova1, 2, Luis E. Hueso1, 2,* 1CIC nanoGUNE, 20018 Donostia-San Sebastian, Basque Country, Spain 2IKERBASQUE, Basque Foundation for Science, 48011 Bilbao, Basque Country, Spain Email: s.parui@nanogune.eu; l.hueso@nanogune.eu KEYWORDS: Schottky barrier, hot electron transistor, spectroscopy Abstract We show the operation of a Cu/Al2O3/Cu/n-Si hot-electron transistor for the straightforward determination of a metal/semiconductor energy barrier height even at temperatures below carrier- freeze out in the semiconductor. The hot-electron spectroscopy measurements return a fairly temperature independent value for the Cu/n-Si barrier of 0.66 ± 0.04 eV at temperatures below 180 K, in substantial accordance with mainstream methods based on complex fittings of either current- voltage (I-V) and capacitance-voltage (C-V) measurements. The Cu/n-Si hot-electron transistors exhibit an OFF current of ~2  10-13 A, an ON/OFF ratio of ~105 and an equivalent subtreshold swing of ~96 mV/dec at low temperatures, which are suitable values for potential high frequency devices. 1 The Schottky energy barrier naturally appearing at a metal/semiconductor (MS) interface is a critical parameter for the performance of many modern electronic devices, from mainstream metal-oxide-semiconductor field effect transistors (MOSFETs) to novel organic-based light- emitting diodes or photovoltaics [1-5]. Hot electron in-device spectroscopy is a powerful technique for determining such energy barrier between a metal and a semiconductor [6-9]. This technique is based in the hot-electron transistor (HET), a classical metal- A/insulator/metal-B/semiconductor (MIMS) device in which a hot electron current coming from a metal-A emitter crosses ballistically a metal-B base before entering into the semiconductor conduction band. The HET is particularly interesting since hot electrons can be collected in the semiconductor without biasing the MS interface, and hence they generate a collector current that can be used to probe the Schottky barrier without the effects of an external electric field at the interface. Moreover, the energy barrier height can be obtained directly from the experimental data with a simple theoretical model [10]. However, and in spite of its power and simplicity, hot-electron spectroscopy has been sparsely used for the study of metal/inorganic-semiconductor [6, 7, 10-14] and, more recently, metal/organic- semiconductor interfaces [8-9]. In this manuscript, we explore the prototypical Cu/n-Si interface and extract its energy barrier by hot-electron spectroscopy. Additionally, our HETs exhibit appealing electronic characteristics such as an OFF current of ~2  10-13 A, an ON/OFF ratio of ~105, and an equivalent subthreshold swing of ~96 mV/dec. The Cu/n-Si combination is particularly interesting since copper grows highly textured on Si [14, 15], has a high electrical conductivity and strong electromigration resistance, properties which make it a model system for investigation compared to other polycrystalline interfaces. 2 Previously, hot electron spectroscopy have been used for exploring the attenuation length of hot electrons in Cu [16, 17], identifying the band structure of the underlying n-Si substrate [14] and the crystallographic orientation of n-Si [16], detecting chemicurrent [18], as well as for spintronic applications [14, 19-20]. However, in our HET device, we capture useful information such as the reduction of thermal-leakage current with decreasing temperature [21] in addition to the reduction of quasi-elastic phonon scattering of hot electrons in the metal base and in the semiconductor collector [22]. Most importantly, we provide reliable determination of the Schottky barrier of Cu/n-Si interface at low temperatures even below carrier freeze-out temperature of Si, a regime that can be hardly accessed by other conventional electrical characterization methods. For putting this technique into the right perspective, we compare the results we obtained by hot-electron spectroscopy with those arising from standard current – voltage (I-V) and capacitance – voltage (C-V) measurements taken on the same device. Our work underlines in-device hot-electron spectroscopy as a method of choice to map out the Schottky barrier physics below carrier freeze-out temperatures in a wide range of semiconducting materials. 3  Figure 1: (a) Schematic diagram of the hot-electron transistor. The dotted line represents the I-V measurement setup to measure the base/collector diode. (b) Optical microscopy image of the real device. (c) Schematic energy diagram of the hot electron transistor, representing the negatively biased Cu emitter, the Al2O3 tunnel barrier, and the grounded Cu base in direct contact with n-Si(100) collector. The base/collector junction is unbiased. Figure 1(a) shows the schematics of the Cu/Al2O3/Cu/n-Si HET device. The collector used is a n-type Si(100) substrate, having resistivity of 5-10 cm (donor concentration, ND  1015 cm-3 at room temperature), with 300-nm-thick thermally grown SiO2 on top. A 11mm2 window was opened on the oxide layer by means of photolithography and buffered hydrofluoric acid etching. The etched area was then hydrogen terminated using 1% hydrofluoric acid, onto which a 20-nm-thick Cu base layer was deposited by e-beam evaporation in ultrahigh vacuum (UHV) through a shadow mask. For the hot-electron 4 injection by quantum tunneling, a 3-nm-thick Al layer was then evaporated all over the device without breaking the vacuum and in-situ plasma oxidized, followed by another 20- nm-thick Cu emitter layer deposition through a different shadow mask. Prior to the evaporation of the second Cu layer, and to avoid voltage breakdown up to an emitter voltage VE 1.4 V of the thin tunnel barrier at the edges of 300-nm-thick SiO2, an additional 10-nm- thick Al2O3 layer was evaporated in the area and surroundings of the Cu layers with the use of another shadow mask [see Fig. 1(a)]. Figure 1(b) pictures the top view of an actual device, with the emitter, base and collector indicated. The active area for hot-electron injection and collection is considered to be confined to the overlapping area between the emitter and base electrodes (400  800 µm2). Once fabricated, the device was then transferred into a variable- temperature probe station (Lakeshore) for electrical measurements with a Keithley-4200 semiconductor analyzer. Figure 1(c) represents the energy diagram of the HET device. The arrows represent the hot-electron transport above the Cu/n-Si barrier (red) and the electron reflection at the barrier interface below the barrier (blue), respectively. At a sufficiently high negative emitter bias (-VE), the electrons tunnel through the barrier and cross the metal base ballistically with enough energy to surpass the base/collector Schottky barrier and get collected into the semiconductor, providing the collector current. The onset of the output characteristic of the device represents accurately in a first approximation the height of the Schottky energy barrier. Those electrons that do not have enough energy to surpass the Cu/n- Si energy barrier are scattered/reflected at the base/collector interface and contribute to the base current directed to ground. 5  Figure 2: (a) Plots of the collector current (IC) versus emitter bias (VE) at different temperatures for grounded base and at zero collector bias. (b) IC-VE plots at different temperatures with 5 K step. (c) IC-T plots at different VE, which describe the reduction of thermal leakage current and collection of hot-electron current. Inset shows the collector current only due to hot-electrons. (d) Plots of the emitter current (IE) versus emitter bias (VE) at different temperatures. In order to investigate the collector current originated from the ballistic transport of hot electrons through the Cu-base, we measure the temperature dependence of the collector current (IC) versus emitter voltage (VE) from 300 K down to 10 K (see figure 2). As noted above, the onset of the collector current represents an adequate measurement for the energy barrier at the Cu/n-Si interface. At 300 K, a slowly varying collector current of ~ 10-5 A is observed without any well-defined onset above VE = -0.3 V, as it can be seen in Figure 2 (a). The magnitude of this collector current decreases as temperature does, confirming that its 6 origin is the regular thermal leakage current at the Cu/n-Si Schottky interface [21]. For temperatures below 180 K, the hot-electron current (ON) arises for VE higher than -0.5 V, reaching values as high as ~ 10-8 A, while the OFF current (for VE lower than -0.5V) is as low as ~2  10-13 A. The OFF current measured is around two orders of magnitude lower than the reported in recent literature [20] and, accordingly, our HET devices exhibit a large ON/OFF ratio of ~105. The equivalent subthreshold swing (defined as the required VE for a ten-fold change in IC) is ~96 mV/dec at ~80 K and below, a very suitable value for applications in high-speed and frequency integrated-circuits [24, 25]. Figure 2 (b) shows in detail (in 5 K steps) the evolution of the hot electron current for temperatures below 190 K, while Figure 2 (c) displays an exponential decrease with decreasing temperature for the thermal leakage down to 160 ± 20 K. However, at temperatures below 160 K, the hot- electron current increases linearly for VE voltages larger than the metal/semiconductor barrier height. Finally, for temperatures below ~20 K the hot electron current decreases due to the carrier freeze out in n-Si where the semiconductor becomes highly resistive (see Figure 2 (c), inset) [26]. Figure 2 (d) represents the tunnel current vs emitter bias at several temperatures. Observation of a large amount of emitter current and also the linearity at high emitter bias suggest the possibilities of unavoidable current through the pinholes in the barrier that might originated from the island growth of Al on top of Cu. However, the thin amorphous Al2O3 barrier in our devices is good enough for the injection of energetic electron from the Cu- emitter into the Cu-base across the MIM structure. The emitter current slowly decreases with decreasing temperatures, whereas the hot-electron current increases, possibly due to the reduction of quasi-elastic acoustic-phonon scattering in the base as well as in the collector at low temperatures [22]. As seen from this set of results, we describe the evolution of the 7 collector current from 300 K to 10 K to interpret the origin of hot electron current and thermal-leakage current, while some previous experiments of hot electron transport across Cu/n-Si interface are reported at temperatures below 150 K [14, 16, 17, 19, 20, 27]. Figure 3: (a) Hot-electron current plotted as (IC/IE)1/2 versus VE superimposed to the linear fit (solid line) at several temperatures. Linear fits are used to extract the barrier heights. (b) Temperature dependent base-collector I-V diode characteristics in semi-logarithmic scale. (c) Base-collector C-2-V characteristics of the diode at a frequency of 1 MHz and in the temperature range from 300 K to 20 K. Solid lines represent the extrapolation of the curvature of the experimental data. (d) Temperature dependence of the barrier heights extracted by different experimental methods i.e., hot-electron, C-V, and I-V (zero-bias and flat-band); measurements are plotted in the left vertical axis. Right vertical axis represents the change in ideality factor of the diode. 8 For the reliable determination of the Schottky barrier at the Cu/n-Si interface, we compare the values extracted from hot-electron spectroscopy at several temperatures with the ones inferred from I-V and C-V methods of the same base/collector junction [see Figure 3]. In hot- electron spectroscopy, the barrier height is extracted as the intercept with the voltage axis for zero collector current. Due to the parabolic conduction band minimum in n-Si, the theoretical dependence (Bell and Kaiser model) of IC with VE [10] is 𝐼𝐶 ∝ 𝐼𝐸(𝑉𝐸 − 𝜙)2, where ϕ is the barrier height of the metal base/semiconductor collector interface. Figure 3 (a) shows the hot- electron spectroscopy data, where (IC/IE)1/2 is plotted with respect to VE and fitted linearly to the experimental curve. According to Schottky-Mott relationship, the value of ϕ is typically the difference between the base-metal work function and the conduction band edge of n- Si(100) collector in the ideal case. However, in general, the energy-level alignment is influenced by additional effects such as Fermi-level pinning, so that the rigid-band approximation is no longer suitable for capturing the actual barrier height [1, 2, 23]. Figure 3 (d) displays experimentally extracted barrier heights by hot-electron spectroscopy and we can observe that its value below 180 K is constant around ~ 0.66 ± 0.04 eV. We can now compare the hot-electron spectroscopy data with other mainstream techniques for obtaining the energy barrier height. In the first place, we show two-probe I-V measurements of the base/collector (metal/semiconductor) junction [see Figure 1 (a) for the setup and Figure 3 (b) for the results]. The data shows clearly the rectifying characteristics of a MS diode. The rectification factor reaches several orders of magnitude and it increases with decreasing temperature due to the sharp decrease of the reverse-bias saturation current (Isat). However, below 40 K the forward saturation current starts to decrease due to the fact that the Si-collector becomes resistive because of carrier freeze-out in Si. The amplitude and 9 temperature dependence of the reverse saturation current varies in a similar way as the hot- electron leakage current, which confirms that the reverse saturation current of the base/collector diode contributes to the leakage current. Using the linear part of the forward bias regime, we extract the barrier height after a fitting with the thermionic emission (TE) equation [1, 2] according to which 𝐼𝐶−𝐵 = 𝐼𝑠𝑎𝑡 [𝑒𝑥𝑝 ( 𝑞𝑉𝐶−𝐵 𝜂𝑘𝐵𝑇 )] for 𝑉𝐶−𝐵 > 3𝑘𝐵𝑇/𝑞, where kB is the Boltzmann constant, T is the temperature and 𝜂 is the ideality factor. Furthermore, 𝐼𝑠𝑎𝑡 = 𝐴𝐴∗𝑇2𝑒𝑥𝑝 (− 𝑞𝜙𝑏0 𝑘𝐵𝑇 ), where A is the base/collector junction area and A* is the effective Richardson constant (~110 A cm-2 K-2 for n-Si) [1]. In principle, 𝜙𝑏0 is meant to be the same Cu/n-Si Schottky barrier as determined by hot electron spectroscopy; however, due to the different measurement methods we denote this as zero-bias barrier height [1] and plot the extracted barrier height in Figure 3 (d) along with the ideality factor 𝜂, a quantity which should be close to 1 for a well-defined Schottky interface. The zero-bias barrier height obtained from the IC-B-VC-B measurement decreases monotonically from 0.60 ± 0.04 eV at 300 K to 0.046 ± 0.003 eV at 10 K, whereas the ideality factor sharply increases from 1.02 ± 0.04 at 300 K to 19.06 ± 0.95 at 10 K. These phenomena are commonly observed in Schottky diodes [28-30], and can be explained by considering an inhomogeneous distribution of barrier heights at the MS interface [31, 32] only in a temperature regime without the carrier freeze-out in the semiconductor. Here, such strong disagreement between these two measurement methods suggests that the zero-bias barrier height decreases artificially at low temperatures below 60 K, where the diode becomes extremely non-ideal and the thermionic emission model fails to explain the experimental data. Furthermore, taking into account the zero-bias barrier height and the ideality factor, the 𝜙𝑏𝑓 (flat-band barrier height) [33] can be expressed as 𝜙 𝑏𝑓 = 𝜂𝜙𝑏0 + (𝜂 − 1)(𝑘𝐵𝑇 𝑞⁄ )[ln (𝑁𝐶 𝑁𝐷⁄ )]. The extracted 𝜙𝑏𝑓 is plotted in 10 Figure 3(d) representing the correction of the non-ideal 𝜙𝑏0 to achieve a more realistic Schottky barrier height. In addition, it is also possible to obtain a Schottky barrier height from the Richardson plots using the reverse bias diode characteristics [2, 4, 34]. We find the Cu/n-Si Schottky barrier height to be 0.52 ± 0.02 eV, fitting the Isat current only in the high temperature regime (300 K to 200 K). However, this method of extracting the Schottky barrier height underestimates the influence of the diode ideality factor. In the second place, we show the capacitance (C)-voltage (V) measurements and how we can extract in this case again the metal/semiconductor energy barrier, which we will denote as ϕ (C-V) [see Figure 3 (c)]. The applied voltages in this particular case range from the reverse- bias condition to a small regime of forward-bias until the diode starts conduction i.e., maximum up to -0.36 V. In the simplest case of a linear relation between C-2 and V we can write [2], 1 𝐶2⁄ = (2 𝑞𝑁𝐷𝜀𝑠𝐴2 ⁄ ) ∙ [𝑉𝐼 + 𝑉𝐶−𝐵], where 𝑉𝐼 = (𝜙 − 𝑘𝐵𝑇 𝑞⁄ − 𝜉) represents the linear intercept on the voltage axis and 𝜀𝑠 is the permittivity of Si. In case of linear dependence, the barrier height can be extracted as 𝜙 = 𝑉𝐼 + 𝑘𝐵𝑇 𝑞⁄ + 𝜉, where 𝜉 = (𝑘𝐵𝑇 𝑞⁄ )[ln (𝑁𝐶 𝑁𝐷⁄ )] is the energy difference between the Fermi level and the conduction band edge in Si. The effective density of states in the conduction band edge is given by NC  2.8  1019 cm-3[1]. However, we notice a strong nonlinear dependence of the C-2-VC-B characteristics. In case of nonlinearity, the C-2 can be expressed as a quadratic function i.e., 1 𝐶2⁄ = 𝑀1[𝑉𝐼 + 𝑉𝐶−𝐵] + 𝑀2[𝑉𝐼 + 𝑉𝐶−𝐵]2, where M1 and M2 are the fitting parameters [35]. Such nonlinear dependence could be due to a contribution of the interface charge capacitance that goes in adverted by other experimental techniques. By using a quadratic dependence around zero-bias voltage, we extract the intercept on the voltage axis, which is 11 then used to determine the barrier as shown in Figure 3 (d). It is worth noting that, in our measurement, the C-2-VC-B characteristics strongly deviate at 20 K, which could be possibly due to carrier freezing in n-Si at ~20 K and below. Figure 3 (d) shows the different values of the barrier height extracted by hot electron spectroscopy, I-V and C-V measurements, i.e., 𝜙(hot-electron), 𝜙 (zero-bias), 𝜙 (flat-band) and 𝜙 (𝐶 − 𝑉), respectively. They all show a similar weak temperature dependence, however 𝜙 (flat-band) and 𝜙 (𝐶 − 𝑉) diverge at very low temperatures, while 𝜙(hot-electron) remains stable for all the temperatures obtained. Consequently, the hot-electron spectroscopy provides a reliable and straightforward method for the barrier height extraction in the absence of an applied electric field at the Cu/n-Si interface and at temperatures below the carrier freeze-out. Being an in-device method, as opposed to photoemission spectroscopic techniques, it provides a realistic approach to the metal/semiconductor barrier height determination for its application in electronic device design. Finally, we summarize the measured Schottky barrier heights by different methods in Table I, where we find a relatively good agreement with the existing literature [16, 17, 36] only in some certain temperatures. TABLE I. Measured Schottky barrier heights by different methods and its comparison with the literature. T 𝜙 𝜙 (C-V) 𝜙 bo (I-V) 𝜙 bf 𝜙 (hot-e) from Ref. 16 𝜙 (hot-e) from Ref. 17 𝜙 b0 (I-V) from Ref. 35 (hot-e) - - 0.67 eV 0.67 eV 0.66 eV 0.66 eV 300 K 200 K 100 K 80 K 40 K 10 K 0.71 eV 0.71 eV 0.70 eV 0.69 eV 0.68 eV - 0.60 eV 0.56 eV 0.39 eV 0.33 eV 0.19 eV 0.05 eV 0.61 eV 0.60 eV 0.59 eV 0.58 eV 0.60 eV 0.74 eV - - 0.62 eV - - - - - - 0.64 eV - - 0.59 eV 0.61 eV 0.63 eV - - - 12 In conclusion, we have demonstrated the operation of a Cu/n-Si-based hot-electron transistor and determined the height of its base/collector (metal/semiconductor) Schottky energy barrier. In this experiment we are able to fully describe the collector current that originates from hot-electron transport and to extract the Schottky barrier height over a wide temperature range. For completeness, we have compared the extracted barrier height with the other measurements methods such as the I-V and the C-V techniques, concluding that hot-electron spectroscopy is the most reliable method to map out Schottky barrier heights at very low temperatures. In addition, our results highlight in-device hot-electron spectroscopy as a straightforward method to determine the metal (base)/semiconductor (collector) energy barrier in real device operation conditions. Our optimized hot-electron transistor, with a current ON/OFF ratio of up to five orders of magnitude, low OFF state current, and an equivalent subthreshold swing of ~96 mV/dec at low temperatures, could be suitable for high-frequency device applications in cryogenic environments. We acknowledge P. Stoliar for his help in the shadow mask design and for discussion regarding the capacitance measurements. We also acknowledge financial support from the European Union’s 7th Framework Programme under the European Research Council (Grant 257654-SPINTROS), under People Programme (Marie Curie Actions) REA grant agreement 607904-13, and under the NMP project (NMP3-SL-2011-263104- HINTS) as well as by the Spanish Ministry of Economy under Project No. MAT2012-37638. References [1] [2] [3] S. M. Sze, Physics of Semiconductor Devices, 2nd ed. 1981. E. H. Rhoderick and R. H. Williams, Metals Semiconductor Contacts, 2nd ed. 1988. R. H. Friend, R. W. Gymer, A. B. Holmes, J. H. Burroughes, R. N. Marks, C. Taliani, 13 S. Parui, L. Pietrobon, D. Ciudad, S. Vélez, X. Sun, F. Casanova, P. Stoliar, L. E. J. P. Spratt, R. F. Schwarz, and W. M. Kane, Phys. Rev. Lett. 6, 341–342, 1961. C. Guinet, Appl. Phys. Lett. 25, 600–602, 1974. J. S. Jiang, J. E. Pearson, and S. D. Bader, Phys. Rev. Lett. 106, 156807, 2011. D. D. C. Bradley, D. A. Dos Santos, J. L. Brédas, M. Lögdlund & W. R. Salaneck, Nature 397, 121, 1999. [4] Hueso, Adv. Funct. Mater. 25, 2972, 2015. [5] M. F. Lo, T. W. Ng, T. Z. Liu, V. A. L. Roy, S. L. Lai, M. K. Fung, C. S. Lee, and S. T. Lee, Appl. Phys. Lett. 96, 113303, 2010. [6] [7] [8] [9] M. Gobbi, L. Pietrobon, A. Atxabal, A. Bedoya-Pinto, X. Sun, F. Golmar, R. Llopis, F. Casanova, and L. E. Hueso, Nat. Commun. 5, 4161, 2014; M. Gobbi, A. Bedoya-Pinto, F. Golmar, R. Llopis, F. Casanova and L. E. Hueso, Appl. Phys. Lett. 101, 102404, 2012. [10] L. D. Bell and W. J. Kaiser, Phys. Rev. Lett. 61, 2368, 1988. [11] W. Yi, A.J. Stollenwerk, V. Narayanamurti, Surface Science Reports 64, 169-190, 2009. [12] K. G. Rana, S. Parui, and T. Banerjee, Phys. Rev. B 87, 085116, 2013. [13] S. Roy, A. M. Kamerbeek, K. G. Rana, S. Parui, and T. Banerjee, Appl. Phys. Lett. 102, 192909, 2013. [14] S. Parui, K. G. Rana, and T. Banerjee, Appl. Phys. Lett. 103, 082409, 2013. [15] B. G. Demczyk, R. Naik, G. Auner, C. Kota, and U. Rao, J. Appl. Phys. 75, 1956– 1961, 1994. [16] S. Parui, J. R. R. van der Ploeg, K. G. Rana, and T. Banerjee, Phys. Status Solidi RRL – Rapid Res. Lett. 5, 388–390, 2011. [17] Appl. Phys. Lett. 96, 062105, 2010. [18] H. Nienhaus, H. S. Bergh, B. Gergen, A. Majumdar, W. H. Weinberg, and E. W. McFarland, Phys. Rev. Lett. 82, 446–449, 1999. [19] [20] Y. Lu, D. Lacour, G. Lengaigne, S. L. Gall, S. Suire, F. Montaigne, and M. Hehn, Appl. Phys. Lett. 103, 022407, 2013. [21] M. Prietsch, Phys. Rep. 253, 163–233, 1995. [22] C. A. Ventrice Jr., V. P. LaBella, G. Ramaswamy, H.-P. Yu, and L. J. Schowalter, Appl. Surf. Sci. 104–105, 274–281, 1996. [23] [24] M. Heiblum, I. M. Anderson, and C. M. Knoedler, Appl. Phys. Lett. 49, 207, 1986. [25] N. Evers, J. Laskar, N. M. Jokerst, T. S. Moise, and Y.-C. Kao, Appl. Phys. Lett.70, 2452, 1997. [26] G. N. Henderson, P. N. First, T. K. Gaylord, and E. N. Glytsis, Phys. Rev. Lett. 71, 2999, 1993. [27] B. Huang, D. J. Monsma, and I. Appelbaum, Phys. Rev. Lett. 99, 177209, 2007. [28] S. Chand and J. Kumar, Appl. Phys. A 63, 171–178, 1996. [29] H.-W. Hübers and H. P. Röser, J. Appl. Phys. 84, 5326–5330, 1998. [30] S. Parui, R. Ruiter, P. J. Zomer, M. Wojtaszek, B. J. van Wees, and T. Banerjee, J. Appl. Phys. 116, 244505, 2014; D. Tomer, S. Rajput, L. J. Hudy, C. H. Li and L. Li, Nanotechnology 26, 215702, 2015. [31] R. T. Tung, Appl. Phys. Lett. 58, 2821–2823, 1991. J. J. Garramone, J. R. Abel, I. L. Sitnitsky, L. Zhao, I. Appelbaum, and V. P. LaBella, I. Appelbaum, B. Huang, and D. J. Monsma, Nature 447, 295–298, 2007. J. Bardeen, Phys. Rev. 71, 717, 1947; R. T. Tung, Phys. Rev. B 64, 205310, 2001. 14 J. H. Werner and H. H. Güttler, J. Appl. Phys. 69, 1522–1533, 1991. [32] [33] L. F. Wagner, R. W. Young, and A. Sugerman, IEEE Electron Device Lett. 4, 320– 322, 1983. [34] S. Mehari, A. Gavrilov, S. Cohen, P. Shekhter, Lett. 101, 072103 (2012). [35] A. M. Goodman, J. Appl. Phys. 34, 329–338, 1963. [36] M. O. Aboelfotoh and B. G. Svensson, Semicond. Sci. Technol. 6, 647, 1991. and M. Eizenberg, Appl. Phys. 15 

1209.2997

1

1209

Self-sustaining dynamical nuclear polarization oscillations in quantum dots

Early experiments on spin-blockaded double quantum dots revealed surprising robust, large-amplitude current oscillations in the presence of a static (dc) source-drain bias [see e.g. K. Ono, S. Tarucha, Phys. Rev. Lett. 92, 256803 (2004)]. Experimental evidence strongly indicates that dynamical nuclear polarization plays a central role, but the mechanism has remained a mystery. Here we introduce a minimal albeit realistic model of coupled electron and nuclear spin dynamics which supports robust self-sustained oscillations. Our mechanism relies on a nuclear-spin analog of the tunneling magnetoresistance phenomenon (spin-dependent tunneling rates in the presence of an inhomogeneous Overhauser field) and nuclear spin diffusion, which governs dynamics of the spatial profile of nuclear polarization. The extremely long oscillation periods (up to hundreds of seconds) observed in experiments as well as the differences in phenomenology between vertical and lateral quantum dot structures are naturally explained in the proposed framework.

cond-mat.mes-hall

cond-mat

Self-sustaining dynamical nuclear polarization oscillations in quantum dots 1 The Niels Bohr International Academy, Blegdamsvej 17, DK-2100 Copenhagen Ø, Denmark M. S. Rudner1,2,3 and L. S. Levitov4 2 Department of Physics, The Ohio State University, 191 W. Woodruff Ave., Columbus, OH 43210 3 Institute for Quantum Optics and Quantum Information of the Austrian Academy of Sciences, A-6020 Innsbruck, Austria 4 Department of Physics, Massachusetts Institute of Technology, 77 Massachusetts Ave., Cambridge, MA 02139 Early experiments on spin-blockaded double quantum dots revealed surprising robust, large- amplitude current oscillations in the presence of a static (dc) source-drain bias [see e.g. K. Ono, S. Tarucha, Phys. Rev. Lett. 92, 256803 (2004)]. Experimental evidence strongly indicates that dynamical nuclear polarization plays a central role, but the mechanism has remained a mystery. Here we introduce a minimal albeit realistic model of coupled electron and nuclear spin dynamics which supports robust self-sustained oscillations. Our mechanism relies on a nuclear-spin analog of the tunneling magnetoresistance phenomenon (spin-dependent tunneling rates in the presence of an inhomogeneous Overhauser field) and nuclear spin diffusion, which governs dynamics of the spatial profile of nuclear polarization. The extremely long oscillation periods (up to hundreds of seconds) observed in experiments as well as the differences in phenomenology between vertical and lateral quantum dot structures are naturally explained in the proposed framework. The coupling of electron and nuclear spin dynamics is responsible for a wide variety of intriguing transport phenomena in semiconductor devices. Spin exchange between electron and nuclear spins provides a mecha- nism for electron spin flips which can dramatically alter the behavior of systems such as spin-blockaded quantum dots, where transport is highly sensitive to spin selec- tion rules[1 -- 7]. Furthermore, the nuclear spins produce a hyperfine (Overhauser) field that shifts the electronic Zeeman energy by an amount corresponding to an effec- tive field that may reach as high as a few Tesla when the nuclei are fully polarized. This Overhauser field can have dramatic consequences for transport in quantum dots, where discrete levels may be shifted in-to or out-of resonance[4, 8 -- 10]. The combination of these two effects -- electron-nuclear spin-exchange which polarizes nuclear spins, and subsequent back-action on energy-dependent spin flip rates -- is responsible for a variety of interesting nonlinear dynamical effects such as multistability, hys- teresis, and intermittency[8 -- 15]. Among all of the nonlinear phenomena which have been observed thus far in transport through double quan- tum dots (DQDs), perhaps the most striking is the ap- pearance of spontaneous, stable current oscillations un- der the application of a dc source-drain bias[8, 16]. This phenomenon is remarkable for a number of reasons. First, the oscillations occur with very long periods ranging from seconds to hundreds of seconds. These timescales are 107− 109 times longer than the (1 pA)/e ∼ 100 ns micro- scopic timescale associated with single electron tunneling through the double dot. Second, the oscillations are ac- companied by long transients and long memory times when the source-drain bias is switched off and on. Fi- nally, after many years of experiments by a variety of groups, the oscillations have only ever been seen in ver- tical DQDs; the phenomenon has never been observed in a gate-defined lateral double quantum dot. FIG. 1: Mechanism of nuclear polarization oscillations in a spin-blockaded double quantum dot. Polarization is driven on a short time scale by resonant hyperfine transitions inside the DQD. Spin injection in the presence of an inhomogeneous Overhauser field leads to a polarization overshoot in the dot. Nuclear spin diffusion homogenizes the Overhauser field on a much longer timescale. Spin-flip transition rates inside the DQD adapt but lead to an overshoot in the opposite direction. Using nuclear magnetic resonance it was shown that the oscillations are in some way driven by nuclear spin dynamics[8]. However, despite wide interest in the prob- lem, a viable mechanism has thus far remained elusive. Here we present a straightforward mechanism which nat- urally produces oscillations with similar phenomenology. The mechanism relies on nuclear spin diffusion [17, 18] and on spin-dependent tunneling rates [19] which are con- trolled by the spatial profile of the Overhauser field. Nuclear spin diffusion, being a slow process, introduces the correct timescale into the dynamics. Furthermore, it also accounts for a sharp difference in predicted phe- nomenology for vertical and lateral DQD structures. The length scale for out-of-dot diffusion is set by the combi- nation of barrier and quantum well half-widths, and is typically a few tens of nanometers. For typical diffusion parameters[17, 18] this translates into diffusion times on the order of 10 seconds, consistent with the observed oscillation period values. These timescales are much longer than those arising from coherent mechanisms[20]. Additionally, in vertical DQDs such as those used in Ref. [8, 16], the edges of the dot are defined by the mesa structure itself. In such structures, nuclear po- enhancesT−loadingDNPovershoots,diffusestobarrierPolarizationbuild-upinDQD,Izdot<0HomogenizedDNPrestoresT+loadingPolarizationinDQDdrivenback,Izdot>0BarrierpolarizationrespondsslowlyDNPovershootsininoppositedirectionLevelalignmentfavorsT−-SflipsInhomogeneousDNPlarization predominantly diffuses in the vertical direc- tion, into the adjacent tunnel barriers. For gate-defined lateral DQDs, however, nuclear polarization can diffuse in all directions[24], and is not expected to flow signifi- cantly into the tunnel barriers. Thus the feedback effect via polarization-dependent tunneling, which is responsi- ble for the oscillations, is expected to occur in vertical DQDs but not in lateral DQDs. This is consistent with the observation that oscillations are frequently observed in vertical structures but never in the lateral structures. Schematically, oscillations arise as described in Fig. 1. An initial imbalance of hyperfine spin flip rates for up and down electron spins leads to a fast build up of nu- clear polarization inside the DQD. The resulting inho- mogeneity of the Overhauser field between the DQD and its surroundings enhances the probability of injecting the spins with the dominant hyperfine rate. This causes the polarization inside the dot to "overshoot." On a much longer time scale, nuclear spin diffusion homogenizes the Overhauser field. As the spin-injection probabilities re- act accordingly, the balance of hyperfine transition rates inside the DQD reverses, and starts to drive the nuclear polarization in the dot back toward zero. In a similar way, the polarization inside the dot again overshoots and then the cycle repeats. Here we describe the coupled electron and nuclear spin dynamics through the simplest possible model which qualitatively captures the behavior of the essential phys- ical degrees of freedom. In principle, detailed numerical modeling of the dot and of the full spatial profile of nu- clear polarization could be attempted. However, such modeling would introduce a much higher level of com- plexity, and would likely cloud rather than clarify the essence of the oscillation mechanism. Instead, we will write a set of dynamical equations for two polarization variables, one representing the polarization within the double quantum dot, and one representing the polariza- tion under the tunnel barrier to the source lead. The intradot polarization variable is driven by hyperfine spin flip processes with electron spins within the DQD. Polar- ization is then transferred to the barrier via spin diffusion with a large time constant. The delayed reaction of the barrier polarization variable to the intradot spin dynam- ics leads to oscillations as outlined above. A key to our mechanism is the difference in probabili- ties for spin-up and spin-down electrons to tunnel into the quantum dot when it is empty[19]. Naively, one might expect the respective tunneling rates to differ due to the application of a homogeneous Zeeman field, since the final state energies are different. However, as shown in Fig.2a, up and down spins tunnel under identical Zeeman-shifted barriers. Provided that the dot levels are set far below the chemical potential of the lead, EF , and that the lead has an approximately constant density of states in the energy range of interest, the tunneling-rates for the two spin species are equal in the case of a homogeneous field. 2 FIG. 2: Spin-dependent tunneling due to inhomogeneous Overhauser field. a) For a homogeneous Zeeman field, up and down spins are subjected to identical barriers and tunnel into an empty dot with equal probabilities. b) When nuclear polarization is nonzero only under the barrier, y (cid:54)= 0, up and down spins are subjected to different barriers (B = 0 for illustration in b-d). c) When nuclear polarization is nonzero only inside the quantum dot, x (cid:54)= 0, up and down spins tunnel in at different relative energies. d) If the nuclear polarizations in the dot and in the barrier are nonzero but equal, x = y, up and down spins tunnel in with equal probabilities. What happens in the case of an inhomogeneous Zeeman-Overhauser field? For demonstration, consider the case shown in Fig.2b, where the nuclear polarization is large under the barrier and zero outside. Here the Overhauser field locally increases the Zeeman energy un- der the barrier, effectively creating a higher barrier for down spins, and a lower barrier for up spins. In this sit- uation, an empty dot is more likely to be filled by a spin- up electron than by a spin-down electron. Similarly, as shown in Fig.2c, nuclear polarization concentrated only inside the dot can also affect the tunneling-in probabil- ities by changing the tunneling energies relative to the tops of the spin-up and spin-down barriers. It is interesting to note the similarity between this ef- fect and the phenomenon of tunneling magnetoresistance (TMR)[21, 22]. In both cases transport is dominated by tunneling through a barrier and spin polarization is used as a knob to control tunneling rate. While in TMR the spin polarization is due to magnetization in the regions surrounding the barrier, in our case the dominant effect is due to under-barrier nuclear spin polarization. The discovery of TMR has had important consequences for magnetic memory applications. One can envision that some of these ideas can be transposed to DQD systems. We now consider sequential electron transport through a spin-blockaded double quantum dot connected to leads with an applied dc source-drain bias, as described for example in Refs. [8, 11, 12]. In the two-electron spin- a)b)c)d)EqualTunnelingSpinUpFavoredSpinDownFavoredEqualTunnelingEFV↑V↓g∗µBBEFEFEFg∗µBBy<0x<0x=yV↓V↑>AyAx3 The intradot polarization x controls feedback through the Overhauser shift of the electronic triplet levels, which can bring these levels into or out-of resonance with the singlet. The energies ε± of the triplet states T±(cid:105), relative to S(cid:105), are given by ε± = ε ± g∗µBB ± Ax, (1) where ε is the singlet-triplet detuning, g∗ is the electronic effective g-factor (g∗ ≈ −0.44 in GaAs), µB is the Bohr magneton, B is the strength of the applied magnetic field, and A ∼ 100 µeV is the hyperfine coupling constant. Each time an electron decays from T+(cid:105) or T−(cid:105) to S(cid:105) via hyperfine exchange, one nuclear spin is flipped from down to up, or up to down, respectively. The probability for an electron that enters the dot to cause a positive (negative) increment to the nuclear polarization during its escape is determined by the probability f+ (f−) that the electron entered into the state T+(cid:105) (T−(cid:105)), and by the probability that the electron escapes via the hyperfine exchange process rather than by alternative nuclear-spin- independent mechanisms[11, 23]. The hyperfine spin flip probabilities are determined by the ratios W HF ± + W in), where W HF ± is the hyperfine decay rate of T±(cid:105) and W in describes the collective effects of spin-orbit coupling, spin exchange with the leads, and cotunneling processes. In our model, we assume that all nuclear spin flips due to hyperfine exchange with the electrons occur within the double quantum dot. Therefore the dot polarization x receives kicks (with magnitude 1/N ) on the timescale of single electron hopping through the dot, 100 ns to 1 µs, while the barrier polarization y has no dynamics on this small timescale. On a much longer timescale, nuclear polarization may diffuse from the dot region into the barrier region, providing a source for y. ± /(W HF Mathematically, it is simplest to analyze the regime where W in (cid:29) W HF ± . Here the total current, i.e. the ef- fective frequency of electrons passing through the dou- ble dot, is determined by W in. Additionally, the hyper- fine decay probabilities reduce to W HF ± /W in. The de- pendence on W in cancels from the nuclear polarization rate, which depends on products of attempt frequencies and spin flip probabilities, leaving behind contributions proportional to the hyperfine rates W HF ± weighted by the loading probabilities f±: − − 2ΓD(x − y) (2) x = f+W HF y = −2ΓDy + ΓDx, + − f−W HF (3) where ΓD ∼ 0.1 s−1 is the inverse of the time constant for diffusion from the dot to the barrier. The hyperfine spin-flip rates W HF ± are given by Fermi's golden rule[11]: W HF (4) ± = A2 N (1 − x) γ ± + γ2 , ε2 where γ is the decay rate of S(cid:105) due to its coupling to the drain. We account for the dependence of the loading FIG. 3: Two-electron energy levels involved in spin- blockaded transport (adapted from Ref.[23]). a) As a func- tion of interdot potential bias, which controls the asymmetry of the double well potential, the (1,1) and (0,2) singlet states exhibit an anticrossing. b) Energy levels at large detuning, indicated by the dashed vertical line in a). The singlet lev- els are broadened due to the coupling of the (0,2) state to the drain lead. Hyperfine-assisted transitions from T±(cid:105) to S(cid:105) provide a source for the nuclear polarization x within the double dot. blockade regime, "(1,1)" orbital configurations with one electron in each dot, and a "(0,2)" configuration with both electrons in the second dot have nearly the same electrostatic energies. In the (1,1) configuration, where overlap between electrons is negligible, all four spin states (one singlet and three triplet states) are nearly degener- ate in energy. For the (0,2) configuration, however, only the spin singlet configuration is allowed due to the Pauli exclusion principle (the single dot orbital level spacing is assumed to be much larger than the applied bias). Inter- dot tunneling hybridizes the (1,1) and (0,2) singlet states, producing the states labeled S(cid:105) and S(cid:48)(cid:105) in Fig. 3. Tunneling out of the double dot occurs from the (0,2) singlet state, which is coupled to the drain lead. Through hybridization, both singlet states S(cid:105) and S(cid:48)(cid:105) acquire finite lifetimes, reflected in their broadened lineshapes as shown in Fig. 3b. When only spin-conserving tunneling processes are taken into account, the triplet states remain decoupled from the drain. Therefore the rate-limiting step which controls current through this system is the decay of the long-lived triplet states through resonant hyperfine-assisted transitions to the singlet states S(cid:105) and S(cid:48)(cid:105), or higher order processes which may also break the conservation of spin within the double dot[6]. Hyperfine assisted transitions from T±(cid:105) to S(cid:105) and S(cid:48)(cid:105) transfer angular momentum from electron to nuclear spins, and thus drive the nuclear polarization dynamics. Here we focus on the regime of large detuning where the level S(cid:48)(cid:105) is far separated in energy from the triplet states and can be ignored in the calculation of hyperfine- assisted triplet-singlet transitions. We seek a coupled set of dynamical equations in two polarization variables. The first variable, x = (N+−N−)/(N+ +N−), represents the fractional polarization inside the double dot, where N+ (N−) is the number of nuclear spins oriented along (against) the external field. For a typical device, N ≡ N+ + N− ≈ 106. The second variable, which we denote by y, represents the fractional polarization within the tunnel barrier connecting the source lead to the first dot. T+\x{FFFF}T−\x{FFFF}T0\x{FFFF}S\x{FFFF}(0,2)S\x{FFFF}(1,1)S\x{FFFF}S+I−S\x{FFFF}\x{FFFF}εS\x{FFFF}S\x{FFFF}\x{FFFF}T+\x{FFFF}T−\x{FFFF}PotentialBiasγEa)b)g∗µBB+AxS−I+probabilities f± on the Overhauser field inhomogeneity in a lowest-order expansion in x and y: f± = 1 4 [1 ± η(x − y)], (5) where η controls the sensitivity of the loading probabili- ties to a polarization gradient. The factors of 2 in front of ΓD in Eqs. (2) and (3) account for the fact that polar- ization diffuses in both directions (up and down). Under what conditions might we expect to find oscilla- tions in the flow defined by Eqs. (2) and (3)? Typically, oscillations are found when the linearized system has the form of an "unstable spiral:" u = αu + v, v = −µu + βv, (6) with (α + β) > 0 and (α− β)2− 4µ < 0. These conditions ensure that the eigenvalues are complex, with positive real part. Comparing with Eq.(3), we see that y ∼ x, with a positive coefficient of x due to the fact that po- larization preserves its sign as it flows into the barrier. Therefore, we need the coefficient of y in Eq.(2) to be negative. Substituting expression (5) for f± into Eq.(2), we thus obtain a threshold η > 4ΓD/W0, where W0 is the hyperfine spin flip rate at the unstable fixed point. Going one step further, we can expand Eqs. (2) and (3) in the deviations x and y from the (unstable) fixed point of the nonlinear system. Notably, because y only appears to linear order in the original expressions, only y-independent or y-linear terms will show up in the ex- pansion. In general, all other terms will appear: y = ΓD x − 2ΓD y. x ≈ c10 x + c01 y + ··· , (7) where the dots represent higher order terms c20 x2 + c11 xy +c30 x3 +c21 x2 y +··· . Comparing to Eq.(6), we see that we need c10 > 2ΓD > 0 to ensure a positive real part of the eigenvalues, and c01 < 0, (c10 + 2ΓD)2 < 4c01ΓD to ensure a negative discriminant. These considerations lead to the oscillatory regime shown in Fig. 4. Using the vast separation of timescales between the hyperfine spin-flip driven polarization dynamics and the slow diffusion processes, we explore another avenue of analysis. We assume that the barrier polarization y is constant on the timescale of changes in the dot polariza- tion and examine the fixed points of the resulting quasi- one-dimensional dynamical system (2). Figure 4 shows the corresponding stable (green) and unstable (red) fixed points, superimposed on the velocity field map of the full system (arrows indicate direction of the polarization ve- locity, and the color scale indicates its magnitude). The oscillations can be seen as resulting from the combination of the circulatory flow around the origin, ensured by the spiral condition above, combined with the existence of the unstable branch of quasi-fixed-points near the origin. Fast horizontal motion towards the quasi-stable points, followed by slow drift along the green curves provides a 4 FIG. 4: Polarization velocity field in the oscillatory regime. Parameter values: ε/A = −3, γ/A = 0.05, B = 0, η = 0.4, ΓD/A = 10−11. Arrows indicate the direction of the velocity field ( x, y), while the color scale indicates the magnitude of the velocity (arbitrary units). The green and red curves indicate branches of stable and unstable fixed points of the quasi-one- dimensional dynamics (2) with y held constant. The blue curve shows the approximate trajectory of the limit cycle. pictorial representation of the oscillation mechanism out- lined in Fig. 1. The oscillation period is dominated by the length of the excursions along the green branches. As a result, the period grows as the oscillation amplitude in- creases, consistent with experiment[8]. As parameters vary, a transition out of the oscillation regime occurs when the unstable branch shrinks and disappears. We have identified a straightforward physical mecha- nism which can produce stable oscillations of dynami- cal nuclear polarization in spin-blockaded vertical double quantum dots. The mechanism relies on nuclear spin diffusion into a tunnel barrier and is active only in ver- tical DQDs. The dependence of spin-injection probabili- ties and spin diffusion times on barrier width provides a clear experimental signature of this mechanism. Persis- tent oscillations can serve as a new probe of nuclear spin diffusion and spin dynamics in vertical structures. We gratefully acknowledge helpful discussions with S. Amaha, D. G. Austing, and S. Tarucha. M. R. thanks the Institute for Quantum Optics and Quantum Information for their generous hospitality and support. [1] K. Ono, D. G. Austing, Y. Tokura, S. Tarucha, Science 297, 1313 (2002). [2] R. Hanson et al. Rev. Mod. Phys. 79, 1217 (2007). [3] W. A. Coish and D. Loss, Phys. Rev. B 72, 125337 (2005). [4] O. N. Jouravlev and Yu. V. Nazarov, Phys. Rev. Lett. 96, 176804 (2006). 0−5−10510BarrierPolarization,y(×10−2)DotPolarization,x(×10−2)02550−25−505 [5] D. Klauser, W. A. Coish, and D. Loss, Adv. in Solid State [16] D. G. Austing, C. Payette, G. Yu, and J. A. Gupta, Phys- Physics 46, 17, (2007). ica E 40, 1118 (2008). [6] F. Qassemi, W. A. Coish, and F. K. Wilhelm, Phys. Rev. Lett. 102, 176806 (2009). [7] J. Danon and Yu. V. Nazarov, Phys. Rev. B 80, [17] D. Paget, Phys. Rev. B 25, 4444 (1982). [18] D. J. Reilly et al. Phys. Rev. Lett. 101, 236803 (2008). [19] P. Stano and P. Jacquod, Phys. Rev. B 82, 125309 041301(R) (2009). (2010). [8] K. Ono and S. Tarucha, Phys. Rev. Lett. 92, 256803 [20] S. I. Erlingsson, O. N. Jouravlev, and Yu. V. Nazarov, (2004). Phys. Rev. B 72, 033301 (2005). [9] F. H. L. Koppens et al. Science 309, 1346 (2005). [21] J. S. Moodera, L. R. Kinder, T. M. Wong, and R. Meser- [10] J. Baugh, Y. Kitamura, K. Ono, and S. Tarucha, Phys. vey, Phys. Rev. Lett. 74, 3273 (1995). Rev. Lett. 99, 096804 (2007). [22] T. Miyazaki and N. Tezuka, J. Magn. Magn. Mater. 139, [11] M. S. Rudner and L. S. Levitov, Phys. Rev. Lett. 99, L231 (1995). 036602 (2007). [23] M. S. Rudner and E. I. Rashba, Phys. Rev. B 83, 073406 [12] J. Inarrea, G. Platero, and A. H. MacDonald, Phys. Rev. (2011). B 76, 085329 (2007). [13] J. Danon and Yu. V. Nazarov, Phys. Rev. Lett 100, 056603 (2008). [14] J. Danon et al. Phys. Rev. Lett. 103, 046601 (2009). [15] M. S. Rudner et al. Phys. Rev. B 84, 075339 (2011). [24] In fact diffusion is fastest in the vertical direction due to anisotropy of the quantum dot, thus making transfer to the barrier region even more unlikely. 

1202.4102

1

1202

Thermopower of Quantum Hall States in Corbino Geometry as a Measure of Quasiparticle Entropy

Using the Onsager relation between electric and heat transport coefficients, and considering the very different roles played by the quantum Hall condensate and quasiparticles in transport, we argue that near the center of a quantum Hall plateau thermopower in a Corbino geometry measures {\it "entropy per quasiparticle per quasiparticle charge"}. This relation indicates that thermopower measurement in a Corbino setup is a more direct measure of quasiparticle entropy than in a Hall bar. Treating disorder within the self-consistent Born approximation, we show through an explicit microscopic calculation that this relation holds on an integer quantum Hall plateau at low temperatures. Applying this to non-Abelian quantum Hall states, we argue that Corbino thermopower at sufficiently low temperature becomes temperature-independent, and measures the quantum dimension of non-Abelian quasiparticles that determines the topological entropy they carry.

cond-mat.mes-hall

cond-mat

Thermopower of Quantum Hall States in Corbino Geometry as a Measure of Quasiparticle Entropy Yafis Barlas1 and Kun Yang2 1Department of Physics and Astronomy, University of California, Riverside, CA 92521 2National High Magnetic Field Laboratory and Department of Physics, Florida State University, FL 32306, USA Using the Onsager relation between electric and heat transport coefficients, and considering the very different roles played by the quantum Hall condensate and quasiparticles in transport, we ar- gue that near the center of a quantum Hall plateau thermopower in a Corbino geometry measures "entropy per quasiparticle per quasiparticle charge". This relation indicates that thermopower mea- surement in a Corbino setup is a more direct measure of quasiparticle entropy than in a Hall bar. Treating disorder within the self-consistent Born approximation, we show through an explicit micro- scopic calculation that this relation holds on an integer quantum Hall plateau at low temperatures. Applying this to non-Abelian quantum Hall states, we argue that Corbino thermopower at suffi- ciently low temperature becomes temperature-independent, and measures the quantum dimension of non-Abelian quasiparticles that determines the topological entropy they carry. I. INTRODUCTION particles and decays exponentially with their separation) due to this degeneracy18: Candidate fractional quantum Hall (FQH) states which exhibit non-Abelian quasiparticle excitations have been predicted to appear in the second Landau level (LL) for certain filling fractions.1,2 Recent proposals for per- forming intrinsic fault-tolerant quantum computation us- ing non-Abelian anyons3,4 have revived interest at these filling factors, and in particular on the nature of their quasiparticles excitations. At present the most promis- ing candidate for non-Abelian FQH state (based on nu- merical studies5) seems to be ν = 5/2, which is thought to be described by the Moore-Read state1 or its particle- hole conjugate.6 Experimentally, the quasiparticle charge has been measured via tunneling between opposite edge states across quantum point contacts7,8 as well as local charge measurement,9 and found to be consistent with theoretical predictions expected for non-Abelian quasi- particles. In particular the observation of an "even-odd" effect10 alternating between e/4 and e/2 quasiparticles,11 as well as the recently reported phase slips12 in the in- terference pattern are suggestive of the non-Abelian na- ture of the quasiparticle excitations. However these ex- periments need to be reconciled with each other, which would require detailed understanding of all aspects of the experiments. Such an understanding requires careful analysis of possible complications due to edge reconstruc- tion,13 nonequilibrium edge distributions and coupling of the edge state to bulk quasiparticles.14 In light of possible complications associated with the edge, alternative approaches using bulk measurements to directly probe the non-Abelian nature of the quasiparti- cles have been proposed.15 -- 17 The basic idea behind them is the observation that in the presence of non-Abelian quasiparticles, the system has a ground state degeneracy Γ which grows exponentially with the number of quasi- particles Nq: Γ ∼ dNq where d > 1 is the quantum di- mension of the non-Abelian quasiparticle. This leads to a temperature independent entropy (except below an en- ergy scale which is related to the coupling between quasi- SD = kβ log Γ = kβNq log d + O(1). (1) This non-Abelian or topological entropy due to ground state degeneracy associated with the presence of non- Abelian quasiparticles, is an important part of the total entropy: Stot = SD + Sn(T ), (2) where Sn(T ) is the temperature dependent entropy due to normal sources. At sufficiently low temperatures Sn(T ) approaches zero and SD dominates Stot, thereby allowing experimentalists to measure the topological en- tropy using probes sensitive to entropy. In particular, it was pointed out in Ref. 15 that thermopower is one of the possible probes, because it measures entropy per mobile electron under suitable conditions. In fact, the topological entropy might already be a significant con- tributor to thermopower measured near 5/2 in a recent experiment.19 We note thermopower had been measured in the quantum Hall (QH) regime before that.20,21 A QH liquid can be viewed as a macroscopic conden- sate which has a gap for charged excitations; in the case of a FQH liquid these quasiparticle excitations carry frac- tional charge and possibly non-Abelian statistics. The QH condensate does not carry any entropy; hence all the entropy must be carried by the quasiparticles. In the presence of a thermal gradient the thermal response (which is proportional to the entropy) can only come from the entropy carried by the quasiparticles; this can be viewed as response to an "entropical force". In a thermopower measurement the application of a temper- ature gradient ∇T leads to an electric potential gradi- ent ∇V , as no current is allowed to flow through the system. The thermal response must be canceled by the electric response to voltage gradient so the net current is zero. In the Hall bar geometry where the thermopower measurement has been considered earlier,15 the electric 2 1 0 2 b e F 8 1 ] l l a h - s e m . t a m - d n o c [ 1 v 2 0 1 4 . 2 0 2 1 : v i X r a 2 simply runs around the annulus. The crucial point here is that the longitudinal current (from both thermal and voltage gradients) comes from quasiparticles only, with no condensate or edge state contribution. As a result dis- order affects the thermal and electric response on equal footing, therefore it does not suppress thermopower even in the presence of localization effects (because we are always working at non-zero temperature, quasiparticles can still hop even if they are localized at T = 0). An- other way to understand this heuristically is that since the quasiparticle current is zero, there is no pinning force on them, thus the electric and entropical forces must be balanced, even in the presence of disorder. The difference between thermopower in Hall bar and Corbino geometries is best illustrated by analyzing the transport equations in the presence of an electric field and a temperature gradient: j = L11∇φ − L12∇T jQ = L21∇φ − L22∇T T T , , (3) (4) where j and jQ are the respective charge and heat cur- rents, ∇φ is the applied electric field, T is the temper- ature and Lαβ are the response coefficients which are tensor quantities in the presence of a magnetic field. The thermopower in a Hall bar geometry, which is given by the zero-current condition in both the transverse and lon- gitudinal directions (j = 0), is also a tensor quantity: QH = L12(L11)−1; 1 T it relates ∇φ with ∇T through ∇φ = QH∇T. (5) (6) FIG. 1. (Color online) Schematic representation of current responses to thermal and voltage gradients in an infinite Hall bar setup in (a) clean and (b) disordered limits. The thermal response is due to entropy carried by quasiparticles, while the voltage response is dominated by the quantum Hall conden- sate. In (a) clean limit both current responses are in trans- verse (Hall) channel. A thermal gradient applied from the left to right leads to a corresponding voltage difference along the same direction. The current induced by the thermal gra- dient (red arrow), is canceled exactly by the voltage-induced current dominated by condensate flow (blue arrow), resulting in zero net current flowing through the sample. The equilib- rium edge current is represented by the thick black arrows. In the disordered limit (b), thermal gradient induced current is strongly suppressed by pinning or localization of quasipar- ticles, leading to a corresponding suppression in the compen- sating condensate current induced by voltage gradient. As a result the voltage response is also suppressed. In the presence of disorder the thermal gradient induced current includes both longitudinal and Hall components, and the voltage gradient is no longer parallel to thermal gradient. response is dominated by the QH condensate. As a re- sult the thermal and electric responses come from very different sources. This difference may lead to substan- tial suppression of the thermopower signal when disor- der is present, as disorder can have a strong effect on the quasiparticles, potentially leading to their localiza- tion and thus suppressing their response to a thermal gradient. On the other hand disorder has virtually no ef- fect on the condensate; a very weak voltage gradient can induce sufficient current response to compensate for the thermal gradient. For this reason the quasiparticle en- tropy is detectable in Hall bar thermopower measurement only when they are mobile, which requires somewhat el- evated temperatures15 (see Fig. 1 for an illustration of this point). The purpose of this paper is to point out that ther- mopower in a Corbino setup21 (see Fig. 2) is a more direct and thus much better measure of quasiparticle en- tropy. This is because in the Corbino geometry the zero- current condition that results from a cancelation of the thermal and voltage responses only applies to the lon- gitudinal current (or current along the radial direction); there is no constraint on the Hall channel current, which On the other hand in a Corbino geometry setup a tem- perature gradient applied in the radial direction leads to a voltage gradient in the radial direction only. The ther- mopower which is defined by the zero-current condition in the radial direction is then given as QC = 1 T L12 rr L11 rr , (7) which reduces to a number that depends on the longitu- dinal components of the electric (L11 rr) and thermal (L12 rr) responses only. These longitudinal components are dom- inated by quasiparticles, and are equally affected by the presence of disorder. The condensate current, which only flows in the Hall channel is in the angular direction and orthogonal to the radial component, therefore does not contribute to the thermopower measurement. The central result of this paper is that the Corbino thermopower is (approximately) equal to "entropy per quasiparticle per quasiparticle charge": QC ≈ Stot e(cid:63)Nq , (8) (cid:68)V(cid:68)Ta
(cid:68)V(cid:162)(cid:68)Tb
II. CORBINO THERMOPOWER 3 We consider a QH plateau at the total filling factor ν in a Corbino geometry setup. At zero temperature for a sufficiently clean and uniform sample, quasiparti- cles are absent at the center of the QH plateau and the system only consists of the QH condensate. As the mag- netic field B is decreased (increased) from its value at the center of the QH plateau B0, quasielectrons (quasiholes) are introduced even at zero temperature, whose numbers grow linearly with the deviation of the magnetic field ∆B = B − B0: (10) (cid:12)(cid:12)(cid:12)(cid:12)(cid:18) e e(cid:63) Nq = (cid:19) ∆B B0 (cid:12)(cid:12)(cid:12)(cid:12)Ne, where Ne is the number of electrons. Additional quasi- particles or quasiholes are thermally activated at finite temperature; for sufficiently low temperature one type dominates the other (depending on which side of the plateau one is at). In the following we assume that this is the case and consider the contribution to transport by either quasiparticles or quasiholes only. We can split the total charge current j = jq +jcond, into a quasiparticle current jq and condensate current jcond. The latter only flows in the Hall channel and gives rise to quantized Hall transport without dissipation. If the temperature of the inner and outer radii in a Corbino thermopower measurement is fixed so that there is no thermal gradient, then under the application of a radial electric gradient the heat current jQ, which is carried by quasiparticles only, will be proportional to the quasipar- ticle charge current. We can thus write jQ = Πjq = Π(e(cid:63)jq n), (11) where jq n denotes the quasiparticle number current. The above equation (11) is similar to the definition of Peltier heat which normally relates the heat current to total cur- rent. In our case Π relates the heat current to quasiparti- cle current only and hence corresponds to the heat carried by each quasiparticle. To proceed further we subtract the QH condensate contribution from the response coefficient of the electric field. As already mentioned in the previ- ous section, only the Hall channel has contributions from both the condensate and quasiparticles, therefore we can separate out the quasiparticle contribution to the Hall conductivity L11 xy via, L11 rθ = L11 rθ − νe2 h , (12) whereas the longitudinal electrical conductivity L11 rr = L11 rr, along with the longitudinal and off-diagonal ther- mal responses L12 rθ, are completely dominated by the quasiparticles. In terms of the quasi- particle transport coefficients, Π can be expressed as rr and L12 rθ = L12 rr = L12 Π = L21( L11)−1. (13) FIG. 2. (Color online) Schematic representation of ther- mopower measurement in a Corbino disk geometry with disor- der. A thermal gradient applied in the radial direction leads to a corresponding voltage difference, with no net current flow- ing along the same direction; there is no constraint on the circular current flowing along the angular direction. Since the condensate current (represented by the black line, including edge current) is always in the Hall channel and thus flowing along the angular direction, it plays no role in the current balance along the radial direction. As a result the zero radial current condition must be satisfied by balancing quasiparticle current responses to both thermal and voltage gradients, and thus dictated by quasiparticle properties only. Since the net quasiparticle current is zero, disorder plays a much lesser role here than in Hall geometry. where e(cid:63) is the quasiparticle charge, which is opposite for electron and hole like quasiparticles. Eq. (8) can be understood heuristically as the consequence of balancing the electric force due to electric field with the "entropical force" due to the thermal gradient experienced by the quasiparticle: e(cid:63)∇φ = ∇T. Stot Nq (9) This is valid when the quasiparticle density is sufficiently low. Since the quasiparticles are not flowing, the above is expected to be valid even in the presence of disorder, and when the quasiparticles are localized at T = 0. In the following sections we explore these ideas in greater detail and use various arguments to establish Eq. (8). By applying Onsager relations in the analysis of transport equations (3) and (4) for the Corbino geome- try, we show in section II that in the dilute quasiparticle regime Eq. (8) is valid. Since the analysis is quite gen- eral and only based on the behavior of the quasiparticle contribution to the transport coefficients it should ap- ply equally for the integer and fractional QH effects. In section III we establish Eq. (8) for a specific model of non-interacting electrons in the integer QH regime and presence of disorder through microscopic calculation. In section IV we explore the possibility of using Eq. (8) to probe the topological entropy carried by the non-Abelian quasiparticles, and measure their quantum dimension d. T0(cid:68)T(cid:43)T0(cid:68)V4 Also due to the fact that on the plateau the quantized Hall conductance due to QH condensate dominates the quasiparticle contributions to the conductivity tensor, the latter is proportional to the quasiparticle contribu- tions to the resistivity tensor, justifying Eq. (16). In the next section we lend further support to Eq. (8) by performing a microscopic calculation for Corbino thermopower treating disorder in the self consistent Born approximation (SCBA) on an integer QH plateau. III. CORBINO THERMOPOWER: IQHE In section III A. we calculate the thermopower on an integer quantum plateau in a Corbino geometry with dis- order treated within SBCA. In section III B. show that for low temperatures the Corbino thermpower scales like the "entropy per quasiparticle per quasiparticle charge". A. Corbino Thermopower on an IQH plateau The electric field response L11 rr can be calculated using the Kubo formula. We then use the relation26 (cid:90) ∞ −∞ (cid:18) (cid:19)
− µ e d
− ∂nF ∂
L12 ij (T, µ) = L11 ij (T = 0,
), (21) (22) From the definition (11) it is clear that Π corresponds to the heat carried by each quasiparticle divided by its charge. While in principle it is a tensor, we expect it to be very close to (if not exactly) a pure number (or scalar), since jQ and jq should be parallel to each other on physi- cal ground. We thus focus on its diagonal component Πrr in the following. Using the fact Π(B) = Π(−B) and by virtue of the Onsager relations,22 which can be expressed for the quasiparticle contribution to the transport coeffi- cients, Lαβ ab (B) = Lβα ba (−B), (14) the diagonal component of the Π in the Corbino geometry setup can be expressed as Πrr = L11 rθ ( L11 L21 θr + L11 rθ)2 + ( L11 L21 rr rr rr)2 = L11 rθ ( L11 L12 θr + L11 rθ)2 + ( L11 L12 rr rr rr)2 . (15) Close to the center of the QH plateau the quasiparticle contributions to the transport coefficients are expected rr → 0 and to behave as those of a Hall insulator23: L11 rθ ∝ ( L11 L11 rr)2, which leads to rθ (cid:28) L11 L11 rr. (16) Therefore, in the dilute quasiparticle regime or close to the center of the QH plateau the off-diagonal component vanishes much faster than the longitudinal component, we can thus neglect terms involving L11 (15), which reduces to rθ in Eq. Πrr = L12 rr L11 rr = L12 rr L11 rr = T QC. (17) The above relation gives an interpretation of ther- mopower in the Corbino geometry QC = Heat per quasiparticle e(cid:63)T = Stot e(cid:63)Nq , (18) which is the "entropy per quasiparticle per quasiparticle charge", as advertised in Eq. (8). Notice that in the above derivation the condition for a Hall insulator is not essential; to satisfy Eq. (8) we only require that close to the center of the QH plateau the quasiparticle contri- bution to off-diagonal conductivity is much smaller than the diagonal conductivity. This condition on the conduc- tivity (Eq. 16)can be empirically justified for the case of QH plateaus by examining the experimentally observed relation between the Hall and longitudinal resistivity,24,25 Rrr = αrB dRrθ dB = αrB d∆Rrθ dB (19) where αr is an order 1 constant independent of the mag- netic field B, and ∆Rrθ is the deviation of Hall resistivity from the quantized plateau value (due to quasiparticle contribution). Since d∆Rrθ dB ∼ ∆Rrθ ∆B , we have αr∆Rrθ (cid:29) ∆Rrθ. Rrr ∼ B0 ∆B to calculate the thermal response coefficient L12 rr. In Eq. 21, nF represents the Fermi-Dirac distribution and µ is the chemical potential. This above relationship, which holds in a magnetic field, was also shown to be valid in the presence of disorder in Ref. 26, where both the off- diagonal and diagonal conductivity on an IQH plateau were calculated treating disorder within SCBA. The lon- gitudinal electrical conductivity is (cid:90) ∞ (cid:18) d
−∞ − ∂nF ∂
(cid:19) L11 rr = e2ω2 c 8πh (cid:88) [(n + 1)ImGn(
)ImGn+1(
)], n where n corresponds to the Landau level (LL) index, ωc = eB/m is the cyclotron frequency and Gn(
) is the SCBA dressed Green's function for the nth-LL given by Gn(z) = 1 z − ωn − Σ(z) , (23) where ωn = (n + 1/2)ωc. Assuming a random impurity potential V with a white noise distribution (cid:104)V (r)V (r(cid:48))(cid:105)av = 2πl2 2σ2δ(r− r(cid:48)) the SCBA self-energy Σ in the high field approximation (σ << ωc) can be writ- ten as B (cid:115) (cid:18) ω − ωL (cid:19)2 2σ ω − ωL 2 − iσ 1 − (20) Σ(ω + iδ) = , (24) 5 FIG. 3. (Color on-line) Disorder broadened density of states for the isolated lowest Landau level within self-consistent Born approximation. The red (vertical) line represents the position of the lowest Landau level energy, which is the tran- sition point between two integer quantum Hall phases, and separates the "quasi-electrons" and "quasi-holes" regions. Re- gions I and II are defined and discussed later in the paper (see text for details). FIG. 4. (Color on-line) Corbino thermopower measured in units of µV /K for the isolated lowest Landau level plotted as a function of (
−0.5ωc)/(σ), for different values of σ/kβT = 0.25, 0.5, 0.75 and 1. where ωL is the energy of the LL nearest to
and σ is a measure of the white noise disorder potential. The disorder broadened density of states (DOS) is given as (cid:115) (cid:18)
− ωL (cid:19)2 2σ ρn(
) = 1 2π2l2 B σ 1 − , (25) is semi-elliptical with a semi-minor axis 2σ. The dis- order broadened LL density of states for the lowest LL is plotted in Fig. 3 which also indicates the quasiparti- cle/quasihole regimes. The expression (22) can be used in (21) to calculate the longitudinal thermal response L12 rr which can then be used to calculate the Corbino thermopower given by (7). For an isolated LL, it can be expressed as a two- paramter scaling function of (µ− (n + 1/2)ωc)/(σ) and σ/kβT with the scaling function plotted for different values of σ/kβT in Fig. 4. In order to better under- stand features the Corbino thermopower of an isolated LL it is best to orient oneself and understand the dis- order broadened density of states (DOS) which in the SCBA is semi-elliptical with a semi-minor axis 2σ as indicated in Fig. 3. Away from the tails of the disorder broadened DOS the thermopower indicates divergent be- havior as a function of the temperature T , reminiscent of an insulator. This behavior is anticipated away from the tails of the disorder broadened DOS and near the center of the QH plateau where quasielectrons (quasiholes) are thermally activated and the thermopower should naively correspond to the entropy per quasiparticle per quasipar- ticle charge. Furthermore, the Corbino themopower changes sign at the position of the LL where the disorder broadened DOS is maximized. This sign change is due to the fact that transport switches from being electron-dominated to hole-dominated regimes, as a result the charge of the relevant quasiparticles change sign. The overall features of the Corbino thermopower of an isolated LL already indicates behavior anticipated by the "entropy per quasi- particle per quasiparticle charge" definition. We plot in Fig. 5 Corbino thermopower with con- tributions from all LLs taken into account, as a func- tion of the chemical potential for fixed disorder strength σ = 0.05ωc at different temperatures. As expected, we see a periodic behavior from one LL to another, fur- ther justifying the absence of the QH condensate contri- bution. The sign change in the thermopower when the chemical potential in half-way between two neighboring LLs, which corresponds to the center of a QH plateau, has a similar origin to the sign change at the QH transi- tion points discussed above: This is due to the fact that transport to the right of the center of the QH plateau is dominated by thermally activated electrons in the higher LL, and holes to the left in the lower LL. As mentioned earlier in this regime the quasiparticle contribution to the transport coefficients exhibit Hall insulating behav- ior and as we show in the next section the thermopower for low temperatures is equal to the "entropy per quasi- particle per quasiparticle charge". In the next section we establish that this relation is exact at low temperatures near the center of the an integer QH plateau. IIIIIIQuasi(cid:45)electronsQuasi(cid:45)holes(cid:45)4(cid:45)2024Ε(cid:45)0.5(cid:209)Ωc(cid:209)ΣΡn0.250.50.751(cid:45)4(cid:45)2024(cid:45)1000(cid:45)50005001000Ε(cid:45)0.5(cid:209)Ωc(cid:209)ΣQCas q (µ) = (−1)α Nφ N α πσ (cid:90) µα −2σα+ωc/2 (cid:115) d
1 − (cid:18)
− 1/2ωc 2σ 6 (cid:19)2 (28) with µ+ ≤ ωc/2 (electron dominating) and µ− ≥ ωc (hole dominating), and with our definition we always have N α q > 0. The thermally activated quasiparticle can then be expressed as (cid:90) ∞ −∞ N α q (T, µ) = d
∂nF ∂
N α q (µ). (29) FIG. 5. (Color on-line) Corbino thermopower calculated for fixed disorder σ = 0.05 in units of ωc at different tempera- tures kβT = 0.0125, 0.025, 0.375, and 0.05. The thermopower QC is measured in µ V/K and the chemical potential
is in units of ωc. B. Corbino thermopower and entropy per quasiparticle Next we analyze the entropy per quasiparticle per quasiparticle charge and its relation to the Corbino ther- mopower. The entropy is the partial derivative of the grand canonical potential Ω with respect to the temper- ature T , S = −∂Ω/∂T . The grand canonical potential in the presence of the disorder broadened LL DOS for an isolated LL can be written as (cid:90) 2σ+ωn (cid:88) −2σ+ωn n Ω = −kBT Nφ πσ d
log(1 + eβ(µ−
))ρn(
), (26) where Nφ = A/2πl2 B is the degeneracy of a single LL. In the following we concentrate on a single LL only, ne- glecting the effects of LL mixing. This assumption can be formally justified by working the in the high field ap- proximation ωc >> σ. In this limit the entropy is a periodic function of the chemical potential µ. The to- tal entropy for an isolated LL S = S+ + S− can be split into contributions from quasielectrons and quasiholes Sα, α = +(−) for quasielectrons (quasiholes). It is advan- tageous for what follows to express Sα in terms of the number of quasielectrons (quasiholes) N α q as (
− µ) kβT 2 Sα(T, µ) = (−1)α+1 (cid:90) ωc/2 ∂nF ∂
−2σα+ωc/2 q (
), N α d
(27) where we have used the fact that ρn(
) = ∂Nq(
)/∂
and performed an integration by parts on Eq. (26), and then taken the derivative with respect to the tempera- ture. The number of quasiparticles N α q can be expressed To compare the Corbino thermopower calculated in the previous section to the "entropy per quasiparticle per quasiparticle charge" in the low temperature limit we start by identifying two different regions as shown in Fig 3. For the sake of clarity we now restrict our analysis to quasielectrons only, since the analysis for quasiholes is similar. We relax this constraint in our final answers which are general and apply equally to both types of quasiparticles. Regions I and II for quasielectrons are given as 0 < µ < −2σ + ωc/2 and −2σ + ωc/2 < µ < ωc/2 respectively(see Fig. In region I the quasiparticles are purely thermally activated and obey Boltzmann statistics, the diagonal response functions ex- hibit behavior similar to an insulator and hence divergent thermpower (see Fig. 4). In contrast, region II contains a finite density of quasielectrons even at T = 0 and the thermopower depends in particular on the details of dis- order strength and the chemical potential. For region I working in the low temperature limit kβT (cid:28) −2σ − µ + ωc/2 the quasiparticles obey Boltz- mann statistics and one can replace the derivative of the Fermi function ∂nF /∂
in Eqs. (21), (22) and (27) by eβ(
−µ). In this limit the entropy can be expressed as (a similar situation arises for quasiholes), 3). (cid:20) (cid:90) δ (cid:21) 0 d
e−β
N + eβ(µ+2σ−1/2) S+ = − ∂ ∂T q (T = 0,
→ 0) (30) where δ is a small region where the derivative of the Fermi function overlaps with the LL broadened density of states q (T = 0,
→ 0) = and with leading order behavior for N + 2/3(x/σ)3/2. With the same approximation for Eq. 29 the "entropy per quasiparticle per quasiparticle charge" in Region I can be expressed as, (cid:20) (cid:18) (cid:19)(cid:21) Sα eN α q = (−1)α k2 βT (cid:90) δ/(kβ T ) e log ∂ ∂T due−uu3/2 eβ(µ+2σα−1/2) . (31) 0 Similarly one can use the same approximation to eval- uate Eqs. (21) and (22) in Region I giving the Corbino 0.01250.0250.03750.0501234(cid:45)1500(cid:45)1000(cid:45)500050010001500ΕQC7 quasielectron and quasihole becomes ambiguous. This is evident in Fig. 6 where the numerically attained ratio of entropy per quasiparticle per quasiparticle charge S/(eNq) and Corbino thermopower QC is plotted as a function of the chemical potential µ. Fig. 6 also indicates the two are equal in Region I which is near the center of the QH plateau. In Region II the two are no longer equal; however they have the same (linear) T dependence and hence a constant ratio in the low T limit. IV. APPLICATION TO NON-ABELIAN QUANTUM HALL STATES It is clear that Eq. (8) can be used to probe entropy carried by non-Abelian quasiparticles in non-Abelian QH liquids, which is dominated by the topological entropy SD at low temperature. The best place to do this is near the center of the QH plateau, where the quasiparticle density is low. This is opposite to the case of Hall bar geometry,15 in which case thermopower measures total entropy carried by the quasiparticles (divided by number of electrons), one thus needs to be near the edge of QH plateau to have higher quasiparticle density and thus en- tropy. Here since it is entropy per quasiparticle that is measured, the low-density regime is preferable. There are several advantages of working near the plateau center, where the physics in general is simpler. Among them, we mention: (i) We do not need to worry about the residue coupling among quasiparticles that can lift the ground state degeneracy, as they decay exponentially with quasi- particle separation. (ii) We do not need to worry about competing states that may appear (possibly in parts of the sample), which can carry substantial entropy. Quantitatively, we expect Corbino thermopower to sat- urate to a finite value in the low temperature limit that depends on the quasiparticle's quantum dimension: QC(T → 0) = (kβ/e(cid:63)) log d. (36) For the specific case of 5/2, we expect e(cid:63) = e/4 and d = QC(T → 0) = (4kβ/e) log 2 ≈ 1.2 × 10−4V /K, (37) 2,1 thus √ √ which is at least one order of magnitude larger than what has been observed in the FQH regime below 0.1K. In the Corbino geometry, the sign of QC changes when one moves through the plateau center, as one goes from the quasiparticle dominated to quasihole dominated regime. To approach the saturation value of Eq. (36), we need the temperature to be sufficiently low such that Sn(T ) (cid:28) SD. Assuming that the quasiparticles form a Wigner crystal in a completely clean sample, this happens for T (cid:28) TD, where the characteristic temperature TD is the Debye temperature of the Wigner crystal, estimated in Ref. 15 [see its Eq. (13)]. We note this would be a lower bound for realistic samples, as disorder can pin the Wigner crystal FIG. 6. Numerical attained ratio of the entropy per quasipar- ticle per quasiparticle charge S/(eNq) and the Corbino ther- mopower QC for the quasi-electron region of an isolated lowest Landau level for kβT /σ = 0.01. thermopower near the center of the QH plateau, (cid:20) (cid:18) (cid:19)(cid:21) log C = (−1)α k2 βT (cid:90) δ/(kβ T ) Qα e ∂ ∂T due−uu eβ(µ+2σα−1/2) 0 (32) where α = +(−) gives the Corbino thermopower expres- sion to the right (left) of the center of the QH plateau in the low temperature regime. This gives an explicit verifi- cation of the interpretation of the Corbino thermpower as the "entropy per quasiparticle per quasiparticle charge" Qα C = Sα eN α q = α k2 β e (µ + 2σα − 1/2) T + ··· . (33) In the opposite regime (region II) where the quasipar- ticles resemble a degenerate Fermi gas for kBT (cid:28) σ, we recover a Mott like expression for the Corbino ther- mopower, . (cid:12)(cid:12)(cid:12)(cid:12)
=µ (cid:12)(cid:12)(cid:12)(cid:12)
=µ QC = α k2 β e T π2 3 1 L11 xx dL11 xx d
(34) Similarly the entropy per quasiparticle per quasiparticle charge be expressed as S+ eN + = α k2 β e T π2 3 1 N + q dN + q d
. (35) In the latter case while we do not have an exact equality, we still find QC scales like the entropy per quasiparticle per quasiparticle charge with the constant of proportionality depending on the specific model for disorder, and other details. The deviation from equality gets progressively worse as the chemical potential ap- proaches the LL energies, where the distinction between (cid:45)5(cid:45)4(cid:45)3(cid:45)2(cid:45)1001234Ε(cid:45)0.5(cid:209)Ωc(cid:209)ΣS(cid:43)eQCNqp(or perhaps individual quasiparticles), and open gaps in the magnetophonon spectra; this tends to suppress Sn(T ) at low temperature, and pushes the saturation tempera- ture higher. One caveat to keep in mind is for samples with some inhomogeneity, there are preexisting quasiparticles and quasiholes (with equal average density) at the center of the QH plateau, and their contributions to QC would cancel due to their opposite charge. In that case one does need to move away from the plateau center, such that the number of quasiparticles or quasiholes induced by the deviation dominates the preexisting ones. This may put some stringent constraints on the sample quality for the observability of the behavior indicated in Eq. (36). 8 ACKNOWLEDGMENTS One of us (KY) thanks R. R. Du for a useful correspon- dence that motivated the present work. This work was supported in part by NSF grant DMR-1004545 (KY). 1 G. Moore and N. Read, Nucl. Phys. B 360, 362 (1991); M. Greiter, X. G. Wen and F. Wilczek, Phys. Rev. Lett. 66, 3205 (1991), 2 N. Read and E. Rezayi, Phys. Rev. B 59, 8084 (1991). 3 A. Kitaev, Ann. Phys. (Leipzig) 303, 2 (2003). 4 C. Nayak, S. H. Simon, A. Stern, M Freedman and S. D. Sarma, Rev. Mod. Phys. 80, 1083 (2008). 5 R. H. Morf, Phys. Rev. Lett. 80, 1505 (1998); E. H. Rezayi and F. D. M. Haldane, Phys. Rev. Lett. 84, 4685 (2000); X. Wan, K. Yang, and E. H. Rezayi, Phys. Rev. Lett. 97, 256804 (2006); X. Wan, Z.-X. Hu, E. H. Rezayi, and K. Yang, Phys. Rev. B 77, 165316 (2008); A. E. Feiguin, E. Rezayi, Kun Yang, C. Nayak, and S. Das Sarma, Phys. Rev. B 79, 115322 (2009); M.R. Peterson and S. Das Sarma, Phys. Rev. B 78, 155308 (2008); M.R. Peterson, Th. Jolicoeur, and S. Das Sarma, Phys. Rev. Lett. 101, 016807 (2008); Gunnar Moller and Steven H. Simon, Phys. Rev. B 77, 075319 (2008); H. Wang, D. N. Sheng, and F. D. M. Haldane, Phys. Rev. B 80, 241311 (2009); M. Storni, R. H. Morf, and S. Das Sarma, Phys. Rev. Lett. 104, 076803 (2010) . 6 S.-S. Lee, S. Ryu, C. Nayak, and M. P. A. Fisher, Phys. Rev. Lett. 99, 236807 (2007); M. Levin, B. I. Halperin, and B. Rosenow, Phys. Rev. Lett. 99, 236806 (2007). 7 M. Dolev, M. Heiblum, V. Umansky, A. Stern and D. Ma- halu, Nature 452, 829 (2008); I. P. Radu, J. B. Miller, C. M. Marcus, M. A. Kastner, L. N. Pfeiffer and K. W. West, Science 320, 899 (2008); X. Lin, C. Dillard, M. A. Kastner, L. N. Pfeiffer, and K. W. West, arXiv:1201.3648. 8 R.L. Willet, L. N. Pfeiffer and K. W. West, Proc. Natl. Acad. Sci. 106 8853 (2009). 9 Vivek Venkatachalam, Amir Yacoby, Loren Pfeiffer, and Ken West, Nature (London) 469, 185 (2011). 10 A. Stern and B. I. Halperin, Phys. Rev. Lett. 96, 061802 (2006); P. Bonderson, A. Kitaev, and K. Shtengel, Phys. Rev. Lett. 96, 061803 (2006). 11 R.L. Willet, L. N. Pfeiffer and K. W. West, Phys. Rev. B. 82, 205301 (2010). 12 Sanghun An, P. Jiang, H. Choi, W. Kang, S. H. Si- mon, L. N. Pfeiffer, K. W. West, and K. W. Baldwin, arXiv:1112.3400. 13 X. Wan, K. Yang, and E. H. Rezayi, Phys. Rev. Lett. 88, 056802 (2002); K. Yang, Phys. Rev. Lett. 91, 036802 (2003); X. Wan, E. H. Rezayi, and K. Yang, Phys. Rev. B 68, 125307 (2003); S. Jolad and J. K. Jain, Phys. Rev. Lett. 102, 116801 (2009); S. Jolad, D. Sen, and J. K. Jain, Phys. Rev. B 82, 075315 (2010). 14 B. Rosenow, B. I. Halperin, S. Simon, and A. Stern, Phys. Rev. B 80, 155305 (2009); W. Bishara, and C. Nayak, Phys. Rev. B 80, 155304 (2009). 15 K. Yang and B. I. Halperin, Phys. Rev. B 79, 115317 (2009). 16 N. R. Cooper and A. Stern, Phys. Rev. Lett. 102, 176807 (2009). 17 G. Gervais and K. Yang, Phys. Rev. Lett. 105, 086801 (2010). 18 K. Yang, arXiv:0807.3341v1. 19 W. E. Chickering, J. P. Eisenstein, L. N. Pfeiffer and K. W. West, Phys. Rev. B 81, 245319 (2010). 20 H. Obloh, K. von Klitzing and K. Ploog, Surf. Sci. 170, 292 (1986); R. Fletcher, M. D'Iorio, A. S. Sachrajda, R. Stoner, C. T. Foxon and J. J. Harris, Phys. Rev. B 37, 3137 (1988); C. Ruf, H. Obloh, B. Junge, E. Gmelin, K. Ploog and G. Weimann, ibid 37 6377 (1988); U. Zeitler, J. C. Maan, P. Wyder, R. Fletcher, C. T. Foxon and J. J. Har- ris, ibid 47, 16008 (1993); V. Bayot, X. Ying, M. B. Santos and M. Shayegan, Europhys. Lett. 25, 613 (1994); X. Ying, V. Bayot, M. B. Santos and M. Shayegan, Phys. Rev. B 50, 4969 (1994); V. Bayot, E. Grivei, H. C. Manoharan, X. Ying, and M. Shayegan, Phys. Rev. B 52, 8621 (1995); B. Tieke, U. Zeitler, R. Fletcher, S. A. J. Wiegers, A. K. Geim, J. C. Maan, and M. Henini, Phys. Rev. Lett. 76, 3630 (1996); Jian Zhang, S. K. Lyo, R. R. Du, J. A. Sim- mons, and J. L. Reno, Phys. Rev. Lett. 92, 156802 (2004). 21 H. van Zalinge, R. W. van der Heijden, and J. H. Wolter, Phys. Rev. B 67, 165311 (2003). 22 L. Onsager, Phys. Rev. 37, 405 (1931); Phys. Rev. 38, 2265 (1931). 23 S. Kivelson, D. H. Lee and S. C. Zhang, Phys. Rev. B 46, 2223 (1992). 24 H. L. Stromer et. al., Solid State Comm. 84, 95 (1992); see also T. Rotger et. al., Phys. Rev. Lett. 62, 90 (1992). 25 S. H. Simon and B. I. Halperin, Phys. Rev. Lett. 73, 3278 (1994). 26 M. Jonson and S. M. Girvin, Phys. Rev. B 29, 1939 (1984). 

1001.2201

1

1001

Strong and weak coupling of two coupled qubits

I investigate the dynamics and power spectrum of two coupled qubits (two-level systems) under incoherent continuous pump and dissipation. New regimes of strong coupling are identified, that are due to additional paths of coherence flow in the system. Dressed states are reconstructed even in the regime of strong decoherence. The results are analytical and offer an exact description of strong-coupling in presence of pumping and decay in a nontrivial (nonlinear) system.

cond-mat.mes-hall

cond-mat

APS/123-QED Strong and weak coupling of two coupled qubits School of Physics and Astronomy, University of Southampton, SO17 1BJ, Southampton, United Kingdom Elena del Valle∗ (Dated: October 3, 2018) I investigate the dynamics and power spectrum of two coupled qubits (two-level systems) under incoherent continuous pump and dissipation. New regimes of strong coupling are identified, that are due to additional paths of coherence flow in the system. Dressed states are reconstructed even in the regime of strong decoherence. The results are analytical and offer an exact description of strong-coupling in presence of pumping and decay in a nontrivial (nonlinear) system. I. INTRODUCTION Quantum information processing [1] presupposes a coher- ent coupling of the fundamental bricks of quantum informa- tion, the qubits. In real systems, however, decoherence, dissi- pation and incoherent coupling to the environment is unavoid- able [2]. In the field of cavity quantum electrodynamics [3], the notion of coherent coupling is known as strong coupling, in this case, between light and matter (e.g., the excited state of an atomic transition [4] or an electron-hole pair in a semicon- ductor [5]). This leads to a quantum superposition of the bare states, resulting in so-called dressed states [6]. This regime is reached in systems of very high quality and under tight exper- imental control, so that intrinsic sources of decoherence are minimized as much as possible and coherent dynamics takes over. The simplest description of strong coupling neglects dis- sipation altogether and thus reduces to that of mere coupling with strength g, introducing the notion of Rabi splitting [7]. Next step in the description includes the decay g i of the bare states, i = 1, 2 [8]. This gives rise to a widely known criterion for strong-coupling: 4g > g 1 − g 2. This is the case of vac- uum Rabi splitting where at most one excitation is involved. At this level, which is the most natural and fundamental since it describes one particle, there is no difference from the under- lying theoretical model. Differences appear at the next step of description when the excitation scheme is taken into ac- count. A typical description of excitation is to consider an initial condition. The coupling is then studied as the sponta- neous emission from this initial excited state. Unless the ini- tial condition is restricted to one excitation (as previously), the underlying theoretical model becomes determinant. Another important description of the excitation process is that of a con- tinuous pumping, for instance a coherent excitation that drives the system [9, 10], or an incoherent pump that feeds excita- tions at a given rate but without any coherent input [11, 12]. The latter is more directly related to the intrinsic dynamics of the system, and is the one that will be considered in this text. With non-negligible pumping, the underlying theoretical de- scription cannot be ignored, and taking it into account leads to strong deviations from the paradigm of strong-coupling as established by the spontaneous emission of one excitation (in any model) [13]. ∗Electronic address: elena.delvalle.reboul@gmail.com FIG. 1: (Color online) Schema of the systems of interest in this text. The main object of study, two coupled qubits, is sketched in (c). It will be compared throughout with other coupled systems, (a), (b), (d) and (e). An immediate extension of light-matter coupling in the linear regime is the linear model, namely, that of two har- monic oscillators with no restriction in the number of parti- cles Fig. 1(a). Its comprehensive description under incoher- ent pumping was given in Ref. [14]. This describes for in- stance exciton-polaritons in planar semiconductor microcavi- ties [5] (where both excitons and photons are bosons). It was shown in this work how pumping calls for extended defini- tions of strong-coupling, essentially requiring absorption of the pumping rates in the decay rates. A more important the- oretical model, known as the Jaynes-Cummings model [15], describes the coupling of a two-level system (such as an atom or a zero-dimensional exciton in a small quantum dot) with a boson mode (typically, cavity photons), Fig. 1(b). It is more important because more closely related to a genuinely quan- tum regime, the linear model being essentially a classical de- scription cast in quantum-mechanical terms. Its description under incoherent pumping was given in Ref. [16], but en- countered various difficulties to offer a complete picture. In particular, dressed modes exhibit complex patterns and a defi- nition of strong-coupling in this system is much more difficult to achieve, since splitting of the dressed states depends on the excitations. In particular, strong-coupling can be enforced by pumping, leading to situations of mixed weak and strong cou- pling, where some of the states are bare while some others are dressed. In this text, I will address the case of two two-level sys- tems (Fig. 1(cde)), which compromises between simplicity of the linear model and richness of the Jaynes-Cummings model. In particular, thanks to the reduced size of the Hilbert space, I will be able to solve the problem fully analytically, as in the linear model, a convenience not afforded by the Jaynes- Cummings model (when including incoherent pumping). This will allow me to provide a complete picture of weak and strong coupling in a nontrivial system, and therefore shed light on more complicated systems. II. THEORETICAL MODEL The Hamiltonian for two coupled qubits reads: 2 My description will address more particularly independent qubits (Fig. 1(c)), for instance superconducting (Josephson) qubits [17], in the sense that their commutation rules will not be those of two fermions, that anticommute (Fig. 1(d)). I will also address the latter case for comparison and completeness, and obtain the elegant result that the expressions describing two coupled fermions are essentially identical to those de- scribing two coupled bosons, although these two systems are very different in character and behavior (for instance bosons accumulate arbitrary number of particles whereas fermions saturate at at most one, a distinction recovered in the formal- ism by merely substituting effective parameters). Also be- cause two two-level systems can be mapped to one four-level system (Fig. 1(e)), I will address this case in detail, finding another fundamental case of interest. The main results, however, and the deepest connections to be made with other models (such as the Jaynes-Cummings) will be obtained from the case of two qubits. Beyond its inter- est for the previous reasons, the study of the coupling between two qubits is interesting in its own right [18 -- 21]: it comes as the fundamental support of entangled states [22, 23], to imple- ment quantum gates [24, 25] and, in this quantum information processing context, the natural model to investigate decoher- ence [26, 27]. In this text, the two coupled qubits will allow us to investigate strong coupling under incoherent pumping. The rest of this paper is organized as follows. In section II, I introduce the model and its parameters and discuss the level structure. In section III, I obtain the single-time dynamics, that will be shown to be the same independently of the under- lying model. In section IV, I obtain the power spectra, that, in contrast with the previous section, depend strikingly on the underlying model. Power spectra (photoluminescence spec- tra in a quantum optical context) are important because this is where the dressed states manifest. I will analyze in detail the case of two qubits (Sec. IV A) and contrast it with that of two fermions (Sec. IV B) and of a four-level system (Sec. IV C). In section V, I describe the strong and weak coupling regimes, first in the absence (Sec. V A), and then including (V B) the incoherent continuous pump. I define new regimes of strong coupling, proper to the two qubits system. In section VI, I il- lustrate the results of previous sections with examples of some interesting configurations: cases where pumping effect is op- timal (Sec. VI A) and detrimental (Sec. VI B) for the coher- ent coupling. In section VII, based on the previous results, I reconstruct the dressed states, uncovering an unexpected manifestation of decoherence -- the emergence of additional dressed states -- to be found only in the coexistence of pump- ing and decay along with the coherent coupling. Finally, in section VIII, I give a summary of the main results as my con- clusions. To avoid distraction in the main text, most technical details appear in appendices (B, C, D) along with further ma- terial outside the scope of this study (A, E). H0 = w 1s † 1 s 1 + w 2s † 2 s 1) , s 2 + g(s † 1 s 2 + s † 2 (1) where s 1,2 are the lowering operators of the qubits, with bare energies w 1,2. They are linearly coupled with strength g. The two modes can be detuned, by a quantity D = w 1 − w 2, that is small enough, D ≪ w 1,2, so that the rotating wave approx- imation is justified. The Hilbert space of the coupled system It can be de- has dimension four, with the structure 2 ⊗ 2. composed in three subspaces (also called manifolds, rungs, etc.) with a fixed number of excitations: the ground state, { 0,0i}, with zero excitation, the excited state of each qubit, { 1,0i, 0,1i}, with one excitation, and the state { 1,1i} with two excitations. An important point throughout this text is the commutation rules in Eq. (1), that are those of two distinguishable systems, i.e., s is i = s † i [s i,s † [s i,s † [s i,s s † i = 0 , i ]+ = s is † i + s † j ] = s is † j − s † j] = s is j − s i = 1,2, , s i = 1 , i s i = 0 , j js i = 0 , i = 1,2 , i 6= j , i 6= j . (2a) (2b) (2c) (2d) Note that two operators from different systems commute. These commutation rules for two-level systems is most com- monly found in the literature of the Dicke model [28], that describes a gas of two-level systems emitting in a common radiation field. In the case of a fermion gas, this commu- tation is an approximation, that is made for the simplicity of the algebra and that is justified for a dilute gas by the fact that anticommuting operators give the same final phys- ical results [29]. Indeed, when the wavefunctions of any two fermions is weakly-overlapping, symmetrized, antisym- metrized and non-symmetrized results are the same [30]. In our case, the two qubits are strongly interacting, which sets our system apart from the Dicke model and its approxima- tions in many respects (see appendix A). For reference, I will analyze the antisymmetrized case of two interacting fermions (Fig. 1(d)) completely. However, the main object of interest in this text is that of two commuting qubits (like in the Dicke model), corresponding to the case of two distinguishable (in the quantum sense) qubits (Fig. 1(c)). The level structure of Hamiltonian (1) is sketched in Fig. 2(a) (at resonance). Such a "diamond-like" configuration can be mapped to a four-level system (4LS), i.e., as a single entity, with Hilbert space structure 1⊕ 1⊕ 1 ⊕ 1 (Fig. 1(e)). This description fits, for instance, the case of a multi-level atom [31] or of a single quantum dot that can host up to two interacting excitons (electron-hole pairs) forming a biexciton state [32]. On the other hand, two coupled qubits, matches the case of two nearby quantum dots directly coupled. The state with double excitation is then called an interdot biexci- ton state [33]. The Hamiltonian H0 can be diagonalized in terms of two intermediate dressed states, +i and −i, as: H0 = w − −ih− +w + +ih+ +w 11 1,1ih1,1 with eigenfrequecies: w w 1 + w 2 ± = w 11 = w 1 + w 2 . 2 ± R , R =sg2 +(cid:18)D 2(cid:19)2 , 3 (3) (4a) (4b) The diagonalized level structure is sketched in Fig. 2(b). These are the same eigenfrequecies w ± and Rabi splitting (given by 2R) than for the dressed states of two coupled har- monic oscillators up to the first manifold [14]. This equiva- lence breaks in the manifold with two excitations where the fermionic nature of the particles reveals and only the state 1,1i is permitted, as compared to three possible states in the second manifold of the linear model: { 2,0i, 1,1i, 0,2i}. The dynamics of dressed states ±i and their spectral shape depend on the amount of decoherence that the dissipative and excitation processes induce in the system. I will consider an incoming flow of excitations that populate the two-level sys- tems at rates P1, P2 and an outgoing flow (given by the in- verse lifetime) at rates g 1, g 2, respectively. This situation cor- responds to an incoherent continuous pump or injection in the qubit that can be varied independently from the dissipation. The steady state reached under the pump and decay corre- sponds to a statistical mixture of all possible quantum states and is described by a density matrix r . The master equation of the system has the standard Liouvillian form [34], with the corresponding Lindblad terms: dr dt + (cid:229) = L r =i[r ,H] g i 2 Pi 2 + (cid:229) i=1,2 i=1,2 (2s irs (2s † i rs (5a) (5b) (5c) s i) † i is † i ) . i † i − s † i − s is † s ir − rs r − rs i This master equation can be exactly solved given the finite and small dimension of the Hilbert space (which is only four). Within the same formalism, we can describe the spontaneous emission from a general initial state by solving the equations for vanishing pumping. I will note the transitions between bare states (cf. Fig. 2(a)) as: 2 s 2 , u1 = 0,1ih1,1 = s 1s † l1 = 0,0ih1,0 = s 1 − u1 , u2 = 1,0ih1,1 = s † s 1s 2 , l2 = 0,0ih0,1 = s 2 − u2 . 1 (6a) (6b) (6c) (6d) (for upper and lower transition). Note that they can all be written in terms of the qubit operators (in normal order) since s i = ui + li (i = 1,2). The Lindblad terms of pump and decay in Eq. (5) are expressed in terms of their s operators. Rewrit- ing them in terms of the transition operators u1,2, l1,2 leads to FIG. 2: (Color online) Energy levels for the two qubits described by Hamiltonian (1). Weak coupling (a) and strong coupling (b) in the absence of pump are well defined in terms of the bare and dressed states ±i, respectively. As shown in Sec. V B, the SC regime gives rise to new regions when pump is taken into account: (c) FSC where dressed states remain ±i , an (d) SSC an MC, where dressed states form a new set I±i, O±i. Only SSC, with splitting of all the new dressed states, is shown. MC corresponds to the case where I±i have closed (both collapsing on w 1). The thickness of the levels represents the uncertainty in energy due to (a,b) the decay and (c,d) both the pump and the decay. The arrows linking the levels due to pump/decay are blue for the lower (li) and red for the upper (ui) tran- sitions. Transitions labeled A, C (involving +i or I±i, in green) occur at frequencies determined by z1, while B and D (involving −i or O±i, in orange) are determined by z2. The same color code is used in the rest of the figures to plot the decomposition of the spectra. cross Lindblad terms that entangle l1 and u1, on the one hand, and l2 and u2 on the other. If the four levels did not correspond to two qubits but to a single entity (a 4LS), such as atomic lev- els or a single quantum dot levels, the Lindblad terms would be written directly in terms of the transition operators, without these cross Lindblad terms. The alternative master equation is discussed in appendix C (cf. Eq. (C1)). We introduce fermionic effective broadenings: G 1 = g 1 + P1 , G 1 ± G 2 ± = 4 , G 2 = g 2 + P2 , g 1 ± g 2 g ± = 4 (7a) (7b) . Equation (7a) is to be compared with bosonic effective broadenings, for boson modes a and b (such as the harmonic G oscillators of the linear model): G a = g a − Pa , G b = g b − Pb . (8) A tilde is being used to denote the bosonic character of the broadening, in the sense that the pumping strength is sub- tracted to the decay. Pumping leads to broadening of the line in the fermionic case and narrowing in the bosonic case, which is a spectral manifestation of Fermi and Bose statistics. Another convenient notation for Eqs. (7) is the expression of pumping and decay rates in terms of G s and a new parame- ter r which represents the type of reservoir that the qubit is in contact with [35]: g 1 = G 1(1− r1) , P1 = G 1r1 , P2 = G 2r2 . g 2 = G 2(1− r2) , (9a) (9b) All possible situations (with 0 ≤ ri ≤ 1), from a medium that only absorbs excitation, ri = 0, to one which only provides them, ri = 1, are thus included in a transparent way. For in- stance, a thermal bath with temperature different from zero corresponds to ri < 1/2. The effect of the medium on the ef- fective broadening is contained in G i. The power spectrum for each qubit, i = 1,2, is defined as: si(w ) = hs † i (w )s i(w )i = 1 2p ´ Z 0 Z 0 G(1) i (t,t + t )eiwt dtdt (10) where i G(1) i (t)s i(t + t )i , (t,t + t ) = hs † (11) is the first order auto correlation function. s(w ) describes how energy is distributed and is thus of fundamental interest. In a quantum optical context, this can be observed directly in the optical emission, but for generality, I will keep the terminol- ogy of power spectrum. For the steady state spectrum, the running time t is taken at infinite values (thereby removing one integral). In appendix B, we make use of the quantum regression formula [34] in its most general form to compute G(1) as well as other two- and one-time correlators (second order correlation functions are given in appendix E). We shall focus on i = 1 in the following, without loss of generality: i [Lp(t) + iKp(t)]e−iw pt e− g p 2 t , p=A,B,C,D (12) is used to normalize the expression for the spectrum (so that the population of qubit 1, , 1 (t,t + t ) = n1 G(1) where n1 = R s1(w )dw R S1(w )dw = 1): S1(w ) = 1 p g p p∈{A,B,C,D}"Lp 2(cid:1)2 (cid:0) − Kp (cid:0) g p 2 + (w − w p)2 w − w p + (w − w p)2# , g p 2(cid:1)2 (13) from Eqs. (10) and (12). 4 The spectrum is composed of four peaks that I label p = A,B,C,D, each of them with a Lorentzian (weighted by the coefficient Lp) and a dispersive part (weighted by Kp). The four resonant frequencies w p and the associated broadenings (full-widths at half maximum) g p, are intrinsic to the system, as they correspond to the four possible transitions in the sys- tem. The coefficients Kp and Lp (derived in appendix B) are the parameters that are specific to the experimental configu- ration (such as channel of detection) or regime (steady state under incoherent pumping or spontaneous emission of an ini- tial state). This form of S(w ) is a general feature for the power spectra of coupled quantum systems [36]. In the case of uncoupled qubits (g = 0), the four resonances reduce to the two bare energies w 1 and w 2, broadened by the effective decay rates G 1 and G 2, respectively. The peaks are, in this case, pure Lorentzians. In the opposite case of very strong coupling (g ≫ g ,P), the spectrum is also well approxi- mated by Lorentzians, but with the resonances w p now at the dressed state frequencies w ±, and broadened by the average rates (G 1 + G 2)/2. Lorentzian lineshapes correspond to the emission of well defined isolated modes of the system Hamil- tonian, weakly affected by other modes. In this case, Kp ≈ 0. The dispersive contribution becomes non negligible in the in- termediate situations when dissipation and decoherence (or dephasing, cf. appendix A) are of the order of the direct cou- pling. The Hamiltonian eigenmodes are then no longer neatly leading the dynamics and, as a consequence, their broad emis- sion lines overlap in energy, producing interferences. This is the regime of interest in our analysis, since new phenomenol- ogy appears for the dressed states and coupling regimes. With this goal in mind, I devote the next two sections to presenting the analytical expressions of all the quantities appearing in the spectrum, Eq. (13). III. SINGLE-TIME DYNAMICS We start by analyzing the relevant average quantities needed to compute the weights Lp and Kp in the spec- trum (13), that is, populations and coherences: n2 = hs † 2 1 s 1i , n1 = hs † s 2i ∈ C , ncorr = hs † 1 n11 = hs † s 1s † 2 1 s 2i ∈ R . s 2i ∈ R , (14a) (14b) (14c) ni is the probability that qubit i is excited. The sum n1 + n2, that can go up to two, is the total excitation in the system. ncorr is the effective coherence between the qubits due to the direct coupling. n11 is the joint probability that both qubits are excited. It is also the population of state 1,1i. If the qubits were uncoupled, we would have n11 = n1n2. It is important here to outline the difference between popu- lation of a mode (say, n1 = hs † s 1i for a qubit and na = ha†ai tion of a state (r 10 and r 10 for the state 1,0i with 1,0i = a† 0,0i and s † 1 0,0i, respectively). The population of the intermediate state 1,0i (resp. 0,1i) is given by n1 − n11 for an harmonic oscillator, cf. Fig. 1(a) and (c)), and popula- 1 ¥ ¥ (cid:229) (cid:229) (resp. n2 − n11). This is also the probability of having only one of the qubits excited. The population of the ground state is given by 1− n1− n2 + 2n11. In the spontaneous emission case, n1, n2 and ncorr have the same solutions (depending only on the initial condition n0 1, n0 2 and n0 corr) than their counterpart for the two coupled linear oscillators, na, nb and nab. However, the two models differ for n11: in the case of coupled qubits, it decays from its ini- tial value, n11(t) = e−4g +tn0 11, whereas in the linear model, the population r 11 oscillates as a result of the exchange with the other states that are available in the second manifold ( 2,0i and 0,2i). Although populations of the modes have the same dynamics both in the two harmonic oscillators and the two coupled qubits, the underlying populations of their states do not. For instance, the decay from the initial condition 1,1i leads to n1 = n10 + n11 = r 10 +r 11 for the qubit, while for the harmonic oscillators, one has, na = ha†ai = r 10 + r 11 + 2 r 20. Since n1 = na (and also n2 = nb) the states are differently pop- ulated (r 10 6= r 10, Etc.). In the steady state, all these mean values can be written in terms of effective pump and decay parameters, as in the two harmonic oscillators, but now following fermionic statistics (i = 1,2): Peff i nSS i = , g eff i + Peff i g 1 + g 2 g eff i = g i + G 1 + G 2 g D − 2iG + nSS corr = (n1 − n2) , (15a) Qi , Peff i = Pi + P1 + P2 G 1 + G 2 Qi , (15b) (15c) (16) (17) with the corresponding generalized Purcell rates Q1 = 4(geff)2 G 2 , Q2 = 4(geff)2 G 1 , and the effective coupling strength g geff = . r1 +(cid:16) D /2 G +(cid:17)2 Finally, n11 takes a simple intuitive form in the steady state, nSS 11 = 1 P2 + nSS nSS G 1 + G 2 2 P1 1 nSS 2 . ≤ nSS (18) This should be contrasted with the counterpart of nSS bosonic modes (a and b), for which: 11 for two ha†ab†biSS = 2nSS a nSS b − a nSS b . b Pa a Pb + nSS nSS G a + G b ≥ nSS of the wavefunction is an that interesting manifestation sym- This of two metry/antisymmetry bosons/fermions, is known to produce such an at- tractive/repulsive character for the correlators. Here we see that quantum (or correlated) averages h n1 n2i (with ni the number operator) are higher/smaller than classical (uncor- related) averages h n1ih n2i, depending on whether they are the for 5 FIG. 3: (Color online) Power spectrum, S1(w ), from a qubit (thick solid black line) in the SC regime (g 1 = g and g 2 = g/2) for the steady state under vanishing pump (P1 = 0.02g and P2 = 0.01g). The spectra is composed of four peaks arising from the lower and upper transi- tions (blue and red thin lines, respectively). In dashed purple, the linear model spectrum for comparison. of a boson or fermion character, respectively. This provides a neat picture of bunching/antibunching from excitations of different modes that are otherwise of the same character. In the most general case, with pump and decay, before the steady state is reached, also the transient dynamics of the mean values n1, n2 and ncorr for the coupled qubits maps to the corresponding averages na, nb and nab of coupled har- monic oscillators, with only G → G . We can conclude then, that the single-time dynamics of the qubit is ruled by a (half) Rabi frequency of the same form than in the boson case [14]: R1TD =qg2 − (G − + iD /2)2 , (20) only with Fermion-like effective broadenings, Eqs. (7). Since, in contrast to the boson case where there is only one Rabi pa- rameter, another expression will arise in the two-time dynam- ics for coupled qubits, I will refer to Eq. (20) as the single-time dynamics (half) Rabi frequency. We conclude this section by noting that the magnitudes studied up to now are independent on having two qubits, two identical fermions or a 4LS. The cross terms appearing in the master equation for the first case, due to the correlations in- duced by the incoherent processes, do not affect the steady state populations, as is shown in appendix C. (19) IV. POWER SPECTRA A. Two qubits The general expressions for the spectrum and two-time cor- relators of two coupled qubits admit analytic solutions at res- onance in the steady state of an incoherent continuous pump. From now on, we will refer always to this situation and, there- fore, I will drop the steady state label in the notation. The four coefficients Lp + iKp appearing in Eq. (12), now defined in the steady state, read: (cf. appendix B) (2z1 + iG 2) + 2(R + z1 + iG +)i ncorr n1 o , (21a) (2z2 + iG 2) + 2(R− z2− iG +)i ncorr n1 o , (21b) LA + iKA = LB + iKB = LC + iKC = LD + iKD = 1 1 G + P1 G + 16Rz1n2(2z1 + iG 2)(R− iG + 2gh− 16Rz2n2(2z2 + iG 2)(R + iG + 2gh P1 16Rz1n2(2z1 − iG 2)(R− iG + 2gh− 16Rz2n2(2z2 − iG 2)(R + iG + 2gh P1 P1 G + G + 1 1 −) + a1 + a2 n2 n1 −)− a1 − a2 n2 n1 −)− a1 − a2 n2 n1 −) + a1 + a2 n2 n1 (2z2 − iG 2)− 2(R + z2− iG +)i ncorr n1 o . (21d) (2z1 − iG 2)− 2(R− z1 + iG +)i ncorr n1 o , (21c) They are defined in terms of the parameters a1 = g2 G 2 + [4G 2 + +2P1(P2−2G +)−P2G 1] , a2 = g2 G 2 + P1(P1−g 1) , (22) and the corresponding frequencies and decay rates, that also appear explicitly in Eq. (12): + iw A = 2G + + iz1 , + iw C = 2G + − iz1 , g A 2 g C 2 They all depend on two complex parameters, z1 and z2: + iw B = 2G + + iz2 , + iw D = 2G + − iz2 . g B 2 g D 2 (23a) (23b) z1,2 =q(Dsg)2 + (iG + ± R)2 . (24) The degree of symmetry, Ds, is a real dimensionless quantity, between 0 and √2, given by Ds = p(g 1P2 + g 2P1)/2 G + . (25) This quantity is proper to the coupled qubits case and its physical meaning will be clarified later. Its value is linked to the symmetry between the different parameters. For in- stance, Ds = 1 when all parameters are equal to each other, g 1 = g 2 = P1 = P2. On the other hand, Ds = 0 if one of the pa- rameters (any of them) is much larger than the others. It leads to a renormalized coupling strength G = Dsg , (26) that reaches a maximum when the parameters are such that Ds = √2. Such an enhancement, by √2, is related to the 6 cooperative behavior of two coupled modes, similarly to the superradiance of two atoms in the Dicke model or the renor- malization with the mean number of photons in the Jaynes- Cummings Model. The last and most important parameter appearing in the pre- vious expressions is a Rabi frequency for the two-time dy- namics, that for coupled qubits differs from its counterpart for single-time dynamics (cf. Eq. (20)): R =qg2 − (Dsg)2 − G 2 − . This is the true analog of the (half) Rabi frequency of the linear model since this value, not its single-time counterpart, determines strong or weak coupling (emergence of dressed states). At vanishing pump, the renormalized coupling G con- verges to g, and both R and R1TD converge to the standard expression for the (half) Rabi splitting [14]: (27) (28) R0 =qg2 − g 2 − . The normalized power spectrum of qubit 1 follows from Eq. (13) with the coefficients we have obtained. The positions and broadenings of the four peaks are given respectively by the real and imaginary parts of z1 and z2. Their expressions remain valid in the spontaneous emission case by setting the pumping rates to zero. Figure 3 is an example of the spectrum S1(w ) (in solid black) and its decomposition in four peaks (thin blue and red). The split positions of the four peaks, which indicate the sys- tem is in the SC regime, are marked with two vertical blue lines. The two peaks that correspond to the lower manifold transitions, A and D in Fig. 2(b), appear with a thin blue line. Upper transitions, B and C in Fig. 2(b), appear with a thin red line. All resonances are at the same positions but the stronger dispersive part of upper transitions leads to a shift of their maximum. The upper transitions are much weaker in intensity (magnified ×30 to be visible) due to the small pump. The double excitation of the system is very unlikely (n11 = 0.0004). The lineshape is therefore close to that of two coupled harmonic oscillators, plotted with a dashed pur- ple line for comparison. The system is in the linear regime where all models of Fig. 1 for two coupled modes converge. In the following sections, we will see how the lineshapes change when entering the nonlinear regime. B. Two fermions As noted before, although the expressions for the single- time dynamics (populations, coherence, Etc.) for two coupled fermions (anticommuting operators for modes 1 and 2) are the same than for the two coupled qubits, their power spectra are different. As compared to the coupled qubits, the coupled fermions spectrum assumes a simple and fundamental form, closely related to that of the two harmonic oscillators: the for- mal expression is the same, differing only in the parameters (effective broadenings, populations, Etc.). In particular, only one Rabi parameter, the single-time Rabi frequency, R1TD, de- termines both the single- and two-time dynamics. The two- fermions spectra are thus obtained by simply substituting the fermionic parameters (Eq. 7) in the expression of the linear model [14]. This simplicity and likeliness to the linear model stems from the fundamental nature of the problem: two identical (in- distinguishable) particles coupled linearly, obeying fully their quantum statistics. The two coupled qubits (or the Jaynes- Cummings model [16]), by mixing different types of particles (distinguishable modes) and therefore breaking commutation rules, result in the more complex description and richer dy- namics presented in the previous section. C. four-level system (4LS) Also in the four-level system (with no cross Lindblad terms), the expressions for the single-time dynamics (popula- tions, coherence, Etc.) are the same than for the linear model, and here also their power spectra are different. The parame- ters for the 4LS spectra are of a bosonic character, cf. Eq. (8): G 1 = g 1 − P1 , G 1 ± G 2 G ± = R =qg2 − G 2 − . 4 , G 2 = g 2 − P2 , (29a) (29b) (29c) The relevant parameters that characterize the coupling sim- plify to: G + G + R , R = z1,2 = R± i G + , (30a) (30b) recovering the conventional strong coupling criterion based on one parameter only, the bosonic (half) Rabi frequency R. The resulting spectral structure then consists of two pairs of peaks sitting at ±´ + iw A = ( R) with: g A 2 g B 2 g C 2 g D 2 + iw B = + iw C = + iw D = 3(P1 + P2) + g 1 + g 2 4 3(g 1 + g 2) + P1 + P2 4 3(g 1 + g 2) + P1 + P2 4 3(P1 + P2) + g 1 + g 2 4 + i R, + i R, − i R, − i R. (31a) (31b) (31c) (31d) Note that g p are always positive for any combination of the pa- rameters, in contrast with those of two bosonic modes, where the system can diverge. Therefore, the values of pump and decay rates here are not limited, always leading to a physical steady state. 7 V. STRONG AND WEAK COUPLING REGIMES The standard criterion for strong coupling (SC) is based on the splitting at resonance of the bare states into dressed states. This manifests in the appearance of t -oscillations in the two- time correlators and a splitting of the peaks that compose their spectrum. In a naive approach to the problem of defining strong- coupling in a system other than the linear model, one could think that the condition for SC is ´ (R1TD) 6= 0 (at resonance), leading to the familiar inequality, g > G −. However, this is not the case whenever pump and decay are both taken into ac- count. Instead, one must find the condition for a splitting be- tween the new eigenstates, that is, the two pairs of peaks form- ing the spectrum. The peaks are positioned symmetrically in two pairs about the origin at w p = ±´ (z1,2) and, therefore, (z1) 6= 0 or (z2) 6= 0 (32) is the mathematical condition for SC in this system. Given that there are two different parameters z1 and z2 on which the condition relies, the SC/WC distinction must be extended to cover new possibilities. Thus, instead of only one relevant parameter, G −/g, as was the case in the linear model, SC be- tween two qubits is determined by three parameters: −/g , G +/g and Ds . (33) This gives rise to the situations listed in Table I, that are discussed in the following sections. R ´ R iR iR iR (z1) ´ 6= 0 6= 0 0 0 (z2) Acronym 6= 0 6= 0 6= 0 0 FSC SSC MC WC Type of coupling First order Strong Coupling Second order Strong Coupling Mixed Coupling Weak Coupling TABLE I: Type and nomenclature of coupling for two coupled qubits. Beyond the usual weak coupling (WC) and strong coupling (here denoted FSC) encountered in the linear model, the system ex- hibits two new regions: Mixed Coupling (coexistence of weak and strong coupling) and Second order Strong Coupling (with two differ- ent splittings of two pairs of dressed states). A. Vanishing pump and spontaneous emission In the case of vanishing pump, that corresponds as well to spontaneous emission, the standard SC and WC hold. In this limit, we recover the familiar expression for the half Rabi frequency R,R1TD → R0. The parameters simplify to z1,2 →p(R0 ± ig +)2 = R0 ± ig + [39]. The positions and broadenings of the four peaks are: g A 2 g C 2 + iw A = g + + iR0 , + iw C = 3g + − iR0 , + iw B = 3g + + iR0 , + iw D = g + − iR0 . (34a) (34b) g B 2 g D 2 ´ ´ G From here, the associated condition for SC reduces to R0 be- ing real, or more explicitly g > g −, as in the linear model at vanishing pump [see Fig. 2(b)]. In SC, the two pairs of peaks p = A,D and p = B,C sit on the same frequencies although they have different broadenings. Excited states have shorter lifetime, since each excitation can decay. From Eqs. (34), the two dressed states undergo the transition into weak coupling (WC) simultaneously [as in Fig. 2(a)]. In WC, R0 → iR0 and both parameters z1,2 become imaginary, giving 8 p = A,B,C,D , w p = 0 , g A 2 g C 2 = g + −R0 , = 3g + +R0 , g B = 3g + −R0 , 2 g D = g + +R0 , 2 (35a) (35b) (35c) with g p ≥ 0, since g + ≥ R0 in this regime. The four peaks collapse into four Lorentzians at the origin, all differing in their broadenings. As a result of the two pairs of peaks sitting always on the same two (or one) frequencies, the final spectra can only be either a single peak or a doublet, both shapes being possi- ble in SC or WC regimes (as in the linear model and for the same reasons [14]). An intuitive derivation and interpre- tation of these results is given in appendix D, based on the so-called manifold picture [13], which consists in considering transitions between eigenstates of a non-hermitian Hamilto- nian, with energies broadened by the imaginary part. In the steady state case but in the limit of vanishing pump (the linear regime), only the vacuum and first manifold are populated. The spectra in this limit converge with the lin- ear model and also it can be analyzed in terms of manifolds by straightforward extension. The spectrum in Fig. (3) is an example of SC for vanishing pump as we can see from the fact that the lower transition peaks, in blue, dominate over the broader and weak upper peaks, in red. In this case, the split- ting of the dressed modes gives rise to a splitting in the final spectrum (in black). In what follows, we take this SC config- uration (g 1 = g and g 2 = g/2) as a starting point to explore the effect of a non-negligible incoherent continuous pump. B. Non-negligible pump When pump is taken into account, all the types of coupling listed in Table I are accessible. These are plotted in Fig. 4 as a function of the pumping rates. The starting point is the Rabi frequency R, Eq. (27), that is either real or pure imaginary. First, let us consider the case where: (36) R = R ⇔ G2 < g2 − G 2 − , from which follows that z1 = z∗2, and therefore ´ (z2). This is the most standard situation that we already found in the absence of pumping. It is sketched in Fig. 2(c). To dis- tinguish it from the other types of coupling to be discussed shortly, I will from now on call it First order Strong Coupling (FSC). Note that condition (36) can only be satisfied if Ds < 1, therefore, when the renormalization of the coupling through (z1) = ´ FIG. 4: (Color online) Phase space of the steady state Strong/Weak Coupling regimes as a function of pump for g 1 = g and g 2 = g/2. In Strong Coupling (SC, blue), one can distinguish two regions, First order (FSC, light blue) and Second order (SSC, dark blue) Strong Coupling. Weak Coupling (WC) is in purple and Mixed Coupling (MC) in green. The dashed blue lines enclose the two regions where two peaks can be resolved in the power spectrum of the first qubit, S1(w ). One falls in SC and the other in WC. the interplay of pump and decay is detrimental (since Eq (36) implies that G < g). It is not possible to reach the optimum effective coupling and maximum splitting of the spectral lines given by 2√2g. Eq. (36) leads to the following explicit con- dition for g: g > . (37) G − p1− (Ds)2 (z1) = ´ If R 6= 0, then, also ´ standard SC regime in the absence of pump (g > G (z2) 6= 0. FSC includes the − that is implied by Eq. (37)). It is the most extended region in Fig. 4, colored in light blue. In this case, the spectrum of emission follows the expected pattern: two pairs of peaks, A, D and B, C, are placed one on top of each other, although they are differently broadened [see the spectra in Fig. 3]. In Fig. 5(a) and (b) we track the broadenings and positions of the four peaks (A and D in blue and B and C in red) as a function of pump, through the SC region of Fig 4, on a diago- nal line defined by P2 = P1/2. Following them from vanishing pump, where the manifold picture is exact, the four peaks can be easily associated with the lower and upper transitions of Fig. 2(b), and that is why we keep the same color code and notation. Dressed states ±i, close to the Hamiltonian ones, can still be defined in the system, but with the modified fre- quencies w 1 ± ´ (z1), both affected equally by decoherence. By construction, the resulting spectra in this regime can only be a doublet or a single peak, depending on the mag- nitude of the broadening of the peaks (that always increases with pump and decay) against the splitting of the lines (that always decreases). As in the limit of vanishing pump, ob- serving a doublet in the spectra does not imply splitting of the dressed states (and thus, SC) [37], but here the tendency is al- ways the same: the lower the pump and the decay, the better the resolution of the splitting. Second, let us consider the situation of the Rabi frequency being imaginary: R = iR ⇔ G2 > g2 − G 2 − . (38) This results in three possibilities, listed in Table I (WC, SSC and MC), that constitute the three remaining regions delimited in Fig. 4. In what follows we find the specificities of each of these three regimes. The Weak Coupling regime (WC, in purple) is characterized by z2 = iz2, (z1) = ´ z1 6= z2 ⇔ G < G +−R (39) z1 = iz1, and therefore ´ (z2) = 0. Note that condition (39) is not analytical in terms of the relevant parameters (33). In WC, the four peaks are placed at the origin with four different broadenings. The dressed states have collapsed in energy to w 1. Up to here, we have remained within the SC and WC re- gions already known from the linear model. We now consider the two new regions of SC, proper to the coupled qubits, that I call SSC and MC, respectively: z2 = z2, z1 < z2 ⇔ G > G + +R . (40) SSC: When both parameters z1,2 are real, then: z1 = z1, We refer to it as Second order Strong Coupling regime (SSC, colored in dark blue in the phase space). Here, the broad- enings of the four peaks are equal, g p/2 = 2G +, but the po- sitions of the pairs of peaks are different, w A,C = ±z1 and w B,D = ±z2. The reason is that the bare energies of the modes undergo a second order anticrossing induced by the interplay between coupling, pump and decay. The energies of the dressed states are affected differently by decoherence, up to the point where we may picture the physics in terms of a new type of eigenstates. The association of the A and D (with w A,D) as the peaks corresponding to lower transitions and B and C (with w B,C) to upper transitions is completely arbitrary in this region, given that the broadenings of the peaks, which led us to such association in FSC, are now equal. This implies that, rather than two dressed states, −i and +i (Fig. 2(b)) as in the conventional strong coupling (FSC), the system now exhibits four dressed states: I−i, I+i, O−i and O+i. They are plotted in Fig. 2(d): I±i (resp. O±i) have energies split at ±z1, giving rise to the inner peaks, in green (resp. ±z2, giving rise to the outer peaks, in orange). The physical ori- gin of this remarkable departure from the conventional strong coupling picture will be discussed in section VII. We can see how peak broadenings and positions change when going from FSC to SSC in Fig. 5(c) and (d). In this case, we track the peaks by varying P2 for a fixed P1, mov- ing upwards in the phase space. The first vertical guideline 9 FIG. 5: (Color online) Broadenings (a), (c), and positions (b), (d) of the lines that compose the spectra as a function of pump for the decay parameters g 1 = g and g 2 = g/2. In the plots of the first column, the pump P1 varies with P2 = P1/2, moving upwards in the phase space of Fig 4. The vertical guideline shows the crossing from FSC to WC. In the plots of the second column, the pump P2 varies with P1 = 0.2g, moving in diagonal in the phase space of Fig 4. The vertical guidelines show the crossing from FSC to SSC and finally to MC. The dashed blue line represents the splitting as it is resolved in the final spectrum S1(w ). The color code (red-blue and orange- green) corresponds to that of the transitions in Fig. 2. marks the border between the two kinds of SC, with the open- ing of a "bubble" for the positions w A and w B (that were equal in the FSC region), and the convergence of all the broaden- ings. In principle, one can expect that quadruplets and triplets may form out of the four peaks. However, the broadenings and contributions of the dispersive parts (given by Kp) are too large to let any fine splitting emerge clearly. The spectra in this region reduce to singlets and doublets. However, we show in Sec. VI through some examples that they may be distorted, doubtlessly reflecting the multiplet structure. MC: When z1 is imaginary and z2 real, or equivalently, z2 = z2 ⇔ G + −R < G < G + +R , z1 = iz1 , (41) we enter the last new region in Fig. 4. This is a Mixed Cou- pling regime (MC, colored in green in the phase space) where the two inner peaks, A and C -- as well as the reconstructed eigenstate I±i -- have collapsed at the origin, like in WC. However, the two outer peaks, B and D -- as well as O±i -- are still split. As in SSC, the broadening of the peaks does not allow for a distinction between upper of lower resonances. The collapsed resonance is in this case at the bare energy w 1 = 0 because that is the total average bare resonance in the system. In Sec. VI A we will see that when the qubits are detuned, this resonance happens at (w 1 + w 2)/2 = −D /2. Again, although one may expect a triplet in MC, only dis- torted singlets are observed in the best of cases due to the broadening and dispersive parts. In Fig. 5(c) and (d) we can see the transition from SSC into MC, at the second vertical line. Note that, in this system, the pumping mechanism is equiv- alent to an upward decay, due to the ultimate saturation of the qubit and the symmetry in the schema of levels that they form. The master equation is symmetrical under exchange of the pump and the decay (g i ↔ Pi) when the two-levels of both qubit are inverted ( 0,0i ↔ 1,1i and 1,0i ↔ 0,1i) [40]. Consequently, the parameters z1, z2 and R, and also the popu- lations of all the levels, are symmetric in the same way, as it happens with just one qubit. In other systems, like the linear model, the Jayne-Cummings model or simply a single har- monic oscillator, the effect of the pump extends upwards to an infinite number of manifolds while the decay cannot bring the system lower than the ground state. There is no natural trun- cation for the pump (that ultimately leads to a divergence), as there is for the decay. But with coupled qubit, state 1,1i is the upper counterpart of 0,0i, undergoing a saturation. This implies, for instance, that the (transient) dynamics in the limit of vanishing decay is exactly the same as that of vanish- ing pump and that in such case we can also apply the mani- fold method to obtain the right positions and broadenings as a function of pump, in the same way that we did as a function of decay only. We only have to take into account the mentioned symmetry consistently. As long as the dynamics moves up- wards or downwards only, even when intermediate states are coupled, the manifold picture is suitable. The manifold di- agonalization breaks, however, in the presence of both non- negligible pump and decay. This is discussed in appendix D. VI. PARTICULAR CASES In this section, we illustrate the rather abstract previous discussions with examples. The symmetry in the decay and pumping rates determines the effective coupling, emission properties and dressed states. Let us start by expressing Ds, the magnitude quantifying such symmetry, in terms of the reservoir parameters (G In this form, its physical meaning is more clear. There are two separate factors to discuss: the symmetry in the strength of the couplings to the reservoirs, given by √G 1G 2/[(G 1 +G 2)/2] and plotted in Fig. 6(a), and the symmetry in the nature of the reservoirs given by √r1 + r2 − 2r1r2, plotted in Fig. 6(b). The coupling g is enhanced when g < G ≤ √2g, that is, 1 < Ds ≤ √2 , (43) which happens when r1 > 1/2 and r2 < 1/2 (or the other way around). This corresponds to the two squared regions with lighter colors in Fig. 6(b). Then, the two reservoirs are of opposite natures: the reservoir of the first qubit provides ex- citations (P1 > g 1) while the other absorbs them (P2 < g 2). If this is accompanied by similar interaction strengths, G 1 ∼ G 2, enhancement occurs. We refer to these situations as optimally pumped and study them in Sec. VI A. On the other hand, if the reservoirs are of the similar na- tures, both r1,r2 ≥ 1/2 or ≤ 1/2, both providing or absorbing particles, then the system is detrimentally pumped: 0 ≤ Ds ≤ 1 . 0 ≤ G ≤ g, that is, (44) It corresponds to the two squared regions with darker colors in Fig. 6(b). We study this in Sec. VI B. Ds = √2 i and ri), √G 1G 2 (G 1 + G 2)/2pr1 + r2 − 2r1r2 . (42) 10 FIG. 6: Factors contributing to Ds: (a) √G 1G 2/[(G 1 + G 2)/2] as a function of G 1,G 2 and (b) √r1 + r2 − 2r1r2 as a function of r1, r2. The values corresponding to the contour lines are marked on the plots. Both functions take values from 0 (dark blue) to 1 (light). A. Optimally pumped cases: g < G ≤ √2g Let us explore the optimally pumped cases by consider- ing parameters on the diagonal G 1 = G 2 in Fig. 6(a) together with the antidiagonal r1 + r2 = 1 in Fig. 6(b). The reservoirs have opposite nature but interact with equal strength with the qubits. This corresponds to the situation where the decay and pumping parameters are equal in a crossed way: P1 = g 2 , and P2 = g 1 . (45) The system has a total input that is equal to the total output, PTOT = P1 + P2 = g TOT = g 1 + g 2, and also equal Purcell rates, Q1 = Q2. The excited and ground states are formally equiva- lent in the dynamics. Figure 7 shows the different coupling regimes accessible with this configuration, as a function of P1 and P2 with the same color code than in Fig. 4. This configuration is in FSC only when all parameters are equal, Ds = 1, (blue line) and there is total symmetry in the system. Otherwise, one of the new type of coupling (SSC in blue or MC in green) is realized as the coupling is effectively improved, G > g. In the inset the type of spectral shapes that results is shown. singlet doublet Lineshape Lsl Lcon 2 1 6 distorted singlet 1 4 3 8 distorted doublet 3 5 6 8 7 quadruplet triplet TABLE II: The lineshapes S1(w ) are defined by two quantities: Lsl is the number of times that S1(w ) changes slope, that is, the number of real solutions to the equation dS1(w )/dw = 0; Lcon is the number of times that S1(w ) changes concavity, that is, the number of real solutions to the equation d2S1(w )/dw 2 = 0. The vertical axis in Fig. 7, with P1 = g 2 = g and P2 = g 1 = 0, is illustrative of all the possible coupling regions and line- shapes. This is the extreme situation of optimal pumping where the two reservoirs interact equally strongly with the 11 FIG. 8: (Color online) (a) Positions of the peaks (w A in thick green, w B in thick orange) and populations (n1 in dashed blue, n2 in dashed purple, ncorr in dotted brown) as function of P1/g = g 2/g = g /g, for P2 = g 1 = 0. The system goes from FSC (at 0) to SSC, to MC, to WC, while the lineshape of the spectra changes as coded in Table II. The most interesting lineshapes, that can only appear in SSC and MC, are the distorted doublet (I) and singlet (II). The total spectra (in black) is decomposed in inner (green) and outer (orange) peaks coming from the transitions sketched in Fig. 2(d). the splitting of the dressed modes is the largest possible, 2√2g (even though the lineshape remains a singlet). Finally, when the coupling becomes very weak, g ≫ g, the first dot satu- rates, quenching the exchange of excitation between the qubit and n2 = n11 = ncorr = 0. Although the underlying physics of coupling is very rich and complex, the spectra do not acquire distinctively marked lineshapes (such as well resolved triplets of quadruplets): a doublet in SSC only gets distorted, but this is unambigu- ously due to the underlying quadruplet structure, as shown in Fig. 8(I), and also a singlet gets distorted due to the underly- ing triplet structure, in (II). Before reaching WC, the spectrum has become a plain singlet. The way to distinguish mathemat- ically the different possible shapes and their origin in under- lying triplets and quadruplets is by counting zeros of first and second derivatives, as given in Table II, of which only the four first lines are realized in the coupled qubit. In Fig. 9 we make the comparison between the results dis- cussed in Fig. 8 and those obtained for a 4LS (without cross Lindblad terms), where the system undergoes a standard tran- sition SC-WC as g increases. From Sec. IV C, we know that, although the populations do not change from those in Eqs. (46), the position of the four peaks differ, as well as the coupling regimes. In this case the parameters z converge to: z1,2 →qg2 − (g /2)2 . (48) This corresponds to a splitting intermediate between those of FIG. 7: (Color online) Phase space of FSC/SSC/MC/WC as function of P1/g = g 2/g and P2/g = g 1/g. The color code is that of Fig. 4. In inset, the possible lineshapes of S1(w ): a doublet (red), a distorted doublet (green), a distorted singlet (blue) and a singlet (white), as explained in Table II. two qubits (G 1 = G 2 = g ) and have completely opposite na- tures: one only providing particles (r1 = 1) and the other only absorbing them (r2 = 0). At this point, there is maximum renormalization of the coupling, G = √2g (Ds = √2) as both factors in Eq. (42) are maximum. The populations and mean values read: 2 n2 = n11 = n1 = 1− n2 , 4 + (g /g)2 , 4 + (g /g)2 6= n1n2 , 1 g /g ncorr = −i 4 + (g /g)2 . − = 0), and z1,2 =qg2 − (g /2)2 ∓ gg . (46a) (46b) (46c) (47) The two qubits share one excitation only. The Rabi frequency also simplifies to R = ig (as G In Fig. 8, we can see some of these magnitudes varying in the different regimes as a function of g /g. In the linear regime limit, g ≪ g, there is FSC with all the levels equally popu- lated (n1 = n2 = 1/2, n11 = 1/4) and ncorr = −i(g /g)/4. As soon as there is pumping, the SSC opens a "bubble" in the eigenenergies with the splitting of inner and outer peaks. The transition into MC, with the collapse of the inner peaks, takes place at g = 2(√2− 1)g, and the transition into WC, closing the bubble, takes place at g = 2(√2 + 1)g. The maximum of z2 = √2g (in orange) takes place at g = 2g, when the coher- ence ncorr = 1/4 is maximum. This is a special point where 12 FIG. 11: (Color online) Spectra for varying positive detuning (an- ticrossings) in SSC (I) and in MC (II) for the corresponding cases in Fig. 9. The inset in (II) shows the positions of the three/four peaks composing the spectra as a function of detuning. Close to res- onance, the inner peaks converge to the average transition energy w 11/2 = D /2. At large detunings, the first dot (that is pumped) emits at w 1 = 0 and dominates over the second dot (that decays) with emis- sion at w 2 = −D . superimpose whenever there are only one or two resonances. Otherwise, green lines represent inner resonances and orange lines the outer ones. Dashed black lines represent the case of a 4LS, which simply splits in two lines close to resonance in SC (the transition to WC happens at g = 2g). In thin black, the bare energies (w 1 = 0, w 2 = −D ) are plotted as a reference, and recovered in all cases very far from resonance or well into WC. The 4LS resonances converge to the bare energies at all detunings almost as soon as they enter WC. However, the resonances of two qubits, when closing in WC (MC in the case of inner resonances), collapse to the average between the bare energies (w 1 + w 2)/2 = −D /2 instead. At resonance these two behaviors are equivalent, but at small detunings they are clearly very different. We will discuss the reason for this puzzling saturation at −D /2 instead of −D in Sec. VII. Even- tually, at very large g , the bare energies will be recovered both by inner and outer peaks. The anticrossing that the lineshapes form when detuning between the modes is varied from zero to some detuning D max, is also peculiar. In Fig. 11(I) and (II) we can see that the dis- torted doublet and singlet keep their features up to D max = g and D max = 2g (resp.). The resonances in Fig. 11 are not equally present in the emission. In both cases, at large de- tuning, the emission at w 1 = 0 of the first qubit is dominant over that of the second qubit at w 2 = −D because the first qubit is being pumped and the second only dissipates the ex- citation. Note that, the singlet in (II) slightly oscillates to the left before joining the origin at large detuning. This is a clear signature that the central peak is not simply a feature of WC, but of a more complicated MC eigenstate structure, as can be seen in the inset. This leftwards shift is a consequence of the splittings saturating at −D /2 instead of the bare energies (0 and −D ). This shift is even more evident when both inner and outer peaks saturate at some detuning, like in case g = 4g. FIG. 9: (Color online) Comparison between the results of Fig. 8 (solid lines) and those obtained for a 4LS (dashed black lines). The positions of the peaks composing the spectra, are marked with ver- tical lines in (I) and (II). In a 4LS, only plain doublets and singlets arise. The populations are the same for both cases. Different types of splitting lead to dramatic differences in the spectral shapes. FIG. 10: (Color online) (a) Positions of the peaks (w A,C in thick green, w B,D in thick orange) as a function of detuning for six cases in Fig. 8 with the specified parameter g . The positions in the absence of crossed terms appear in dashed black and in the absence of coupling in thin black, for comparison. The first two plots are those labeled (I) and (II) in previous Figures. the inner and outer peaks, as we can see in dashed black in Fig. 9(a). Although the 4LS splitting is smaller than the split- ting between outer peaks (in orange), the SC doublet in the spectrum is better resolved than in the case of two qubit, due to the large intensity of the inner peaks. Fig. 10 displays the behaviors of all the resonances as a function of detuning for six cases in Fig. 8 (see insets), in- cluding (I) g = g/2 and (II) g = g. Inner and outer peaks B. Detrimentally pumped cases: 0 < G ≤ g To explore the detrimentally pumped cases, we consider pa- rameters on the diagonal r1 = r2 = r in Fig. 6(b), where the reservoirs have the same nature. If the reservoirs also have the same interaction strength with the qubit, G 1 = G 2, the two qubits become indistinguishable. This corresponds to equal dissipation and pump, g 1 = g 2 = g and P1 = P2 = P . (49) The symmetry in the system is not total as the income- outcome flows of excitation may not be balanced, g TOT 6= PTOT. The steady state populations are the same as for the un- coupled case with n1 = n2 = P/(g + P), n11 = n1n2, ncorr = 0, as happens with any two indistinguishable coupled modes (like, for instance, harmonic oscillators). This does not mean that the qubit are uncoupled, in fact, the system can be con- sidered always in strong coupling (FSC) with g = 2r− 1 2g , R = g − P g + P √g P Ds = (g + P)/2 = 2pr(1− r). (50a) (50b) If P ≫ g (r = 1), or the other way around (r = 0), the symme- try between the flows (inwards and outwards) is completely broken and Ds = 0, although, at the same time, the splitting in dressed modes is maximum R = g (although not enhanced). On the other hand, if P = g (r = 1/2), the symmetry is total (Ds = 1). Here, all the levels are equally populated (n1 = n2 = 1/2, n11 = 1/4) and the Rabi frequency vanishes R = 0. The SC/WC regimes become conventional with z1,2 =pg2 − g 2. The SC condition is simply given by g > g . This particular situation is depicted in Fig. 7 with a thick dark blue line in SC that goes into WC when the parameters equal 1. A second possibility that results in degrading the effective coupling, still considering reservoirs of the same nature (r1 = r2 = r), is when r = 1/2, which means: g 1 = P1 = G 1/2 and g 2 = P2 = G 2/2 . (51) Although the parameters may be different, G 1 6= G 2, there is compensation between the flows (inwards and outwards) g TOT = PTOT. This configuration arises when both qubits are in a thermal bath of infinite temperature. The popula- tions are also those of the uncoupled system, n1 = n2 = 1/2, n11 = n1n2, ncorr = 0. The Rabi and degree of symmetry read R = G − G + qg2 − G 2 + and Ds = The FSC condition reads now g > G +. √G 1G 2 (G 1 + G 2)/2 . (52) VII. DRESSED STATES 13 We have pointed out that the manifold picture, based on the Hamiltonian dynamics, breaks in presence of non-negligible pumping. To define dressed states in such an essentially non- Hamiltonian system, one must recourse to another formalism, presented in this section, based on studying the flow of coher- ence in the system. This allows a reconstruction of dressed states that recovers the manifold picture for vanishing pump and extends it otherwise. To analyze how coherence flows in the system, let us ad- dress the dynamics of coherence between states, given by off- diagonal elements of the density matrix r . We thus consider the vector of the relevant transitions:   hu1i hl1i hu2i hl2i h1,1 r = h1,0 r h1,1 r  h0,1 r vcoh =  dvcoh/dt = −Mcohvcoh 0,1i  0,0i  1,0i 0,0i . (53) (54) The dynamics of this vector is ruled by a matrix Mcoh: that reads: Mcoh =  G 1 2 + g 2 −P2 G 1 −g 2 2 + P2 −ig 0 0 ig −ig 0 G 2 2 + g 1 − iD −g 1 0 ig −P1 G 2 2 + P1 − iD .   (55) The diagonal terms in Mcoh give the decay rate of each coher- ence and the off-diagonal terms, the rates at which coherences is transferred from one transition to another. Its eigenvalues recover the positions and broadenings of the spectrum peaks in Eq. (23), with eigenstates that I note hTpi, p = A,B,C,D. Therefore, the single time dynamics of hTpi is free: hTp(t)i = e−(iw p+g p/2)(t−t0)hTp(t0)i , (56) and their spectral shape is a Lorentzian: sTp(w ) = hT† p(w )Tp(w )i = hT† pTpi p g p/2 (g p/2)2 + (w − w p)2 . (57) In this way, one can reconstruct the dressed states from the above expressions by inspection of the hTpi. First, let us consider the simplest case of Hamiltonian dy- namics (without pumping and decay), for which we already obtained the dressed states through the manifold picture (by diagonalization of the Hamiltonian with only direct coupling g). The following discussion will thus reconstruct Fig. 2(b). The eigenenergy w A = w B = +ig corresponds to the two eigenstates: hTAi = hd1 + d2i hTBi = hu2 − u1i The eigenenergy w C = w D = −ig corresponds to the counter- (lower transition 0,0i ↔ +i) , (upper transition −i ↔ 1,1i) . (58a) (58b) part eigenstates: Now that we have described the new regimes appearing in this system due to the incoherent pumping, SSC and MC, and their spectral features, we can discuss their origin. hTCi = hd1 − d2i hTDi = hu2 + u1i (transition 0,0i ↔ −i) , (transition +i ↔ 1,1i) . (59a) (59b) 14 FIG. 12: (Color online) (a) Contributions from state 1,0i to the dressed states appearing in an optimally pumped case: P1 = g 2 = g , P2 = g 1 = 0. The x-axis is g /g, as in previous Figs. 8 and 9 describing the same case. In thin dashed black lines, the contribution to the 4LS dressed states ±i. In thick lines, the qubit dressed states: orange for O±i and green for I±i. The splitting of each pair of lines ± from equal to different contributions, marks their transition into WC. All the "−" states tend towards 1,0i in the very weakly coupled regime. Therefore, the system resonances hTpi give away the existence of the dressed states: ±i (cid:181) ± 1,0i+ 0,1i (60) (left not normalized for convenience here and further-on). We can clearly see by inspection of Mcoh how the direct coupling transfers coherence between the two lower transitions li, on the one hand, and the two upper transitions ui, on the other hand. In this case, coherence can follow a closed path forming loops that flow always at the same rate 2g (in between u1 ↔ u2 and d1 ↔ d2) and give rise to Rabi oscillations and dressed states. In the case of a 4LS, the situation is not much different, even when adding pump and decay: dressed states are also identified through the same procedure, diagonalizing the cor- responding matrix without incoherent crossed terms: G 1 2 + g 2 0 −ig 0 G 1 2 + P2 0 0 ig −ig 0 G 2 2 + g 1 − iD 0 Mcoh =  (61) In this case, the eigenenergies we found in Eq. (31) corre- spond to the same combination of transitions, Eqs. (58-59) with: 0 ig 0 G 2 2 + P1− iD .   ±i (cid:181) G − ± i R g 1,0i + i 0,1i . (62) If G coherence. Up to the transition into WC at G − = 0 ( R = g), we recover the dressed states without de- − = g, both coefficients in Eq. (62) are complex numbers with unit norm and, therefore, they give rise to a balanced contribution of the two states 1,0i and 0,1i equal to 1/2. Their mean energy, h± H ±i = ± R, is exactly the one appearing in the spectral resonances w p. The straightforward identification of dressed states, in the same way as when there is only direct coupling, makes it possible to describe the dynamics in the manifold picture as explained in appendix D. When the system is in WC ( R = i R), the dressed states become essentially the bare ones, 1,0i and 0,1i, since G −/g → ¥ . That is, in this regime, the populations become less and less balanced, as we can see in the example of Fig. 12, where the probability to find the system in 1,0i is plotted in thin dashed for both ±i. On the other hand, in the case of two qubits which are only pumped and decaying, with no direct coupling, the eigenstates of Mcoh are combinations of the transitions corresponding to each qubit: hP2u1 + g 2d1i, hP1u2 + g 1d2i, hu1 − d1i and hu2 − d2i. No new dressed states arise although coherence is being transferred (incoherently) between two transitions. From these three limiting situations, we learn that direct coupling is essential for the appearance of dressed states but also that pump and decay can transfer coherence in the direc- tions u1 ↔ d1 and u2 ↔ d2, in an incoherent continuous way. In spite of this, they do not induce oscillations in the dynam- ics. Even in the most symmetrical case where g 1 = P1 and g 2 = P2, pump and decay rather move around coherence in an FIG. 13: (Color online) Circulation of coherence among the transi- tions between bare states for two qubits in the case where g 1 = P2 = 0. A loop is formed that involves all the transitions, thanks to the pump- ing and decay terms that participate to the flow of coherence in a con- structive way. The direct coupling goes both in tune with the loop or against it. When g 2 = P1 = 2g the transfer is optimal. incoherent way, basically disrupting the Rabi oscillations, as we observed at the end of Sec. VI B: G ≤ g. Exceptionally, if the coherence transfer induced by pump and decay happens in a constructive, rather than independent, way as compared to the one induced by the direct coupling, Rabi oscillations and dressed states can be enhanced. This is the case of the optimally pumped configurations studied in Sec. VI A. Let us analyze, for instance, the best situation, g 2 = P1 = 0 and g 1 = P2 = g , where G = √2g. In this case, the coherence transfer in the system can take place in a loop that involves not only the direct coupling but also incoherent pro- cesses, as sketched in Fig. 13. This results in the appearance of four different resonances in the system, that can be written in terms of two pairs of new intermediate dressed states, I±i (cid:181) O±i (cid:181) g /2± iz1 g /2± iz2 g g 1,0i + i 0,1i , 1,0i + i 0,1i , (63a) (63b) named after inner and outer eigenenergies. They correspond to g ± iz1 and g ± iz2, respectively, where z1,2 are those in Eq. (47). The level scheme that these new coherence loops produce is plotted in Fig. 2(d). In SSC, the dressed states ap- pear in the eigentransitions as: hTAi = h I−ih1,1 +i 0,0ihI+ i , hTBi = h O−ih1,1 −i 0,0ihO+ i , hTCi = h I+ih1,1 +i 0,0ihI− i , hTDi = h O+ih1,1 −i 0,0ihO− i . (64a) (64b) (64c) (64d) The normalization of these dressed states also depends on the coupling region. In SSC, inner states are normalized by divid- ing bypg(2g− g ) and outer bypg(2g + g ), instead of √2g as was the case of the 4LS. These states do not have a bal- anced contribution from the bare states due to the saturation brought by pump and decay. The inner states have a smaller contribution from the first qubit than the outer states: h1,0 I±i2 = h1,0 O±i2 = g 1 1 2(cid:0)1− 2(cid:0)1 + 2g− g (cid:1) , 2g + g (cid:1) . g (65a) (65b) The 4LS dressed states correspond to an intermediate situation between the inner and outer dressed states in terms of popu- lations (equal to 1/2 for both qubit) and energy (see Fig. 10). This appears clearly in Fig. 12 where the first qubit contribu- tion to each dressed state is plotted as a function of g /g. The two transitions composing each of the eigentransitions in Eq. (63) have the same energy, which does not coincide with the average Hamiltonian value as soon as decoherence appears: hI± H I±i = ±´ (z2)/[1 + g /(2g)]. O±i = ±´ (z1)/[1 − g /(2g)] and hO± H When entering WC, the splittings between each pair of dressed states close, while the eigentransitions are linked each to a single state: hTAi = h I−ih1,1 +i 0,0ihI− i , hTBi = h O−ih1,1 +i 0,0ihO− i , hTCi = h I+ih1,1 +i 0,0ihI+ i , hTDi = h O+ih1,1 +i 0,0ihO+ i . (66a) (66b) (66c) (66d) Note that the inner peaks are always the most intense ones and that, when entering the WC, it is the "−" states that domi- nate the spectrum due to the saturation of the system into their limiting state 1,0i. VIII. CONCLUSIONS I described fully analytically the weak and strong coupling of two qubits in presence of dissipation and pumping. This 15 configuration is realized by two two-level systems that com- mute. This was contrasted with other models of quantum cou- pling between two modes, namely, two bosons (that commute) and two fermions (that anticommute). I also addressed the case of a four-level system (4LS), which has the same level structure. Whereas the coupling of bosons, fermions and a 4LS mani- fest a conventional phenomenology, well known from the nor- mal coupling of two modes and differing only in bosonic or fermionic effective parameters, I showed that cross-Lindblad terms in the two coupled qubits give rise to a rich and com- plex new nomenclature of coupling. Namely, beyond stan- dard strong coupling (SC) where two dressed states emerge from quantum superposition of the bare states, I found that four dressed states, I±i and O±i, can be realized in the system with two different Rabi splittings, as a result of pump- ing and decay establishing new paths of coherence flow in the system. This is a new paradigm of strong coupling that leads to new regimes that I called "Second order Strong Coupling" (SSC), when the four dressed states are split (in two pairs), and "Mixed Coupling" (MC), when one pair of dressed states, the inner I±i, have closed. In this wider picture, I called "First order Strong Coupling" (FSC) the conventional type of strong coupling with only one pair of dressed states. These results show and spell out the considerable complexity of coherence transfer in other systems such as the Jaynes-Cumming model in presence of incoherent pumping, that are not amenable to analytical treatment. Acknowledgments I thank F. P. Laussy for fruitful discussions and critical read- ing of the manuscript and T. Ostatnick´y for useful comments. This research was supported by the Newton Fellowship pro- gram. Appendix A: Interactions and dephasing If the two qubits are close enough in space, they may interact. The energy levels are affected by the interaction (Coulomb interaction in the case of atoms or excitons) and the Hamiltonian must be corrected with the term: s 1s † 2 † 1 s 2 , Hb = −cs (A1) where c is the binding energy between two excitations (c > 0 corresponds to attraction). If c ≪ w 1,2, interactions have a negligible effect on the populations and statistics, but not on the spectral features, starting with Eq. (4b) becoming w 11 = w 1 + w 2 − c . The dynamics in the presence of interactions is derived in appendix B although the expressions for spec- trum and dressed states are not presented in the text as they are not analytical in the most general case, and bring little to the discussion. The main effect that it produces is that upper and lower transitions can be identified thanks to the different energy of the upper transitions and this was used to check the consistency of the results. In FSC and WC, the difference is already present due to the broadenings associated to each tran- sition and interactions only accentuate it. However, in SSC and MC, interactions break the symmetry between upper and lower transitions (e.g., w A was linked to both upper or lower transition from inner states in Fig. 2(d)) depending on the sign of c : • if c > 0, w A,D correspond to the upper transitions and w B,C to the lower transitions • if c < 0, w A,D correspond to the lower transitions and w B,C to the upper transitions Naturally, strong interactions produce splittings between the peaks corresponding to upper and lower transitions, which may turn the spectral doublets into fully formed quadruplets. A second element that may be important in the system is pure dephasing. It may be caused, depending on the physi- cal system, by interaction with phonons or other external el- ements. This effect has proved to be relevant in state of the art semiconductor experiments [38]. Noting the rates of de- phasing of each qubit as d 1, d 2, one must solve the following extended master equation from Eq. (5): dr dt = L r + (cid:229) s i) . (A2) s i−s † s ir −rs (2s † i d i 2 s irs † i † i i i=1,2 In the formalism, as a consequence, Eq. (7a) becomes: 1 = g 1 + P1 + d 1 , G d G d 2 = g 2 + P2 + d 2 . (A3a) The dynamics in the presence of pure dephasing is again de- rived in appendix B and not discussed in the main text for the same reason as before, that it spoils analyticity of the re- sults and brings little to the discussion. The main effect of pure dephasing is to weaken the coherent coupling and is of a quantitative nature, not relevant enough to discuss it further in the present text. Appendix B: Quantum regression formula and time correlators In this appendix I obtain, thanks to the quantum regression formula, the first order differential equations for the one and two-time correlators required for the spectrum in Eq. (10). The quantum regression theorem states that, once we find the set of operators C{h } and the regression matrix M that sat- isfy Tr(C{h }L O) = (cid:229) }Tr(C{l }O) for a general op- erator O, then the equations of motion for the two-time corre- lators (t ≥ 0) read: {l } M{hl ¶ ¶t hO(t)C{h }(t + t )i = (cid:229) {l } M{hl }hO(t)C{l }(t + t )i (B1) The most general set of operators for the two coupled qubits is C{m,n,m ,n } = s † 2 , with m, n, m , n ∈ {0,1}. The ms n 1 s † 2 m s 1 n corresponding regression matrix M is: (m − n )2 , d 2 2 d 1 2 G 2 2 (m + n)− (m− n)2 − (m + n )− mnmn = iw 1(m− n) + iw 2(m − n )− ic (mm − nn ) Mmnmn G 1 − 2 M mnmn 1−m,1−n,mn M mnmn mn,1−m ,1−n M mnmn m,1−n,1−m ,n M mnmn 1−m,n,m ,1−n M mnmn 1−m,n,1−m ,n M mnmn m,1−n,m ,1−n = P1mn− ic (1− m)(1− n)(m − n ) , = P2mn − ic (1− m )(1− n )(m− n) , = 2ig(n − m )(1− n)(1− m ) , = 2ig(n− m )(1− n )(1− m) , = ig[m(1− m ) + m (1− m)] , = −ig[n(1− n ) + n (1− n)] , 16 (B2a) (B2b) (B2c) (B2d) (B2e) (B2f) (B2g) and zero everywhere else. 1 , as G(1) We concentrate on computing G(1) 2 can be obtained from it by exchanging the indexes 1↔ 2. It corresponds to set- ting O = s † 1 and having {m,n, m ,n } = {0,1,0,0} in Eq. (B1). Other two-time correlators are linked to the one of interest by the regression matrix, and therefore are also needed to com- pute the spectrum. All correlators can be grouped in mani- folds that, following Refs [14, 16, 36], I denote Nk, where k is the minimum number of particles that allows the corre- lators to take a nonzero value. As we can see schematically in the left part of Fig. 14, the first manifold N1 is composed of the correlators with {0,1,0,0} and {0,0,0,1}, linked only by the coherent coupling (red arrows). In the linear regime (very low pump), it is enough to solve the equations trun- cating in this manifold, which is the same as in the linear 1 (t ). The next (and last) manifold N2, model and includes G(1) where the two models differ, is composed of the correlators with {1,1,0,1} and {0,1,1,1}. The links between manifolds are also of an incoherent nature, due to the pump (green and blue arrows). Gathering the four correlators in a vector v(t,t + t ) =  hs † 1 (t)s 1(t + t )i  1 (t)s 2(t + t )i hs †  s 1s 2(t + t )i hs † 1 (t)s † hs † 1 (t)s 1s † s 2(t + t )i 1 2 their equations of motion read in matricial form: ¶ ¶t v(t,t + t ) = −M1v(t,t + t ) , , (B3) (B4) (B5a) with M1 = G d 1 2 iw 1 + ig 0 −P2   G d 2 2 ig iw 2 + −P1 0 −2ig −ic i(w 2 − c ) + −ig 2G 1+G d 2 2 −ic −2ig −ig i(w 1 − c ) + 2G 2+G d 1 2 .   with 2 1 1 2 u(t) =  hs † hs † hs † hs † G 1 0 G 2 0 ig −ig −ig ig M0 =  s 1i(t)  s 2i(t)  s 2i(t) s 1i(t) ig −ig G d 1+G d 2 − iD 2 0 , P1 P2 0 0 p =  −ig ig 0 1+G d G d 2 + iD 2 ,     17 (B8a) . (B8b) FIG. 14: (Color online) Chain of correlators -- indexed by {h } = (m, n, m ,n ) -- linked by the Hamiltonian dynamics with pump and the decay for two coupled qubits. On the left (resp., right), Nk) involved in the equations of the two-time set Sk Nk (resp., Sk (resp., single-time) correlators. In green are shown the first man- N1 that correspond to the linear model [14], and ifolds N1 and N2. The equation of mo- in blue, the second manifold N2 and tion hs † 1 (t)C{h }(t + t )i with h ∈ Nk requires for its initial value the correlator hC{ h }i with { h } ∈ Nk defined from {h } = (m, n, m ,n ) by { h } = (m + 1, n, m ,n ), as seen on the diagram. The thick red ar- rows indicate which elements are linked by the coherent (SC) dynam- ics, through the coupling strength g, while the green/blue thin arrows show the connections due to the incoherent quantum dot pumpings. The sense of the arrows indicates which element is "calling" which in its equations. The self-coupling of each node to itself is not shown. This is where w 1,2 and G 1,2 enter. These links are obtained from the rules in Eqs. (B2), that result in the matrices M1 and M0. The number of correlators needed to compute the spectrum is truncated naturally (with four elements) thanks to the saturation of the qubit. The general solution (for t ≥ 0), v(t,t + t ) = e−M1t v(t,t), leads to the correlator in Eq. (12) with only four contributions, indexed by p = A, B, C and D. The frequencies w p and rates g p are the imaginary and real parts of the eigenvalues of the ma- trix M1. The dimensionless coefficients Lp(t) and Kp(t), also real, are linear combinations of the single-time average quan- tities, This equation is similar to that of the linear model, only changing the effective fermionic rates into bosonic ones G i → G i [14]. On the other hand, the correlator with {1,1,1,1} (that is, hs † s 2i(t)), which is the only one that is not N2, finds its expression separately, only in terms of zero in the operators labeled {1,1,0,0} and {0,0,1,1}, through the equation s 1s † 2 1 d dt hs † 1 s 1s † 2 s 2i(t) = −4G +hs † +P2hs † s 1s † 2 s 1i(t) + P1hs † s 2i(t) s 2i(t) . 1 1 2 (B9) In order to insert these averages in the expression (10) for the spectrum, they must be either time integrated, in the sponta- neous emission case (SE), to give vSE =R 0 v(t,t)dt (and the coefficients LSE p ) or computed directly in the steady state (SS) to give vSS = limt→¥ v(t,t) (and the coefficients p , KSS LSS p ). p and KSE 1. Two fermions The regression matrix M1 (cf Eq. (B5)) truncates to its first submatrix, 1 s 1i(t) hs † v(t,t) =  s 2i(t) hs † 1   s 1s 2i(t) s † hs † 1 s 1s † hs † s 2i(t) 2 1 1 . (B6) M1 = iw 1 + ig G d 1 2 ig iw 2 + 2 ! . G d 2 (B10) recovering the exact mapping to the linear model, as discussed in the text. These mean values can also be found through the quantum re- gression formula, setting the operator O = 1 and deriving the corresponding regression matrix from the rules in Eq. (B2). Nk. All They can also be grouped in manifolds, denoted nonzero single-time average quantities, all of them needed to compute the spectrum, are shown in the right part of Fig. 14. N1 can be obtained independently solving a second matricial equation: s 1s 2i(t) = 0. The correlators in Note that hs † s † 1 1 du(t) dt = −M0u(t) + p (B7) 2. Matrix of coherences The matrix of coherence in Eq. (54) can be found by rewrit- ing the regression matrix M1 in the basis of projectors for each transition, presented in Eq. (53): Mcoh = A−1M1A with A some transformation matrix. For this reason, they have the same eigenvalues, which are the resonances of the sys- tems. Since the eigenstates of Mcoh are the transitions be- tween dressed states, their study enables us to reconstruct them even in a highly dissipative environment. ¥ Appendix C: Cross Lindblad terms In the case where the four levels of the system do not cor- respond to two qubits but are the levels of a single entity [32], the corresponding Liouvillian of the system is most conve- niently written in terms of the four-level operators in Eq. (6). The final master equation in that case can be found from L r in Eq. (5) by removing cross terms that appear in the two cou- pled qubits: dr dt = L r −g 1(l1r u† 1 + u1r l† r l1 + l† 1 1l1r u† 2l2r u† 1)− g 2(l2r u† r u1)− P2(u† 2 1u1r l† 1l1) 2u2r l† 2l2) . 1u1 + u† 2u2 + u† 1 −P1(u† −d 1(l† −d 2(l† 2 + u2r l† 2) r u2) r l2 + l† 2 (C1) (C2) (C3) (C4) Such cross terms represent effective couplings between the coherences, in the case of pump and decay, and the states, in the case of pure dephasing, associated to each qubit. Ev- idently, they can only have consequences on the dynamics when the two qubits are also directly coupled. The rules to add to those in Eq. (B2) in order to obtain a new general re- gression matrix M + d M are: d Mmnmn mnmn = P1(m + n− 1)(m − n )2 + P2(m + n − 1)(m− m)2 , (C5a) d M mnmn 1−m,1−n,mn d M mnmn mn,1−m ,1−n = −(m − n )2[(P1 + d 1)mn− G 1(1− m)(1− n)] , (C5b) = −(m− n)2[(P2 + d 2)mn − G 2(1− m )(1− n )] . (C5c) With these new rules, the matrix for the one-time correlators, M0, is still the same as in Eq. (B8b) and, therefore, the mean values in u(t), Eq. (B8a), do not differ from those for two qubits. The two-time regression matrix M1 changes into M1 + d M1 with d M1 =  P2 0 0 d 2 + P2 0 0 g 1 − P1 P1 d 1 + P1 −P1 0 0 g 2 − P2 0 0 −P2 .   Appendix D: Dressed states in the manifold picture We can give an alternative derivation and intuitive interpre- tation of the results in section V A by assuming complex fre- to the complex dressed frequencies: quencies w 1,2 → W 1,2 = w 1,2 − ig 1,2/2 in Eq. (3). This leads w + − → W + w 11 → W = w 1 − ig + ± R0 , − 11 = 2w 1 − 2ig + . (D1b) (D1a) From here, we can compute positions and broadenings of the four possible contributions to the spectra. The positions are 18 given by the difference in energy between the levels involved in the transitions while the broadenings are given by the sum of the imaginary parts (the uncertainties). If we apply this principle between the levels in Eq. (D1), we find the same result than in Eq. (34) -- (35). This allows us to identify the contributions A, B, C, D to the SC spectrum with the mani- fold transitions in the order that we now explain. In Fig. 2(b) we can follow the four possible transitions in the manifold picture. The lower transitions (in blue), from ±i, coincide, respectively, with the expressions for peaks A and D, given by Eq. (34). The upper transitions (in red), towards ±i, coincide with the expressions for peaks B and C, given by Eq. (34). The upper transitions have a larger broadening than the lower due to the addition of the uncertainties in energy of the levels in- volved, brought by the spontaneous decay. In order to understand better some features of the spectra of the steady state under incoherent pump in the different regions that we have defined in section V B, we will now attempt to push the manifold method -- adequate for vanishing pump -- a bit further, to the case of non-negligible pump. If we combine their effects in the non-hermitian Hamiltonian as P1 + P2 , 2 0,0 = −i ± = w 1 − iG + ± R1TD , 11 = 2w 1 − i g 1 + g 2 2 , (D2a) (D2b) (D2c) and apply their sum and subtraction to obtain positions and broadenings of the spectral contributions of each of the four transitions: g A 2 g B 2 g C 2 g D 2 + iw A = + iw B = + iw C = + iw D = 3(P1 + P2) + g 1 + g 2 4 3(g 1 + g 2) + P1 + P2 4 3(g 1 + g 2) + P1 + P2 4 3(P1 + P2) + g 1 + g 2 4 + iR1TD , + iR1TD , − iR1TD , − iR1TD , (D3a) (D3b) (D3c) (D3d) we do not obtain the right results of four -- in principle different -- peaks. This discussion brings us back to the naive association of SC with ´ (R1TD) 6= 0 we discarded in Sec. III. It is evident that using only the single-time dynamics fre- quency R1TD to describe the splitting between the dressed states, cannot give the variety of situations than the two com- plex parameters z1,2 give. In the case of a 4LS, the results obtained in Eq. (31) are comparable to those in Eq. (D3), even though the positions of the peaks are given by R instead of R1TD. The broaden- ings are exactly the same. We can conclude that without the cross Lindblad terms, the coupling acquires a bosonic charac- ter: the correlations between levels are relaxed from those of two qubits into those of one (larger) four-level system where the manifold picture holds. W W W Appendix E: Second order correlations functions In terms of the following vector of mean fluctuations, 19 In this appendix, I obtain the second order direct and cross correlation functions defined, respectively, as 1 (t,t + t ) = hs † G(2) 12 (t,t + t ) = hs † G(2) 1 (t)s † 1 (t)s † 1 (t + t )s 1(t + t )s 1(t)i , 2 (t + t )s 2(t + t )s 1(t)i . (E1a) (E1b) These two quantities are proportional to the probabilities of two t -delayed emissions from the same qubit (G(2) 1 ) or two different qubit emissions (G(2) 12 ). FIG. 15: (Color online) Steady state values of g(2)(t ) (solid blue) 12 (t ) (dashed purple) at resonance. (a) The system is in FSC: g 1 = g(2) g, g 2 = g/2, P1 = g/2 and P2 = g/10. (b) In MC with optimum coupling G = √2g: g 1 = P2 = 0 and P1 = g 2 = 2g (see Fig. 8). In order to compute such two-time correlators, we apply the quantum regression formula presented in an alternative form to Eq. (B1): ¶ ¶t }hO(t)C{l }(t + t )Q(t)i hO(t)C{h }(t + t )Q(t)i = (cid:229) {l } (E2) where Q is a second arbitrary operator. The equations we must solve setting O = s † 1 and Q = s 1, are the same as those used to compute the mean values in Eq. (B7) but now with the vector of correlators M{hl w(t,t + t ) =  1 1 (t)s † hs † 1 (t)s † hs † hs † 1 (t)s † hs † 1 (t)s 1s † s 1(t + t )s 1(t)i  s 2(t + t )s 1(t)i  s 2(t + t )s 1(t)i 2 (t + t )s 1(t)i 2 1 , (E3) and the vector n1(t)p instead of p. In the steady state (t = 0), the solution for w(0,t ) requires knowing the initial condition: wSS = w(0,0) =  0 n11 0 0 .   (E4) fSS = −(wSS − uSSnSS , (E5) the steady state normalized correlations functions (for t > 0) read 1 ) =  n2 1 n1n2 − n11 ncorrn1 n∗corrn1   g(2)(t ) = 12 (t ) = g(2) 1 (0,t ) G(2) n2 1 12 (0,t ) G(2) n1n2 = 1− = 1− [e−M0t fSS]1 n2 1 ∈ [0,1], (E6a) [e−M0t fSS]2 n1n2 ∈ [0,1] (E6b) where [x]1 means that we take the first element of the vec- tor x. At infinite delays, the two functions reach the value of uncorrelated emissions, 1 (t → ¥ ) , g(2) g(2) 12 (t → ¥ ) → 1 . (E7) At zero delay, it is evident that g(2)(0) = 0, as two excita- tions cannot exist in the same qubit. At intermediate delays, the emission presents antibunching: g(2)(0) < g(2)(t ), the sec- ond excitation from the same qubit has more probability to be emitted after some delay (see Fig. 15). At zero delay, g(2) 12 (0) = n11 n1n2 . (E8) It always is possible a second emission coming from a differ- 12 (t = 0) is 1/5). When ent qubit (the minimum value for g(2) 12 (t ) = 1 all along. the qubit behave as independently, g(2) Being the second order correlation functions given by basi- cally the same equation as the populations and depending only 12 (t ) are the same for the two coupled on them, g(2) fermions, qubits and 4LS. It is not possible to distinguish from their dynamics the four coupling regions described in Table I, appearing in the case of coupled qubits. 1 (t ) and g(2) [1] M. A. Nielsen and I. L. Chuang, Quantum computation and [4] L. Allen and J. H. Eberly, Optical Resonance and Two-Level quantum information (Cambridge University Press, 2000). Atoms (Dover, 1987). [2] G. W. Gardiner and P. Zoller, Quantum Noise (Springer-Verlag, [5] A. Kavokin, J. J. Baumberg, G. Malpuech, and F. P. Laussy, Berlin, 2000), 2nd ed. [3] S. Haroche and J.-M. Raimond, Exploring the Quantum: Atoms, Cavities, and Photons (Oxford University Press, 2006). Microcavities (Oxford University Press, 2007). [6] C. Cohen-Tannoudji, J. Dupont-Roc, and G. Grynberg, Photons et atomes (EDP Sciences, 2001). 20 [7] J. J. Sanchez-Mondragon, N. B. Narozhny, and J. H. Eberly, Phys. Rev. Lett. 51, 550 (1983). Katzer, D. Park, C. Piermarocchi, and L. J. Sham, Science 301, 809 (2003). [8] H. J. Carmichael, R. J. Brecha, M. G. Raizen, H. J. Kimble, and [26] I. A. Grigorenko and D. V. Khveshchenko, Phys. Rev. Lett. 94, P. R. Rice, Phys. Rev. A 40, 5516 (1989). [9] B. R. Mollow, Phys. Rev. 188, 1969 (1969). [10] C. Cohen-Tannoudji and S. Reynaud, J. phys. B.: At. Mol. Phys. 10, 345 (1977). [11] J. I. Cirac, H. Ritsch, and P. Zoller, Phys. Rev. A 44, 4541 (1991). [12] L. Tian and H. J. Carmichael, Quantum Opt. 4, 131 (1992). [13] E. del Valle, Microcavity Quantum Electrodynamics (VDM Verlag, 2009). [14] F. P. Laussy, E. del Valle, and C. Tejedor, Phys. Rev. B 79, 235325 (2009). [15] E. Jaynes and F. Cummings, Proc. IEEE 51, 89 (1963). [16] E. del Valle, F. P. Laussy, and C. Tejedor, Phys. Rev. B 79, 235326 (2009). 040506 (2005). [27] Y. M. Galperin, D. V. Shantsev, J. Bergli, and B. L. Altshuler, Europhys. Lett. 71, 21 (2005). [28] R. H. Dicke, Phys. Rev. 93, 99 (1954). [29] M. I. Shirokov, J. phys. B.: At. Mol. Phys. 23, 1923 (1990). [30] A. Messiah, Quantum Mechanics (Dover, 1999). [31] G. Morigi, S. Franke-Arnold, and G.-L. Oppo, Phys. Rev. A 66, 053409 (2002). [32] E. del Valle, S. Zippilli, F. P. Laussy, A. Gonzalez-Tudela, G. Morigi, and C. Tejedor, Phys. Rev. B 81, 035302 (2010). [33] B. D. Gerardot, S. Strauf, M. J. A. de Dood, A. M. Bychkov, A. Badolato, K. Hennessy, E. L. Hu, D. Bouwmeester, and P. M. Petroff, Phys. Rev. Lett. 95, 137403 (2005). [34] H. J. Carmichael, Statistical methods in quantum optics 1 [17] Y. A. Pashkin, O. Astafiev, T. Yamamoto, Y. Nakamura, and (Springer, 2002), 2nd ed. J. S. Tsai, Quantum Information Processing 8, 55 (2009). [18] Y. A. Pashkin, T. Yamamoto, O. Astafiev, Y. Nakamura, D. V. [35] H.-J. Briegel and B.-G. Englert, Phys. Rev. A 47, 3311 (1993). [36] E. del Valle, F. P. Laussy, and C. Tejedor, AIP Conference Pro- Averin, and J. S. Tsai, Nature 421, 823 (2003). ceedings 1147, 238 (2009). [19] J. B. Majer, F. G. Paauw, A. C. J. ter Haar, C. J. P. M. Harmans, [37] F. P. Laussy, E. del Valle, and C. Tejedor, Phys. Rev. Lett. 101, and J. E. Mooij, Phys. Rev. Lett. 94, 090501 (2005). [20] T. Hime, P. A. Reichardt, B. L. T. Plourde, T. L. Robertson, C.- E. Wu, A. V. Ustinov, and J. Clarke, Science 314, 1427 (2006). [21] A. O. Niskanen, K. Harrabi, F. Yoshihara, Y. Nakamura, S. Lloyd, and J. S. Tsai, Science 316, 723 (2007). [22] W. K. Wootters, Phys. Rev. Lett. 80, 2245 (1998). [23] A. J. Berkley, H. Xu, R. C. Ramos, M. A. Gubrud, F. W. Strauch, P. R. Johnson, J. R. Anderson, A. J. Dragt, C. J. Lobb, and F. C. Wellstood, Science 300, 1548 (2003). [24] T. Yamamoto, Y. A. Pashkin, O. Astafiev, Y. Nakamura, and J. S. Tsai, Nature 425, 941 (2003). [25] X. Li, Y. Wu, D. Steel, D. Gammon, T. H. Stievater, D. S. 083601 (2008). [38] A. Laucht, N. Hauke, J. M. Villas-Boas, F. Hofbauer, G. Bohm, M. Kaniber, and J. J. Finley, Phys. Rev. Lett. 103, 087405 (2009). [39] When taking square roots, we always follow the prescription √c2 = c, for c ∈ C. Considering this solution is enough because in any case all quantities depend on both ± [40] That is, pictorially, when the structure of levels and transitions are rotated by 180◦ or, mathematically, when the rising and low- ering operators are inverted. √c2. 

1807.07312

1

1807

Spin-wave amplification and lasing driven by inhomogeneous spin transfer torques

We show that an inhomogeneity in the spin-transfer torques in a metallic ferromagnet under suitable conditions strongly amplifies incoming spin waves. Moreover, at nonzero temperatures the incoming thermally occupied spin waves will be amplified such that the region with inhomogeneous spin transfer torques emits spin waves spontaneously, thus constituting a spin-wave laser. We determine the spin-wave scattering amplitudes for a simplified model and set-up, and show under which conditions the amplification and lasing occurs. Our results are interpreted in terms of a so-called black-hole laser, and could facilitate the field of magnonics, that aims to utilize spin waves in logic and data-processing devices.

cond-mat.mes-hall

cond-mat

Spin-wave amplification and lasing driven by inhomogeneous spin transfer torques R.J. Doornenbal,1 A. Rold´an-Molina,2 A.S. Nunez,3 and R. A. Duine1, 4 1Institute for Theoretical Physics, Utrecht University, Leuvenlaan 4, 3584 CE Utrecht, The Netherlands 2Universidad de Aysn, Calle Obispo Vielmo 62, Coyhaique, Chile 3Departamento de F´ısica, Facultad de Ciencias F´ısicas y Matem´aticas, Universidad de Chile, Casilla 487-3, Santiago, Chile 4Department of Applied Physics, Eindhoven University of Technology, P.O. Box 513, 5600 MB Eindhoven, The Netherlands (Dated: January 22, 2020) We show that an inhomogeneity in the spin-transfer torques in a metallic ferromagnet under suitable conditions strongly amplifies incoming spin waves. Moreover, at nonzero temperatures the incoming thermally occupied spin waves will be amplified such that the region with inhomogeneous spin transfer torques emits spin waves spontaneously, thus constituting a spin-wave laser. We determine the spin-wave scattering amplitudes for a simplified model and set-up, and show under which conditions the amplification and lasing occurs. Our results are interpreted in terms of a so-called black-hole laser, and could facilitate the field of magnonics, that aims to utilize spin waves in logic and data-processing devices. PACS numbers: 85.75.-d, 75.30.Ds, 04.70.Dy Introduction. -- Spin waves are collective excitations in magnetically-ordered materials. At the semi-classical level, spin waves in ferromagnets correspond to a wave- like pattern of precessing spins in which the relative phase of the precession of two spatially separated spins is determined by the ratio of their distance to the wave- length of the spin wave. When the exchange interactions dominate, the spin precession is circular. Anisotropies and dipolar interactions, however, generically lead to elliptically-precessing -- in short, elliptical -- spin waves. Though spin waves are neutral excitations, they are able to transfer angular momentum. Magnonics [1, 2] is named after the quasi-particle, the magnon, associ- ated with a spin wave. This field has the ultimate goal of controlling and manipulating spin waves to the point that they can be used to realize energy-efficient data-processing and logic devices. One hurdle to real- ize technology based on spin waves is that they have a finite lifetime as a result of processes that lead to loss of spin angular momentum and relax the magnetization. Hence, experimental progress has been nearly exclusively made using a unique low magnetic-damping material: the complex magnetic insulator Yttrium Iron Garnet (YIG), thereby limiting the process as it is difficult to fabricate and pattern at high quality in reduced dimensions. The relaxation of spin waves can be counteracted by injection of spin angular momentum to compensate for the losses. This has been demonstrated in YIG/Pt-based material systems [3], in which a charge current, driven through the Pt and tangential to the interface with YIG, excites a -- via the spin Hall effect [4] -- spin current that is absorbed by the magnetization in the magnetic insula- tor. A similar result has been obtained with the magnetic metal permalloy and Pt [5]. In a different implementation [6], the spin current was injected by a thermal gradient FIG. 1: The set-up that is considered in this Letter: A ferro- magnetic wire with magnetization saturated in the z-direction is subjected to a current density driven in the long direction of the wire. The wire has an indent such that the current density in this narrow region is larger than in the wider parts of the wire. For large enough currents, the resulting current- induced spin-wave Doppler shift pushes the spin-wave energies indicated in red to negative values in the narrow part of the wire while the spin-wave energies in the wide parts of the wire remain positive. The spin-wave frequency ω is sketched as a function of the wave number k in the three different regions. via the spin Seebeck effect [7], and increased spin-wave propagation lengths were also observed. In these exam- ples, the amplitude enhancement of the spin waves is proportional to the applied charge current or tempera- ture gradient, which may be a limiting factor in case the damping that needs to be overcome is large, or because of the associated heating. In this Letter, we propose a different way to amplify spin waves. It makes use of the spin transfer torques that arise in the bulk of ferromagnetic metals as a result of the interaction between the spin-polarized electronic current and the magnetization [8]. The basic set-up we 8 1 0 2 l u J 9 1 ] l l a h - s e m . t a m - d n o c [ 1 v 2 1 3 7 0 . 7 0 8 1 : v i X r a ferromagnetic metal electronszxykkkwwwvs<vscvs>vscvs<vscRT1x=0x=Lconsider is sketched in Fig. 1. It consists of a ferromag- netic metallic wire with a constriction. A charge current driven through the wire will have a larger current den- sity in the narrow part of the wire, as compared to the wider parts. As a result, the velocity that characterizes the spin transfer torques and determines, for example, the current-induced spin-wave Doppler shift [9] will be larger in the narrower part of the wire. For sufficiently large current densities, the Doppler shift will make the spin-wave energies and frequencies in the narrow part of wire negative (indicated in red in Fig. 1), while in the wider parts of the wire they remain positive. Moreover, in the case of a finite spin-wave ellipticity, the spin waves with positive and negative energy are coupled in the re- gions where the width of the wire changes. As a result of this coupling, spin waves can be created simultaneously in the wide and narrow parts of the wire without chang- ing the total magnetic energy. For the spin-wave modes that bounce back and forth in the narrow part of the wire, the coupling between negative-energy and positive- energy spin waves can be resonant. We show that this leads to an enormous enhancement of the transmission and reflection amplitudes for spin waves that are scat- tered off the narrow part of the wire and that fulfil the resonance condition. Moreover, this enhancement is so large that the thermally-excited spin waves that fulfil the resonance condition will be amplified and emitted from the narrow part of the wire and overwhelm all other spin waves, thus constituting a spin-wave laser. Below, we discuss our set-up in more detail and provide analytical results to underpin our proposal. Model and set-up. -- We consider a ferromagnetic metal subject to an external field B in the z-direction, and charge current j. The magnetization direction n(x, t) obeys the Landau-Lifshitz-Gilbert equation with spin transfer torques that is given by [8] (cid:18) (cid:19) (∂t +vs·∇)n = n×Heff −αn× ∂t + vs · ∇ β α n . (1) (cid:21) (cid:20) (cid:90) dV a3 E = This equation describes precession of the magnetization around the effective field Heff = −δE/δn, where E is the magnetic energy functional. We consider a generic energy functional that incorporates exchange, anisotropy, and the external field and is given by − 1 2 Jn · ∇2n + 1 2 Kxn2 x + Kyn2 y − Bnz 1 2 . (2) In this expression, a is the size of the unit cell, J is the exchange constant, and Kx and Ky are anisotropy con- stants. In case that the anisotropy constants are equal, the spin waves are circular. Due to shape and crystalline anisotropy the spin wave are typically elliptical, corre- sponding to the case that Kx (cid:54)= Ky. The spin transfer torques in the Landau-Lifshitz- Gilbert equation (1) are characterized by the velocity 2 vs = −gP µBj/2eMs, that is proportional to the current density and further determined by the current polariza- tion P , the Land´e factor g, the Bohr magneton µB, the elementary charge e, and the saturation magnetization Ms. The adiabatic spin transfer torque appears on the left-hand side of Eq. (1), whereas the non-adiabatic spin transfer torques appears on the right-hand side and is parametrized by the dimensionless constant β (cid:28) 1. The Gilbert damping constant α (cid:28) 1 determines the rate of decay of the magnetization direction. We take Kx, Ky, B > 0 so that the equilibrium di- rection of magnetization is the z-direction. Linearizing around this equilibrium direction yields the dispersion relation ωk − vs · k = ω0 k − iαω0 k − i(α − β)vs · k , with k =(cid:112)(ω0)2 + 2Jω0k2 + J 2k4 − ∆2 , ω0 (3) (4) the real part of the spin-wave dispersion in the absence of current, and where ω0 ≡ B + (Kx + Ky)/2 and ∆ ≡ (Ky − Kx) /2 and we assumed ∆ > 0 without loss of generality. This parameter is too some extent tunable by the wire geometry. In deriving the above dispersion relation we took vs constant, but we will drop this as- sumption shortly. (cid:113)ω0 +(cid:112)(ω0)2 − ∆2, the For vs > vs,c =(cid:112)2J/2 real part of the dispersion in Eq. (3) becomes negative. This signals an energetic instability as the system may lower its energy by creating negative-energy excitations. From now on, we assume that β ≈ α. This implies that the system remains dynamically stable even when it is energetically unstable, because small-amplitude fluctua- tions are damped out. This results from the imaginary part of the dispersion relation in Eq. (3) which remains negative when β ≈ α. In the remainder of this Letter we consider the system sketched in Fig. 1, in which a local increase in the ve- locity vs is accomplished by a narrow region in a wire of the metallic ferromagnet. Moreover, we assume that the current density is such that vs is above the critical value vs,c in the narrow part of the wire, whereas it is below the critical value in the wider parts of the wire. The result- ing local spin-wave dispersions are also sketched in Fig. 1 and correspond approximately to shifted parabolas. The negative-energy modes in the narrow part of the wire are indicated by the red dispersion curve. Scattering solutions. -- We now proceed to construct spin-wave scattering solutions to the Landau-Lifshitz- Gilbert equation. We neglect in first instance the mag- netization relaxation and put α = β = 0. We assume, moreover, that the transverse dimensions of the wire in the y and z-direction are very small so that we may drop the dependence of n on y and z, and may, moreover, take 3 vs = vs(x)x. We further assume that the regions where the wire becomes wider and narrower are very small so that we may put vs ≡ vL < vs,c independent of x in the wider part of the wire and vs ≡ vM > vs,c independent of x in the narrow part of the wire. We take the nar- row part of the wire between x = 0 and x = L. As we discuss later on when we give an interpretation of our results, the assumptions on the transverse dimensions of the wire and on the position dependence of vs(x) are not essential but are made to facilitate the computations. To construct the scattering solutions it is convenient to introduce ψ ≡ nx − iny. The linearized solutions of the Landau-Lifshitz-Gilbert equation are then given by ψ(x, t) = u(x)e−iωt − v∗(x)eiωt , (5) where u(x) and v(x) obey (cid:18)ω + ivs(x)∇ + J∇2 − ω0 −∆ −∆ −ω − ivs(x)∇ + J∇2 − ω0 (cid:19)(cid:18)u(x) (cid:19) v(x) = 0 . (6) Because we have taken vs(x) to be piecewise constant, the above equations for u and v are solved by a plane- wave ansatz in both the central narrow region of the wire, and in its wider parts. Scattering solutions are now con- structed by taking a spin wave incoming from the left, and calculating its reflection and transmission amplitude, by matching at the position where vs changes. The plane-wave solutions take the form (cid:18)u(x) (cid:19) v(x) (cid:18)F (cid:19) G = eikx , where F, G are complex coefficients. Using Eq. (6) we obtain the dispersion relation (ω − vsk)2 = (ω0)2 − ∆2 + 2Jω0k + J 2k4 . (7) (8) From now on we take ω > 0 without loss of generality. At a given frequency ω, there are in general four (complex) values of k that satisfy the dispersion. These are denoted by ki, with i = 1, 2, 3, 4 and are labelled according to Fig. 2. Different ω regimes need to be distinguished. Firstly, for vs < vs,c, the dispersion exhibits a gap ωmin. We require ω ≥ ωmin in order for scattering solutions to exist. For the regions where vs < vs,c there are then two propagating modes with real k and two growing/decaying modes with imaginary k. Secondly, for vs > vs,c, there exists a range of ω from ωmin to ωmax within which there exist four real wave vectors k that satisfy the dispersion relation. For ω exceeding ωmax, two of the solutions for k are real and two are imaginary. In what follows we will look at scattering solutions and hence assume ω > ωmin. The coefficients of the respective growing modes for x < 0 and x > L must vanish. In addi- tion, we impose matching conditions at both jumps in vs. The functions u(x) and v(x), as well as their first deriva- tives, are required to be continuous. This leads to a sys- tem of linear equations that can be solved for the reflected and transmitted amplitudes. Here, the reflection ampli- tude R is defined as the ratio between the F -amplitudes FIG. 2: Dispersion relation for vs < vs,c (red) and vs > vs,c (blue). Black dots mark the four wavenumbers corresponding to a generic value ω < ωmax in the vs > vs,c sector. Parameter values: vL = 0.5(cid:112)2Jω0/, vM = 3.0(cid:112)2Jω0/, ∆ = 0.3 ω0. [Eq. (7)] of the incoming and reflected wave, whereas the transmission amplitude T is defined as the same ratio but for incoming and transmitted wave. (One could also consider similar ratios of the G-amplitudes. This choice does not affect the location of the resonances.) Results. -- In Fig. 3 we show the results for the spin- wave transmission and reflection probabilities as a func- tion of frequency, with the choice L = 12(cid:112)2J/ω0. For distinct frequencies both reflection and transmission are strongly enhanced. We refer to the peaks in the reflec- tion and tranmission amplitudes as resonances. We have found that the resonance condition is well approximated by the equation k4 − k3 = + O 2πn L (9) (cid:18) 1 (cid:19) L2 where n = 1, 2, . . .. The first term of this equation has the physical interpretation that the counterpropagating 4 FIG. 3: Transmission (broken line) and reflection (solid line) probabilities off the spin-transfer torque inhomogeneity, see Fig. 1, as a function of spin-wave frequency. Parameter values: vL = 0.5(cid:112)2Jω0/, vM = 3.0(cid:112)2Jω0/, ∆ = 0.7 ω0. waves corresponding to k3 and k4 interfere constructively between x = 0 and x = L. Near ω = ωmax, the disper- sion relation is approximately parabolic. This allows us to describe the resonant frequencies ωn with the simple approximate formula ωn = ωmax − n2Γ2, (10) ωmax − ωmin/Γ(cid:99) and Γ ≈ π(cid:112)J/L2 + O(cid:0)1/L2(cid:1). The value of Γ is only weakly sensitive to the where n = 1, ...,(cid:98)√ parameters vL, vM , ∆. Varying the parameters vL, vM affects the location of the resonances significantly only by shifting them all, via a change in the value of ωmax. For ω > ωmax, there are no resonances at all. This is explained by the fact that the wave vectors k3, k4 are complex in this case. In Fig. 4 we have zoomed in on a single resonance peak for different values of ∆. We find that smaller ∆ results in higher and narrower peaks, but no resonances are present if ∆ = 0. Given the strong enhancement of the reflection and transmission amplitudes, the occupation of the incom- ing spin waves by thermal fluctuations leads to emission of amplified spin waves thus forming a spin-wave laser. We denote with un and vn the scattering solutions cor- responding to the resonant energies for a spin wave in- coming from the left. Furthermore, u(cid:48) n denote the resonant scattering solutions for spin waves incom- ing from the right. These occur at the same energies as the scattering solutions for waves incoming from the left, which follows from the condition in Eq. (9). We have that the magnetization direction, parametrized by ψ(x, t), is n and v(cid:48) FIG. 4: Transmission probabilities near a resonance, for dif- ferent ∆. The resonance peak is sharper and higher for small ∆ > 0, but is not present if ∆ = 0. Parameter values: vL = 0.5(cid:112)2Jω0/, vM = 3.0(cid:112)2Jω0/. ψ(x, t) ∝(cid:88) nB (ωn)(cid:2)un(x)e−iωnt − v∗ approximately given by n(x)e−iωnt − (v(cid:48) u(cid:48) n(x)eiωnt n)∗(x)eiωnt(cid:3) . (11) n √ ωmax − ωmin/Γ < 2, there is only a single reso- If 1 < nance peak. Based on this, one can experimentally en- gineer spin-wave laser that emits spin waves at a single frequency. Discussion and outlook. -- Our results can be inter- preted as follows. The left transition region, i.e., where vs changes from vs < vs,c to vs > vs,c, is a black-hole event horizon for spin waves [10, 11] coming in from the left (referring to Fig. 1). The right transition from vs > vs,c to vs < vs,c is a white-hole horizon for spin waves com- ing from the right. Because the spin waves do not dis- perse linearly, these horizons are referred to as disper- sive horizons [12]. In the field of analogue gravity, such a pair of black-hole-like and white-hole-like event hori- zons, discussed in Ref. [11] for spin waves, is known to be able to give rise to so-called black-hole lasers [13, 14], which exhibit the resonant amplification and lasing we have discussed in our specific set-up and model. Within this interpretation, the resonant amplification occurs as a result of the constructive interference of particle-hole coupling processes that arises at each horizon. (This particle-hole coupling gives rise to Hawking radiation in the quantum regime [15].) For the system in Fig. 1, the negative-energy modes are the holes, while the particles correspond to positive-energy modes. The interpretation as a black-hole laser points to some essential ingredients for the spin-wave amplification and lasing. First of all, the negative-frequency and positive- frequency modes need to be coupled. This coupling oc- curs only for elliptical spin waves because these are a superposition of positive and negative frequencies, see Eq. (5). Secondly, though we have assumed a step-like current-density, our results are more general as any tran- sition where vs goes from below (or above) to above (or below) vs,c will couple negative-energy and positive- energy modes and thus lead to amplification and las- ing. A unique ingredient of magnetic systems is the way the horizons are implemented, i.e., using electric current rather than flow of the spin waves themselves. This gives rise to the non-adiabatic spin-transfer torque, determined by the parameter β, which has no counterpart in other analogue gravity systems. The existence of these non- adiabatic spin transfer torque is crucial to make the sys- tem dynamically stable. Typical experimentally accessible values are J ∼ 10−39 Jm2 and B/µB ∼ K/µB ∼ 0.1 − 1 T [11], so that ω0 ∼ 10 − 100 GHz, and the length scale(cid:112)J/ω0 ∼ 10 − 100 nm. This means that resonances should be visible for systems in the range L ∼ 10 − 1000 nm, which is eas- ily accessible experimentally. While we have in most of our treatment ignored magnetization relaxation (except for requiring that β ≈ α, which is a typical situation), our results remain valid provided the spin-wave coher- ence length is much long than L. This translates to the condition that 1/(αki) (cid:29) L. This condition is easily satisfied given that α (cid:28) 1. For the above-mentioned typical values of anistropies and external fields, the cur- rent density corresponding to vs,c is on the order of 1012 A/m2 [11]. Though large, this current density is reached often, e.g., in experiments using pulsed current-driven domain-wall motion [8]. This critical current density may be lowered in systems that involve Dzyaloshinskii-Moriya interactions. In such systems, the spin-wave dispersion has a linear part in the dispersion without current [16]. As a result, the current required to make some of the spin-wave energies negative is smaller . In conclusion, we have presented a simple set-up and model for spin-wave amplification and lasing by inhomo- geneous spin-transfer torques. We note that the amplifi- cation and lasing that we have discussed occurs already within the linear spin-wave approximation. As we have mentioned, the energy of the emitted spin waves is pro- vided by creating negative-energy excitations in the nar- row region of the wire. Future research could focus on more accurate modelling of the set-up in Fig. 1, includ- ing non-linearities, dipolar interactions and more compli- cated current patterns. These issues could be addressed by numerical solution of the stochastic Landau-Lifshitz- Gilbert equation. Recently, numerical results were re- ported that show spin-wave emission in a similar set-up as the one we consider [17]. A direct comparison between these results and ours is postponed to future work. Acknowledgements. -- We thank Andrei Slavin for 5 sharing unpublished results and Reinoud Lavrijsen for useful discussions and comments on our manuscript. R.D. is member of the D-ITP consortium, a program of the Netherlands Organisation for Scientific Research (NWO) that is funded by the Dutch Ministry of Educa- tion, Culture and Science (OCW). This work is in part funded by the Stichting voor Fundamenteel Onderzoek der Materie (FOM) and the European Research Council (ERC). [1] V.V. Kruglyak, S.O. Demokritov, and D. Grundler, J. Phys. D: Appl. Phys. 43, 264001 (2010). [2] A. V. Chumak and H. Schultheiss, J. Phys. D: Appl. Phys. 50, 300201 (2017). [3] V. E. Demidov, S. Urazhdin, A. B. Rinkevich, G. Reiss, and S. O. Demokritov, Appl. Phys. Lett. 104, 152402 (2014). [4] Jairo Sinova, Sergio O. Valenzuela, J. Wunderlich, C.H. Back, and T. Jungwirth, Rev. Mod. Phys. 87, 1213 (2015). [5] O. Gladii, M. Collet, K. Garcia-Hernandez, C. Cheng, S. Xavier, P. Bortolotti, V. Cros, Y. Henry, J.-V. Kim, A. Anane, and M. Bailleul, Appl. Phys. Lett. 108, 202407 (2016). [6] E. Padr´on-Hern´andez, A. Azevedo, and S. M. Rezende, Phys. Rev. Lett. 107, 197203 (2011). [7] Uchida K, Xiao J, Adachi H, Ohe J, Takahashi S, Ieda J, Ota T, Kajiwara Y, Umezawa H, Kawai H, Bauer G E W, Maekawa S and Saitoh E 2010 Nature Mat. 9 894 -- 897 [8] D. C. Ralph and M. D. Stiles, J. Magn. Magn. Mater. 320, 1190-1216 (2008). [9] Vincent Vlaminck and Matthieu Bailleul, Science 322, 410 (2008). [10] G. Jannes, P. Massa, T. G. Philbin, and G. Rousseaux Phys. Rev. D 83, 104028 (2011). [11] A. Rold´an-Molina, Alvaro S. Nunez, and R.A. Duine, Phys. Rev. Lett. 118, 061301 (2017). [12] Jennifer Chaline, Gil Jannes, Philppe Maıssa, andGer- in D. Faccio, et al., Eds., Analogue main Rousseaux, Gravity Phenomenology (Springer, Heidelberg, 2013). [13] S. Corley and T.D. Jacobson, Phys. Rev. D 59, 124011 (1999). [14] J. Steinhauer, Nature Phys. 10, 864 (2014). [15] S.W. Hawking, Nature 248, 30 (1974); S.W. Hawking, Commun. Math. Phys. 43, 199 (1975). [16] Jung-Hwan Moon, Soo-Man Seo, Kyung-Jin Lee, Kyoung-Whan Kim, Jisu Ryu, Hyun-Woo Lee, R. D. McMichael, and M. D. Stiles, Phys. Rev. B 88, 184404 (2013). [17] M. Dvornik, R. Khymyn, J. kerman, V.S. Tiberkevich and A.N. Slavin, Intrinsic generation of propagating spin waves by a magnonic "black hole in a current-driven nano-constricted Wire, HD-10, Abstracts of the Interna- tional Magnetics Conference INTERMAG-18, Singapore, Singapore, April 2018. . 

1602.08091

2

1602

All electrical propagating spin wave spectroscopy with broadband wavevector capability

We develop an all electrical experiment to perform the broadband phase-resolved spectroscopy of propagating spin waves in micrometer sized thin magnetic stripes. The magnetostatic surface spin waves are excited and detected by scaled down to 125 nm wide inductive antennas, which award ultra broadband wavevector capability. The wavevector selection can be done by applying an excitation frequency above the ferromagnetic resonance. Wavevector demultiplexing is done at the spin wave detector thanks to the rotation of the spin wave phase upon propagation. A simple model accounts for the main features of the apparatus transfer functions. Our approach opens an avenue for the all electrical study of wavevector-dependent spin wave properties including dispersion spectra or non-reciprocal propagation.

cond-mat.mes-hall

cond-mat

a All electrical propagating spin wave spectroscopy with broadband wavevector capability Imec, Kapeldreef 75, B-3001 Leuven, Belgium and F. Ciubotaru∗ KU Leuven, Departement Electrotechniek (ESAT), Kasteelpark Arenberg 10, B-3001 Leuven, Belgium Institut d'Electronique Fondamentale, CNRS, Univ. Paris-Sud, Universit´e Paris-Saclay, 91405 Orsay, France T. Devolder M. Manfrini, C. Adelmann, and I. Radu Imec, Kapeldreef 75, B-3001 Leuven, Belgium (Dated: August 27, 2018) We develop an all electrical experiment to perform the broadband phase-resolved spectroscopy of propagating spin waves in micrometer sized thin magnetic stripes. The magnetostatic surface spin waves are excited and detected by scaled down to 125 nm wide inductive antennas, which award ultra broadband wavevector capability. The wavevector selection can be done by applying an excitation frequency above the ferromagnetic resonance. Wavevector demultiplexing is done at the spin wave detector thanks to the rotation of the spin wave phase upon propagation. A simple model accounts for the main features of the apparatus transfer functions. Our approach opens an avenue for the all electrical study of wavevector-dependent spin wave properties including dispersion spectra or non-reciprocal propagation. Spin wave based computing1 -- a paradigm-shifting tech- nology that uses the interference of spin waves -- offers po- tential for significant power and area reduction per com- puting throughput with respect to complementary metal- oxide-semiconductor (CMOS) transistor technology. Effi- cient solutions for spin wave routing2,3, spin wave emission4, amplification1, and spin wave combination5,6 have been de- veloped. However, these solutions often rely on materials that are difficult to integrate7 into a CMOS environment. More- over, they are often demonstrated only for long wavelength (≥ 1 µm) spin waves, for which the low group velocity lim- its the speed of computation and communication. Efficient methods to generate and detect spin waves with short wave- lengths are still lacking. Inductive methods have commonly been employed for long wavelength spin waves as has Bril- louin light scattering spectroscopy8, which is however diffrac- tion limited and requires complex procedure to retrieve phase information9. Alternative spin wave generation and detec- tion methods based on magneto-elastic coupling in surface- acoustic wave devices10 are still under development and raise questions regarding their high frequency capability11. Long- to-short wavelength conversion can be done in magnonic crystals12 for a geometrically limited discrete set of wavevec- tors at the expense of high conversion loss. A better conver- sion efficiency can be obtained by periodically folded copla- nar antennas13,14 but this at the expense of any flexibility in the generated wavevevector. Overall, none of the above meth- ods has so far demonstrated the combination of phase reso- lution, broad frequency coverage, high sensitivity, and large wavevector (short wavelength) capability. In this work, we demonstrate that the use of deep sub- micron inductive antennas can circumvent these limitations and allow for the generation and detection of spin waves with ultra-wide frequency band capability and broad wavevector capability up to 15 − 20 rad/µm. We illustrate our method on micrometer-sized Permalloy stripes and describe its be- havior within an analytic framework. Our method comple- ments the advantages of Brillouin Light scattering in a com- pact all electrical device in which the phase resolution is in- herent and the operation is convenient as it benefits from the speed -- and sensitivity -- of electronic measurement systems. Our geometry paves the way for all electrical characterization of wavevector-dependent properties of spin waves like chi- ral effects16, magnon-magnon interactions and spin current to spin wave interactions13 that need to be understood to assess the technological potential of spin wave based computing. Our on based are (Permalloy), sputter-deposited devices Ta 3 nm/Ni80Fe20 t = 17 nm/Ta 3 nm films grown on top of a 300 nm layer of SiO2 on Si substrate. Their hysteresis loops indicated a very soft and quasi- isotropic behavior, consistent with a polycrystalline and not a textured structure. A saturation magnetization of 771 kA/m and a damping of 0.008 was extracted from ferromagnetic resonance (FMR) experiments. The Permalloy layer was patterned into stripe-shaped spin wave conduits with widths in the range of w ∈ [0.25, 5 µm] using electron beam lithography and standard ion-beam patterning. The conduits were then covered by 100 nm of hydrogen silsesquioxane (HSQ) for electric isolation. Spin wave excitation and detection was performed via two identical Ti/Au antennas with widths in the range of L ∈ [0.125, 1 µm] and center-to-center distances ranging from 1.2 to 10 µm, denoted below as r. Each antenna is connected in series between a ground pad and another coplanar contact pad (Fig. 1). The antennas are RF powered and thus generate an oscillating RF field hx located essentially under the an- tennas and a weaker and anti-symmetric RF field hz located at the edges of the antennas [see example in Fig. 1(b)]. In this geometry, the RF fields will thus excite spin waves with wavevectors purely along the x direction. Static longitudinal fields saturating the magnetization along the stripe length lead to signals related to spin waves that are excited by hz at the antenna edges. However the associ- ated signals (not shown) were one to two orders of magnitude Network analyzer (a) z w y x hrf j S11 Port 1 Port 2 S22 L Spin waves H r h x h z (b) 1.00 0.75 0.50 0.25 0.00 -0.25 -0.50 f o n o i t u b i r t s d l a i t a p S i ) s t i n u . b r a ( f r h d e z i l a m r o n ) t i n u . b r a ( e d u t i l p m A 100 10-1 10-2 10-3 10-4 10-5 10-6 -4 -2 0 2 4 Distance from antenna x ( m)μ Permalloy conduit (c) 0 25 50 75 100 Wavenumber (rad/ m)μ (d) 2.5 mμ 125 nm Spin wave antenna FIG. 1. (Color online) Experimental configuration: (a) Schematic of the studied devices consisting of inductive antennas (orange) and a magnetic stripe conduit (blue) with propagating spin waves (wavy arrows).(b) Spatial distribution of the hx and hz components of the magnetic fields induced by the microwave current flowing through a 125 nm wide antenna placed at 100 nm above the permalloy stripe. (c) Wavenumber distribution of the hx field in (b). (d) Scanning electron micrograph of a 2.5 µm wide Permalloy conduit with two 125 nm antennas. smaller than when a transverse magnetic field was applied in the configuration depicted in Fig. 1(a). In the remainder of the paper, we focus on the case of a static field Hy saturating the stripe magnetization in the transverse direction (Damon- Eshbach configuration), and thus study the properties of mag- netostatic surface spin waves (MSSW) that are excited in this geometry. The devices were characterized by measuring the magnetic field dependence of their complex scattering matrix Si,j with i, j ∈ {1, 2} under 10 mW of excitation power. An on- chip full two-port calibration routine with a load-match-reflect standard impedance calibration kit was performed to correct for imperfections of the network analyzer, the cable assem- bly, and the RF probes. The pads connecting the device were short enough (300 µm) that no deembedding was required. The phase error associated with this approximation is negli- gible compared to the phase accumulation due to propagation of spin waves through our devices, as shown below. The signal transmission through the device comprises two su- perimposed signals: (i) a direct antenna-to-antenna parasitic coupling of typically -30 to -40 dB and (ii) smaller superim- posed spin wave related signals. We construct an approxi- mation of the scattering matrix of the parasitic transmission by calculating the average of the scattering matrix over all applied magnetic fields. We then isolate the spin-wave re- lated signals by defining the complex admittance normalized to 1/50 Ω as: 2 Y (Hy) = 1 2h S(Hy) − 1 2Hmax Z Hmax −Hmax S(Hy) dHyi (1) The typical order of magnitude of Y is 10−5 for r = 1.1 µm, L = 125 nm and w = 2.5 µm, and seems to depend on the spin wave conduit width. The basic behavior of the devices is illustrated in Figs. 2(a- c), which display the transmission and reflection properties for a stripe width of w = 5 µm, a propagation distance of r = 4 µm, and antenna widths of L = 1 µm. The field-independent cross-talk is subtracted using Eq. 1. For a given magnetic field, the spin-wave transmission signal exists only above a certain frequency threshold and stays detectable within a frequency transmission band of width ∆f . In all cases, the onset of finite transmission comes together with a drop of the reflection signal [red line in Fig. 2(c)]. The frequency threshold shifts to higher frequencies upon increasing the magnetic field strength. We have found that it matches quantitatively the FMR condition ωFMR ≈ γ0pHy(Hy + MS) [see Fig. 2(b)]. Here, γ0 is the gyromag- netic ratio and the shape anisotropy of the stripe has been ne- glected. Conversely, the transmission band [fFMR, fFMR+∆f ] gets nar- rower as we increase the field strength. This behavior can be understood by looking at the expected dispersion relation of spin waves15 for several magnetic fields [see e.g. Fig. 2(d)]. The narrowing of transmission band stems from the fact that the antenna can excite spin waves only up to a finite wavenum- ber kmax. The flattening of the dispersion relations at larger magnetic fields [Fig. 2(d)] then leads to a narrowing of the transmission band ∆f = f (kmax) − f (k = 0) when the mag- netic field is increased. The shapes of the reflection and transmission signals [Fig. 2(c)] differ substantially: the transmission signals are complex numbers with real and imaginary parts that oscillate and decay with frequency, while the reflection signals do not oscillate and show a slightly less pronounced decay with fre- quency. The oscillation of the transmission signal is thus a phase rotation inherent to the propagation of the spin waves. The decay of the envelope of the reflection and transmission signals stems from the wavevector-dependent excitation effi- ciency. The faster decay of the propagation signals may be indicative of additional losses upon spin wave propagation. A final striking point is the amplitude non reciprocity. For positive fields, the forward transmission is typically five times larger than the backward transmission, while the situation is reversed upon a change of the applied field direction (Fig. 2). This feature recalls the surface character of the MSSW of fi- nite wavevector, and it will be discussed further below. The main experimental features -- frequency threshold to al- low transmission, phase rotation upon transmission, and non reciprocity -- can be accounted for by the following simple (a) -200 -150 -100 -50 Field (mT) 50 0 100 150 200 ) z H G ( y c n e u q e r F 12 8 4 12 8 4 Low intensities Re(Y )21 Δf3 High intensities Δf2 Δf1 Re(Y )12 High intensities Low intensities FMR frequency 3 (d) (c) Macrospin Re(Y )21 Im(Y )21 5xRe(Y )12 Im(Y )/411 ) z H G ( y c n e u q e r F 14 12 10 8 6 4 2 1 3 3 m T 0 m T T 0 m 5 1 Δf3 Δf2 Δf1 kmax (b) -200 -150 -100 -50 0 50 Field (mT) 100 150 200 10 11 12 13 0 Frequency (GHz) 5 15 Wavenumber (rad/µm) 10 20 FIG. 2. (Color online) Field dependence of the real parts of the forward transmission (a) and backward transmission (b) signals (Eq. 1) for a Permalloy stripe with w = 5 µm, a propagation distance of r = 4 µm, and antenna widths of L = 1 µm. Light blue color corresponds to a zero spin-wave transmission, while the dark blue and the white stands for propagating spin waves. The dashed black curve in (b) is the analytical field-dependency of the ferromagnetic resonance frequency. (c) Comparison of the forward transmission (black symbols), backward transmission (green symbols) and forward reflection (red symbols) for a field of 140 mT along the stripe width. The latter have been rescaled for readability. The lines superimposed on the experimental response curve are the responses modeled according to Eq. 6 with β = 0.15. The top curves are the theoretical macrospin response using material parameters determined by FMR. (d) Spin wave dispersion relations calculated analytically for the structure in (a) with (dashed curves) and without (continuous lines) taking into account the exchange interactions for three values of the applied magnetic field as indicated. kmax = 6.28 rad/µm is the maximum spin-wave wavevector that can be excited by a 1 µm wide antenna. model. The pumping field has its components and its gradi- ents along the stripe axis x and stripe normal z [Fig. 1(a)]. The z dependence is due to RF absorption by Eddy currents and excitation of the (lossy) magnetic precession17, and it has its importance since MSSW waves tend to localize at either the top or the bottom surface of a ferromagnetic film18,19, depend- ing whether kx, M and z form a direct or an indirect trihedron. The demagnetizing field within the stripe is quasi-uniform, except for a y gradient of its y component near the stripe edges. We have performed micromagnetic simulations20 and have found that a transverse bias field of 10 mT is enough to saturate the sample over 75% of its width. All together, this provides in principle possibilities of exciting spin waves with propagating kx and kz components and with standing waves with ky near the edges. Since spin waves with wavevectors kz perpendicular to the film plane have frequencies substantially above our region of investigation, we shall restrict our study to the case of kz = 0. The main transmission channel relies on spin waves extending over the whole stripe width, so only spin waves with wavevec- tors purely along x (Damon-Eshbach modes18) will be in- cluded in the model. For long wavelength (i.e. kxt << 1) spin waves of frequency ω/(2π) the real part of the complex wavevector can be approximated by an exchange-free formu- lation: dispersion relations calculated with and without taking into account the exchange interaction in the studied wavevector range. We approximate the Oersted field profile below the an- tenna by assuming that its x-component hmax(z) is uniform in an interval [−Leff/2, Leff/2] and vanishes everywhere else. Leff ≈ L + h is the typical lateral extension of the RF field at the stripe altitude. The description is accurate only when L ≥ h [Fig. 1(b)] but we shall see that the details beyond the lateral extension of the RF field are not critical. Equivalently, the RF field can be described in reciprocal space by: hx ∝ sinch kxLeff 2 i (3) This field profile changes sign at a wavevector of 2π/Leff: our excitation efficiency vanishes at this wavevector and its mul- tiples [Fig. 1(c)]. We shall account for the overlap between the RF field pro- file and the exponential profile of the surface spinwaves by defining an ad-hoc kx-dependent pumping (or pick-up) effi- ciency factor: χ = 1 t Z t/2 −t/2 hmax(z) exp[−kxz]dz (4) kx = 2 t ω2 − ω2 0 M 2 γ 2 S FMR H(ω − ωFMR) (2) where H is the Heaviside distribution. From Fig. 2(d) one can observe that there is nearly no difference between the If the center-to-center distance between the antennas is r, the spinwave undergoes a phase rotation of kxr along its prop- agation path, together with an exponential decay of rate βkxr, where β is a mode-dependent function of the Gilbert damping. This can be described by a propagation operator Y 21× ω ω( - 2 2 ) FMR ) 1 2 ( Y m I Rescaled (75 mT) 75 mT 50 mT 15 10 5 0 ) m µ / d a r ( r e b m u n e v a W 6 8 10 12 14 Frequency (GHz) FIG. 3. (Color online) Examples of spin wave signals transmis- sion (bold lines) and wavenumber (dashed lines, from Eq. 2) for the narrowest antenna L = 125 nm, the shortest propagation distance r = 1.125 µm and a stripe width of 5 µm. The top curve has been rescaled proportionally to the wavenumber to better evidence the sig- nal oscillations at higher frequencies. exp[−ikxr − βkxr] (5) Note that each maximum in the real part of the propagation operator corresponds to a total propagation over an integer number of wavelength. This will be a convenient way to di- rectly measure kx from the oscillatory experimental signals. Overall, the scattering admittance elements of our system at a given frequency are proportional to the product of the stripe width, the kx dependence of the excitation field (from Eq. 3) at the corresponding wavevector (from Eq. 2), the detection sensitivity (also Eq. 3), the square of the pumping efficiency (Eq. 4) and the complex propagation operator (Eq. 5). This yields: Yi6=j ∝ w χ2 sinc2h kxLeff 2 i exp[−ikxr − βkxr] (6) Note that because of the surface nature of MSSWs, χ2 will vary faster than exp[−2kxt] and thus it will differ for forward (S21) and backward (S12) transmission coefficients. The dif- ference increases with the wavevector, hence with frequency at a given applied magnetic field. This faster decay becomes clear in Fig. 2(c) when comparing forward (black) and back- ward (green) transmission signals. Note also that the transmis- sion coefficients are complex numbers, with real and imagi- nary parts that rotate in quadrature thanks to the propagation operator. This is also in line with experiments. The reflection coefficients S11 and S22 can be obtained from Eq. 6 by setting the propagation distance r to zero. Equation 6 is compared quantitatively to the experimental data in Fig. 1(c). The main experimental features -- oscilla- tions in quadrature of the two components of the transmission parameters and overall decay with the frequency -- are well reproduced with the propagation loss β as single fitting pa- rameter, which confirms that the essential physics is included in Eq. 6. 4 However, the onset of transmission near ωFMR is much more gradual in the experiments than in the modeling. This shortfall of Eq. 6 results from the assumption of a bijective relation be- tween wavevector and frequency (Eq. 2) while in reality any given mode has a finite susceptibility also below and above its eigenfrequency. For k = 0 (FMR macrospin mode) the susceptibilities are symmetric and antisymmetric Lorentzian lines that show a finite spread around the mode center fre- quency (see Fig. 2(c) blue curves). The rigorous description of the onset of transmission at the threshold goes beyond the scope of this paper, but would require to include a convolu- tion of the mode susceptibilities within the kernel of Eq. 6. One can notice that the experimental signals indeed resem- ble the convolution of the modeled response with a macrospin response [see Fig. 2(c)]. To be able to excite spin waves with an ultra-wide wavenumber band one must use antennas with smaller widths. For example, a 125 nm wide antenna can potentially excite spin waves with wavenumbers up to 50 rad/µm (Fig. 1(c)). Figure 3 demonstrates that spin waves can be efficiently gen- erated and sensed inductively by a such a narrow antenna. For the shortest propagation distance of r = 1.125 µm, few oscillations of the transmission signals were detected. The corresponding signals could be satisfactorily accounted for an effective Leff = 500 nm (fits not shown) which corre- sponds to a first lobe of Eq. 3 at 12 rad/µm. From the analysis of the dispersion relations calculated for different magnetic bias fields, we could estimate that the highest de- tected wavevector was kh,30 mT ≈ 17 rad/µm (wavelength λ ∼ 370 nm), kh,50 mT ≈ 15 rad/µm (λ ∼ 420 nm) and kh,75 mT ≈ 12 rad/µm (λ ∼ 525 nm), respectively. Our ap- proach is very advantageous because it identifies the routes to access even higher wavevectors. An improvement of the wavevector capability requires the minimization of the effec- tive antenna size Leff by thinning the insulator that separates the antenna from the stripe, as well as a further shrinking of the antenna. Nevertheless, this will increase the parasitic cross-talk and thus a geometrical compromise will have to be found. In summary, we have presented an inductive method to per- form phase-resolved spectroscopy of propagating spin waves in thin Permalloy stripes. We demonstrate that very narrow antennas can be successfully used to generated and detect spin waves in a very wide range of wavevectors (up to 17 rad/µm), which compares well with other techniques. Furthermore, we developed an analytic model to describe the main features of the spin-wave transmission functions as the dispersion rela- tion or the non-reciprocity in propagation. This work was supported by imec's Industrial Affiliation Program on Beyond CMOS devices and by the French Na- tional Research Agency (ANR) under contract No. ANR-11- BS10-0003 lead by Matthieu Bailleul. F.C. thanks J. Loo for e-beam lithography, Rudy Caluwaerts for SEM images and imec's clean room technical support. ∗ Florin.Ciubotaru@imec.be 1 A.V. Chumak, V.I. Vasyuchka, A.A. Serga, and B. Hillebrands, Nature Physics 11, 453 (2015). 2 K. Vogt, H. Schultheiss, S. Jain, J.E. Pearson, A. Hoffmann, S.D. Bader, and B. Hillebrands, Applied Physics Letters 101, 042410 (2012). 3 K. Vogt, F.Y. Fradin, J.E. Pearson, T. Sebastian, S.D. Bader, B. Hillebrands, A. Hoffmann, and H. Schultheiss, Nature Com- munications 5, 3727 (2014). 4 V.E. Demidov, S. Urazhdin, and S.O. Demokritov, Nature Materi- als 9, 984 (2010). 5 O. Rousseau, B. Rana, R. Anami, M. Yamada, K. Miura, S. Ogawa, and Y. Otani, Scientific Reports 5, 9873 (2015). 6 S. Klingler, P. Pirro, T. Brcher, B. Leven, B. Hillebrands, and A.V. Chumak, Applied Physics Letters 105, 152410 (2014). 7 V. Cherepanov, I. Kolokolov, and V. L'vov, Physics Reports 229, 81 (1993). 8 S.O. Demokritov, B. Hillebrands, and A.N. Slavin, Physics Re- ports 348, 441 (2001). 9 F. Fohr, A.A. Serga, T. Schneider, J. Hamrle, and B. Hillebrands, Review of Scientific Instruments 80, 043903 (2009). 10 S. Cherepov, P.K. Amiri, J.G. Alzate, K. Wong, M. Lewis, P. Upadhyaya, J. Nath, M. Bao, A. Bur, T. Wu, G.P. Carman, A. Khitun, and K.L. Wang, Applied Physics Letters 104, 082403 5 (2014). 11 C. Jegou, G. Agnus, T. Maroutian, V. Pillard, T. Devolder, P. Crozat, P. Lecoeur, and P. Aubert, Journal of Applied Physics 116, 204102 (2014). 12 H. Yu, G. Duerr, R. Huber, M. Bahr, T. Schwarze, F. Brandl, and D. Grundler, Nature Communications 4 (2013). 13 V. Vlaminck and M. Bailleul, Science 322, 410 (2008) 14 V. Vlaminck and M. Bailleul, Physical Review B 81, 014425 (2010). 15 B.A. Kalinikos and A.N. Slavin, Journal of Physics C: Solid State Physics 19, 7013 (1986). 16 M. Belmeguenai, J.-P. Adam, Y. Roussign, S. Eimer, T. Devolder, J.-V. Kim, S.M. Cherif, A. Stashkevich, and A. Thiaville, Physical Review B 91, 180405 (2015). 17 M. Bailleul, Applied Physics Letters 103, 192405 (2013). 18 R.W. Damon and J.R. Eshbach, Journal of Physics and Chemistry of Solids 19, 308 (1961). 19 M. Kostylev, Journal of Applied Physics 113, 053907 (2013). 20 The simulations were performed using the OOMMF open code: M.J. Donahue and D.G. Porter, NISTIR Report No. 6376, 1999. 21 C. Bayer, J. Jorzick, B. Hillebrands, S.O. Demokritov, R. Kouba, R. Bozinoski, A.N. Slavin, K.Y. Guslienko, D.V. Berkov, N.L. Gorn, and M.P. Kostylev, Physical Review B 72,064427 (2005). 

1305.0906

1

1305

A new model for the recombination and radiative lifetime of trions and biexcitons in spherically shaped semiconductor nanocrystals

In this letter, we propose a new model to determine the recombination oscillator strength of trions and biexcitons for bound and unbound cases in the effective mass approximation. The validity of our model has been confirmed by the radiative lifetime of the trion and biexciton in a spherical quantum dot. The results show that the model works sufficient accuracy in comparison with results of more complex methods such as quantum monte carlo techniques and atomistic calculations.

cond-mat.mes-hall

cond-mat

A new model for the recombination and radiative lifetime of trions and biexcitons in spherically shaped semiconductor nanocrystals Mehmet Sahin1, 2, a) and Fatih Ko¸c2 1)Department of Material Science and Nanotechnology Engineering, Abdullah Gul University, Kayseri, Turkey 2)Department of Physics, Faculty of Sciences, Selcuk University, 42075, Konya, Turkey In this letter, we propose a new model to determine the recombination oscillator strength of trions and biexcitons for bound and unbound cases in the effective mass approximation. The validity of our model has been confirmed by the radiative lifetime of the trion and biexciton in a spherical quantum dot. The results show that the model works sufficient accuracy in comparison with results of more complex methods such as quantum monte carlo techniques and atomistic calculations. 3 1 0 2 y a M 4 ] l l a h - s e m . t a m - d n o c [ 1 v 6 0 9 0 . 5 0 3 1 : v i X r a a)Electronic mail: mehmet.sahin@agu.edu.tr; mehsahin@gmail.com 1 Semiconductor quantum dots (QDs) are very promising structures to confine basic carriers like electrons, holes, or excitons. Recent studies show that the exciton mechanisms in QDs play a very important role in revealing extraordinary optical properties and some quantum mechanical effects which can not be observed in bulk semiconductor materials.1 -- 6 Quantum dots are currently the subject of intense research because of their controllable properties such as a size- and shape-tunable energy levels, lifetimes, high quantum yields, and chemical processability. These adjustable properties allow control of the dynamics of both single- and multi-exciton states by engineering the electronic and optical properties such as electron- hole wave function overlap, Coulomb interaction between the electron and hole, tunability between type-I and type-II localization regimes etc.3,7 -- 11 One of the best known optical properties is radiative decay of the excitons. In order to control the exciton radiative decay, one needs to tune the overlap between electrons and holes in QDs and this tuning is named as wave function engineering by Klimov et al.10 The overlap is very important for oscillator strength and lifetime of the excitonic structures which can be tuned depending on the overlap. A number of studies have been reported on oscillator strength and lifetimes of neutral of charged exciton complexes.12 -- 14 As is well known, respectively, the oscillator strengths of the exciton and biexciton struc- tures in a quantum heterostructure are15 fX ∝ < ψXPcv0 > 2, fXX ∝ < ψXX PcvψX > 2, (1) (2) where ψX > and ψXX > are the exciton and biexciton wave functions, respectively, and Pcv is the momentum operator. Although these expressions are used for the absorption phenomena, the same forms are also used in the recombination process. In this letter, we suggest an approximation to calculate the recombination oscillator strength of biexciton (XX), and negatively (X −) and positively (X +) charged excitons for bound and unbound cases in the effective mass approximation (EMA). In order to verify the model, we have calculated the radiative lifetime of the X −, X + and XX for bound and unbound cases in a spherical QD. We have shown that the recombination oscillator strengths 2 and radiative lifetime of trions and biexcitons can be determined with sufficient accuracy by our simple model in comparison with results of more complex calculations. In the electronic structure calculations, we have solved the Poisson-Schrodinger equations self-consistently in the single band EMA to determine the energy levels and corresponding wave functions. Also, the quantum mechanical many-body interactions between the same kinds of particles have been taken into account in the local density approximation (LDA). We consider a spherically symmetric QD structure. In the EMA and BenDaniel-Duke boundary conditions, the single particle Schrodinger equations of a multi-exciton complex can be written as (cid:20)− 2 2 ~∇r(cid:18) 1 m∗ e(r) ~∇r(cid:19) + Ve(r) − qeΦh + qeΦe + V e−e xc [ρe(r)](cid:21) Re(r) = εeRe(r), and (cid:20)− 2 2 ~∇r(cid:18) 1 m∗ h(r) ~∇r(cid:19) + Vh(r) − qhΦe + qhΦh + V h−h xc [ρh(r)](cid:21) Rh(r) = εhRh(r), (3) (4) where  is the reduced Planck constant, m∗ e(r) and m∗ h(r) are the position dependent effective mass of the electron and hole, respectively, Ve(r) is the electron confinement potential and Vh(r) is the hole confinement potential, qe (qh) is charge of the electron (hole), and Φe and Φh are the electrostatic Coulomb potentials of the electron and hole, respectively. The Vxc[ρ(r)] potentials are the exchange-correlation potentials between the same kinds of particles, and Re(r) and Rh(r) are the radial part of the electron and hole wave functions, respectively. εe is the energy eigenvalue of the electron and similarly, εh is the hole energy. These two equations become coupled via attractive Coulomb terms (qeΦh and qhΦe) and must be solved simultaneously with each other. At the same time, the self-consistency requirement in these calculations should be provided by the repulsive Coulomb potential terms (qeΦe and qhΦh) in Eqs. (3) and (4). In this way, all Coulomb effects on the energy eigenvalues and wave functions are taken into account. The electrostatic Coulomb potentials are calculated from the Poisson equations ~∇κ(r)~∇Φe = qe ε0 ρe(r) ~∇κ(r)~∇Φh = − qh ε0 ρh(r), 3 (5) where ρe and ρh are the density16 of the electron and hole, respectively, ε0 is dielectric permit- tivity of the vacuum and κ(r) is the position dependent dielectric constant of the structure. These equations contain the image potential contributions due to surface polarization at the interfaces. For the exchange-correlation potential in the trion and biexciton problem, the Perdew- Zunger17 expression, which is a parametrization of the Monte Carlo results of Ceperley and Alder18, is employed. Also, this formulation contains the self-interaction correction. The last three equations, Eqs.(3), (4), and (5), are solved self-consistently by the full numeric matrix diagonalization technique. It should be noted that the repulsive Coulomb and exchange-correlation potential terms in both Eqs. (3) and (4) must be reduced for X, since there are only one electron and one hole in the X. Similarly, these potential terms must be omitted in only Eq. (4) for X − because of single hole and in only Eq. (3) for X + because of single electron. The binding energy expressions of X −, X +, and XX are given, respectively, as19 Eb(X −) = Etot X + εe(0) − Etot X −, Eb(X +) = Etot X + εh(0) − Etot X +, EXX b = Etot XX − 2Etot X , (6) where Etot X is single exciton total energy, Etot positively charged excitons, respectively, Etot X − and Etot X + are the total energy of negatively and XX biexciton total energy and εe,h(0) is isolated single electron (hole) energy in the QD. As is well known, when a photon interacts with a QD, an electron-hole pair is formed in the QD. In this case, the single exciton oscillator strength, as is well known, is given by20 fX = Ep 2EX (cid:12)(cid:12)(cid:12)(cid:12) 2 Z r2drRe(r)Rh(r)(cid:12)(cid:12)(cid:12)(cid:12) , (7) where Ep is the Kane energy, EX is the exciton transition energy, Re(r) and Rh(r) are the radial part of electron and hole wave functions of the exciton, respectively. This equation can be easily used in both absorption and recombination processes of a single exciton. If a second photon is absorbed by the QD, this process results in a second exciton as illustrated in Fig. 1. This second exciton can occur in two possible situations, unbound 4 FIG. 1. Schematic representation of unbound and bound biexciton structure formation. or bound biexciton cases. In unbound biexciton case, the repulsive Coulomb interaction is predominant and this structure is considered as two-exciton isolated from each other. However, if the attractive Coulomb interaction is dominant, this complex is called as the bound biexciton and considered as a single particle. The oscillator strength of bound or unbound biexcitons is calculated by means of Eq. (2). These forms of oscillator strengths are also used in recombination processes of any biexciton structure.21,22 This formulation is right for an absorption phenomena. On the other hand, we suggest that the oscillator strength of recombination is assumed as different in a trion or a biexciton structure. Because, in contrast to the absorption process, there are various situations in a recombination phenomena. As seen from the left panel of Fig. 2, the unbound X − structure is established a single exciton and a one electron isolated from each other. In the unbound X − case, while one of the electrons (for example first electron) has the highest recombination probability with the hole, the recombination probability of second electron is approximately zero. This is similar to a single exciton recombination. However, in the bound X − case, all charges are bound with each other and considered as a single particle as seen the left panel of Fig. 2. Hence, both of the electrons have same recombination probabilities to the hole. Similar situations are valid for unbound and bound X +, respectively, as seen in the middle panel of the figure. That is, while there is one recombination probability in the unbound trions as is in a single exciton, the recombination probability is two times higher in bound trions. Similar discussions can be made for the biexciton structure. In an unbound XX case, while the recombination probability of first electron with first hole is the highest, the recombination probability of second electron with second hole is the highest. In this case, the recombination probability of an unbound XX is two times larger than that of a single exciton. In a bound XX case, the recombination probabilities of first electron with both holes are equivalent. Similarly, second electron has the same probability as that of the first one. As a result, the recombination probability of the bound XX is two times greater than that of the unbound 5 FIG. 2. Schematic representation of probable recombination processes in the unbound and bound X −, X + and XX structures, respectively. one. In the light of all these discussions, we propose a different model to calculate the recom- bination oscillator strength of trions and biexcitons. In this model, the oscillator strength is to be computed using single particle wave functions of trions or biexcitons. The wave functions include all probable Coulomb interactions and the surface polarization effect be- cause of the self-consistent electronic structure calculation procedure. The recombination oscillator strengths of trions and biexcitons are proposed as f(X −,X +,XX) = A , (8) Ep 2E(X −,X +,XX) (cid:12)(cid:12)(cid:12)(cid:12) 2 Z r2drRe(r)Rh(r)(cid:12)(cid:12)(cid:12)(cid:12) where EX −, EX +, EXX are the transition energies of negative or positive trion, or biexciton, respectively. Re(r) and Rh(r) are the radial part of the electron and hole wave functions of the considered structure, X −, X + or XX. Here, the A is a recombination probability factor and A ≃ 2 for bound and A ≃ 1 for unbound trions. Similarly, the factor A ≃ 4 for bound and A ≃ 2 for unbound biexcitons. The radiative lifetime of exciton is an important quantity for some device applications and therefore a number of studies on lifetime have been reported both theoretically and experimentally.1,25 -- 29 In order to check the validity of our model, we calculate the radiative lifetime of bound and unbound trions and biexciton in a core/shell spherical QD. As is well known, the radiative lifetime is inversely proportional with the oscillator strength and it is defined as23,24 where ε0 is the dielectric permittivity of the vacuum, m0 is the free electron mass, c is the τ = 6πε0m0c32 e2nβsE2f , (9) 6 FIG. 3. The overlap integrals (a), oscillator strength (b) and lifetime (c) of the X, X −, X + and XX as a function of the core radius in CdSe/CdS QD. light velocity, e is the electronic charge, f is the oscillator strength, n is the refractive index, E is the transition energy and βs is the screening factor.24 As a model structure, we use CdSe/CdS QD for type-I confinement regime. In this structure, both electron and hole are localized in the CdSe core region. The atomic units have been used throughout the calculations,  = m0 = e = 1. The material parameters are taken from Refs. 5. The effective exciton Bohr radius is 48.75 A and the effective exciton Rydberg energy is 15.9 meV. In the CdSe/CdS QD, the calculations are performed as a function of the core radius for fixed shell thickness to 2 nm. In this structure, both of the electron and hole are confined 7 to the CdSe core region. As can be seen from Eqs. (7) and (8), the oscillator strength expressions are directly proportional to the electron-hole overlap integral. Figure 3 (a) shows the overlap integral of the electron-hole wave functions for X, X −, X +, and XX. As is well known, in perturbation calculations, the overlap integral of the wave functions are same for the considered complexes, since there is no perturbative correction on the wave functions. Because all Coulombic effects in the electronic structure calculations are taken into account, the overlap integrals are different from each other. We conclude that the effects of the attractive and repulsive Coulomb potentials on the wave functions have drastically changed the overlap characteristic and hence these effects on the wave functions must be taken into account for more realistic calculations. The overlaps are small for all exciton complexes in small core radii and increase with increasing core radius except the X −. The reason for this behavior in the X − can be explained with repulsive Coulomb energy between the electrons. Because the electrons are more energetic than the hole(s), their wave functions expand much more to the CdS shell region5 and this process reduces the overlap of the wave functions. In the case of X, X + and XX, the kinetic energy term decrease with increasing core radii and the attractive Coulomb energy is a bit more dominant according to the X − and hence their overlap integrals increase. The oscillator strength of X, X −, X + and XX are shown in Fig. 3 (b). The oscillator strengths of the X −, and especially X +, have approximately the same values as that of X because the X − and X + are unbound in all core radii. When the core radius increases, the oscillator strength of X − decreases, while the oscillator strengths of X and X + increase slightly. The XX structure is unbound until the core radius is approximately equal to 4 nm. Therefore the recombination probability factor is A ≃ 2. In further increasing of the core radii, the XX becomes bound structure and therefore A ≃ 4. The recombination oscillator strength of each exciton is fX −/fX ≃ 0.92 and fX +/fX ≃ 1.02. This relationship for biexciton is fXX/fX ≃ 1.92 in unbound cases and fXX /fX ≃ 3.94 in bound cases. These ratios increases with increasing core radius in both bound and unbound XX cases. The lifetime of the X, X −, X + and XX, calculated by means of Eq.(9), is shown in Fig. 3 (c) as a function of the core radius. As seen from the figure, the single exciton lifetime is approximately equal for all core radii. Similar trends have been reported both theoretically and experimentally by some authors28,30. Some changes have been observed in the trions radiative lifetimes with the core radius, but these variations are not very evident especially in 8 X +. Although, in some theoretical studies25,28, it has been reported that the trion lifetime is almost half that of the single exciton lifetime, in our results, the trions' lifetimes are approximately equal to the exciton lifetime since the trions are not bound. The our model predicts τ −1 X − ≃ 0.95 τX two times (i.e. τ −1 XX = 2.14 τX and τ −1 . As regards to the biexciton, its lifetime is approximately X + ≃ 1.05 ) shorter than that of single exciton in case of unbound biexciton. τX This ratio is τ −1 XX = 4.17 τX for the bound biexciton structure. Similar results are found from quantum monte carlo and atomistic calculations12,28,31,32. In conclusion, we have suggested a new model to determine the recombination oscillator strength for the bound and unbound excitons in QDs and provided a different perspective for their radiative lifetimes. The model has been tested in determining the radiative lifetimes of the X −, X + and XX in a type-I spherical QD heterostructures and it is seen that the results are in a very good agreement with quantum monte carlo and atomistic calculations results. We have not compared our results with the results of first-order perturbation calculations in the EMA because, in that method, there is not taken into consideration the Coulombic effect corrections on the wave functions. In addition, the binding energy results are used a different manner except traditional one and as depending on the binding energy, we have decided the probability of recombination type. The presented model can be expanded to type-II QD strucutures or to more excitonic complexes than a biexciton, for example three or four excitons and so on. It should be noted that the success of the model is strongly dependent on taking into consideration of Coulomb effects on the wave functions in the electronic structure calculations. This study was supported by Turkish Scientific and Technical Research Council (TUBITAK) with Project Number 109T729. REFERENCES 1S. A. Ivanov and M. Achermann, ACS Nano 4, 5994 (2010). 2R. W. Meulenberg, J. R. I. Lee, A. Wolcott, J. Z. Zhang, L. J. Terminello, T. V. Buuren, ACS Nano 3, 325 (2009). 3S. Kim, B. Fisher, H.-J. Eisler and M. Bawendi, J. Am. Chem. Soc. 125, 11466 (2003). 4V. A. Fonoberov and A. A. Balandin, Appl. Phys. Lett., 85, 5971, (2004). 5S. Brovelli, R.D. Schaller, S.A. Crooker, F. Garcia-Santamaria, Y. Chen, R. Viswanatha, 9 J.A. Hollingsworth, H. Htoon and V.I. Klimov, Nature Communications 2, Article num- ber:280 (2011). 6M. Achermann, J. A. Hollingsworth, and V. I. Klimov, Phys. Rev. B 68, 245302 (2003). 7A. R. Kortan, R. Hull, R. L. Opila, M. G. Bawendi, M. L. Steigerwald, P. J. Carroll, L. E. Brus, J. Am. Chem. Soc. 112, 1327 (1990). 8L. P. Balet, S. A. Ivanov, A. Piryatinski, M. Achermann and V. I. Klimov, Nano Lett. 4, 1485 (2004). 9D. Oron, M. Kazes, and U. Banin, Phys. Rev. B 75, 035330 (2007). 10A. Piryatinski, S. A. Ivanov, S. Tretiak, and V. I. Klimov, Nano Lett. 7, 108 (2007). 11E. J. Tyrrell and J. M. Smith, Phys. Rev. B 84, 165328 (2011). 12G. Bacher, R. Weigand, J. Seufert, V. D. Kulakovskii, N. A. Gippius, A. Forchel, K. Leonardi, and D. Hommel, Phys. Rev. Lett. 83, 4417 (1999). 13M. Combescot, and J. Tribollet, Solid State Commun. 128, 273 (2003). 14M. Combescot, and O. Betbeder-Matibet, Phys. Rev. B 80, 205313 (2009). 15T. Takagahara, Phys. Rev. B 39, 10206 (1989). 16M. Sahin , S. Nizamoglu , O. Yerli, and H.V. Demir, J. Appl. Phys. 111, 023713 (2012). 17J. P. Perdew and A. Zunger, Phys. Rev. B 23, 5048 (1981). 18D. M. Ceperley and B. J. Alder, Phys. Rev. Lett. 45, 566 (1980). 19T. Tsuchiya, Physica E 7, 470 (2000). 20V. A. Fonoberov and A. A. Balandin, J. Appl. Phys. 94, 7178 (2003). 21V. M. Fomin, V. N. Gladilin, J. T. Devreese, E. P. Pokatilov, S. N. Balaban, and S. N. Klimin, Phys. Rev. B 57, 2415 (1998). 22T. Feldtmann, L. Schneebeli, M. Kira, and S. W. Koch, Phys. Rev. B 73, 155319 (2006). 23M. Califano, A. Franceschetti and A. Zunger, Phys. Rev. B 75, 115401 (2007). 24B. Alen, J. Bosch, D. Granados, J. Martinez-Pastor, J. M. Garcia and L. Gonzalez, Phys. Rev. B 75, 045319 (2007). 25F. Rajadell, J. I. Climente, J. Planelles, and A. Bertoni, J. Phys. Chem. C 113, 11268 (2009). 26M. Gong, W. Zhang, G. C. Guo, and L. He, Appl. Phys. Lett. 99, 231106 (2011). 27P. P. Jha and Philippe Guyot-Sionnest, ACS Nano 3, 1011 (2009). 28G. A. Narvaez, G. Bester, and A. Zunger, Phys. Rev. B 72, 245318 (2005). 29C. H. Wang, T. T. Chen, Y. F. Chen, M. L. Ho, C. W. Lai, and P. T. Chou, Nanotechnology 10 19, 115702 (2008). 30C. Bonati, M. B. Mohamed, D. Tonti, G. Zgrablic, S. Haacke, F. van Mourik, and M. Chergui, Phys. Rev. B 71, 205317 (2005). 31G A. Narvaez, G. Bester, A. Franceschetti, and A. Zunger, Phys. Rev. B 74, 205422 (2006). 32M. Wimmer, S. V. Nair, and J. Shumway, Phys. Rev. B 73, 165305 (2006). 11 

1503.03458

1

1503

Electron-beam evaporated cobalt films on molecular beam epitaxy prepared GaAs(001)

We have deposited Co films on the GaAs(001) surface by using an e-beam evaporation method. The thicknesses of the Co films are measured by using x-ray reflectivity and Rutherford backscattering. The magnetic properties of the films have been measured using superconducting quantum interference device. The magnetization of the films was found to decrease with increasing film thickness. The slight degradation of magnetic properties is attributed to increasing roughness on the Co surface or the Co/GaAs interface during the Co deposition.

cond-mat.mes-hall

cond-mat

Electron‐beam evaporated cobalt films on molecular beam epitaxy prepared GaAs(001)  Z. Ding and P.M. Thibado  Department of Physics, The University of Arkansas, Fayetteville, Arkansas 72701   C. Awo‐Affouda and V.P. LaBella  School of NanoSciences and NanoEngineering, State University of New York at Albany, Albany, New York 12203 We have deposited Co films on the GaAs(001) surface by using an e-beam evaporation method. The thicknesses of the Co films are measured by using x-ray reflectivity and Rutherford backscattering. The magnetic properties of the films have been measured using superconducting quantum interference device. The magnetization of the films was found to decrease with increasing film thickness. The slight degradation of magnetic properties is attributed to increasing roughness on the Co surface or the Co/GaAs interface during the Co deposition. I. INTRODUCTION  Research in the area of spintronic devices made using ferromagnetic/semiconductor heterostructures is of growing interest.1 For example, there is a current drive to make a spin filter device.2 It is also believed that spintronic devices will predominate in the electronics industry in the future. Injection of electrons from a magnetic material to a semiconductor is considered to be the most challenging task in spintronics. To achieve high efficiency of spin injection and high spin polarization, it is thought that an atomically abrupt interface is important.3 Therefore; it is essential to study the electronic and magnetic properties of ferromagnets in ferromagnetic/semiconductor structures. One approach to developing spintronic devices is to replace the normal metal contacts used on semiconductor device structures with ferromagnetic metal contacts. This can provide a source of spin-polarized current. Ideally, one wants to use a ferromagnetic metal that provides 100% spin-polarized current, shares the same crystal structure, and shares the same lattice constant as the semiconductor. Heusler alloys are a group of ferromagnetic metals that have 100% spin-polarized electrons at the Fermi level.4 Several share a compatible crystal structure and lattice constant with the III–V semiconductors, such as Co2MnAl. Another key factor in developing spintronics is the quality of the interface between the ferromagnetic metal and the semiconductor. The interface between the semiconductor and the ferromagnetic material plays a critical role in the efficiency of spin transmission.5 The structure of Co2MnAl consists of alternating planes of Co and MnAl. Thus, it may be important to first study Co grown on GaAs(001). In this article, we report on the quality of these films and compare them to bulk GaAs and bulk cobalt. In addition the evolution of the interface and film microstructure during growth and its correlation to magnetic behavior are discussed. II. EXPERIMENT  Experiments were performed in an ultrahigh vacuum (UHV) multi-chamber facility (5–810-11 Torr throughout) which contains a solid-source molecular beam epitaxy (MBE) chamber (Riber 32P) with a substrate temperature determination system accurate to +/- 2 °C,6 and an arsenic cell with an automated valve and controller. The MBE chamber also has an all UHV connection to a surface analysis chamber, which contains a custom integrated scanning tunneling microscope (STM) (Omicron).7 Commercially available ‘‘epi-ready’’, n+ (Si doped 1018/cm3) GaAs(001) +/- 0.05° substrates were loaded into the MBE system without any chemical cleaning. The surface oxide layer was removed and a 1.5-m-thick GaAs buffer layer was grown at 580 °C using an As4 to Ga beam equivalent pressure (BEP) ratio of 15 and a growth rate of 1.0 m/h as determined by reflection high-energy electron diffraction (RHEED) oscillations. After growth, the surface was annealed under an As4 BEP of 1 Torr for 15 min at 600 °C followed by another at 570 °C under the same conditions. This procedure improves the RHEED pattern and prepares the surface for STM measurement and e-beam evaporation of Co. For STM observation of the surface morphology, the samples were transferred to the STM without breaking UHV, and imaged at room temperature. After imaging the GaAs(001) surface with the STM, the substrate was transferred to another chamber equipped with a mini e-beam evaporator, without breaking UHV, for the deposition of Co. The experimental set up for the e-beam evaporation is illustrated in Fig. 1. During the evaporation of Co, the GaAs substrate was kept at room temperature and the deposition rate of Co was measured using a quartz crystal monitor. The electron beam is emitted from a hot filament and accelerated by a high voltage to hit the target Co material and heat it. Before the deposition of the Co on the sample, the quartz crystal monitor is placed in the position of the sample to measure the deposition rate of the Co. After the measurement of the deposition rate, the quartz crystal monitor is moved to an adjacent location where a reduced deposition rate can be monitored and scaled to yield the original rate. The sample is then positioned in the same place that the quartz crystal monitor was originally placed, so the deposition rate of the Co is known. By doing this the deposition rate of the Co is believed to be the same as measured by the quartz crystal monitor. By controlling the deposition time, different thicknesses of the Co film can be obtained. The thicknesses of the Co films were measured by both x-ray reflectivity (XRR) and Rutherford backscattering (RBS) experiments. These also provide a calibration for the calculated quartz crystal monitor measurement of the Co film thickness. The magnetic properties of the Co films on GaAs were measured by a superconducting quantum interference device (SQUID). The magnetization of the Co films is calculated from the SQUID data combined with the RBS data. III. RESULTS  The ex situ XRR and RBS experiments provided the thickness of the Co films as shown in Fig. 2 (RBS data not shown). These results are in good agreement with each other and are consistent with the result from the deposition time and rate obtained by the quartz crystal monitor. In the XRR data, the reflectivity of the Co film oscillates with the incident angle of the x-ray beam. The adjacent oscillation peak spacing is related to the thickness of the film, while the decreasing intensity of oscillations indicates the presence of roughness on the surface and/or the interface. Typically, surface roughness decreases the intensity of the whole curve, while interfacial roughness is primarily responsible for the decay in the amplitude of oscillations. From the STM images we know the starting GaAs surface is extremely flat8 (in Ref. 8 see STM picture labeled Tl) so the roughening is generated by either the cobalt not growing smoothly or is strain induced. The magnetic properties of the Co films were measured by SQUID. Hysteresis curves are shown in Fig. 3 for the 50 and 560 nm thick Co films acquired at room temperature. The shape of the hysteresis curve near zero field is nearly a rectangle for the Co film with 50 nm thickness. The magnetic moment of the thicker Co film is much higher (almost ten times higher) than that of the thin film. Interestingly, the magnetic moment under higher magnetic field decreases slowly with the applied field for the thin film, while for the thick films, the magnetic moment increases slowly with the applied field. The magnetization, which is derived from the SQUID and RBS data of the Co films, is shown in Fig. 4. The thicknesses of the three samples are 50, 350, and 560 nm, respectively. The magnetization decreases linearly with the increase of thickness of Co film as shown in the figure. The line that is fit by the least-squares method has been added to the figure. Finally, the magnetization of bulk Co is also shown in the figure near the top as a reference.9,10   IV. DISCUSSION  The thicknesses of the Co films were measured ex situ by RBS and XRR. The disadvantage of an ex situ measurement is that the Co films are exposed to air and this can introduce contamination. By using the quartz crystal monitor, one can determine the thickness of the Co films in situ by measuring the deposition time and rate. On the other hand, the XRR and RBS measurements are a direct thickness measurement and can provide a calibration of the thickness. In the XRR data, the oscillations have damping intensities for the peaks, this would indicate the presence of interface roughness between deposited Co film and GaAs substrate as seen in Fig. 2. This may be due to the deposition of Co being carried out at room temperature. The interface roughness may also develop with the increase in film thickness. This is because of the mismatch of lattice constant for Co and GaAs, which produces strain on the interface of Co and GaAs. The strain will cause some defects such as dislocations or stacking faults after the thickness of Co film is grown thicker than a critical value of about 5 nm.11 This critical value represents the film thickness when the morphology changes from bcc to hcp, while the critical thickness of 50 nm represents the onset of dislocation formation in the film. Our thinnest film is around the critical thickness, while the other two films are seven and ten times this value. As the film grows beyond the critical thickness the dislocation density will increase, and this may play a role in the degradation of the magnetic properties. By comparison of the hysteresis curves of different samples, one can uncover the magnetic properties of the Co films. Also, one can speculate on the surface roughness of the Co films and the interface roughness between the Co and the GaAs. The interface roughness will affect the crystalline quality of the Co films, which will affect the magnetic properties of Co film in return. From Fig. 4, one can see that the magnetization of the thin film is higher than that of the thick film; this indicates that the thinner film has better crystalline quality and lower surface and/or interfacial roughness. The decrease in magnetization may also indicate an anisotropy in the Co film exists. Practical applications of ferromagnetic materials may require a magnetic thin film to have a single magnetic domain state with a preferential magnetization axis. This feature is typical for a film with good surface ordering. The material needs to be a single crystal and the surface should be well ordered over a long range, which should be achievable with cobalt since it only has a lattice mismatch of 0.2% yielding a critical thickness of about 50 nm before dislocations will form.12 A film with a disordered surface, by contrast, may alter the magnetic phase on the Co film.13 In some experiments about magnetic anisotropy of Co films, the experiments and theoretical calculation show that Co films on GaAs(001) and GaAs(110) are in the bcc phase during the early stage of growth.11,14 –16 The bcc phase is believed to originate from imperfections such as overlayer–substrate interaction, substrate atoms segregation, and surface roughness.11 After the thickness of the Co film is thicker than a critical thickness of about 5 nm, there coexist a polycrystalline hexagonal close packing (hcp) and bcc structures.12 Some other studies show that after the growth of bcc Co on GaAs(001) in the early stages, there exist two perpendicularly oriented hcp Co domains in the final stages.17,18 The hcp layer is found to be directly on top of the bcc layer in some other studies.19,20 For our samples, the thicknesses are much higher than 5 nm, so the crystalline phase of Co films may be hcp on top of bcc. The growth of Co on the oxide-desorbed GaAs(001) surface is more three-dimensional growth mode similar to that observed for Fe on GaAs(001) because the surface after oxide removal is still rough.21 For our samples, the starting surface is extremely flat and the starting surface roughness should be very low. Also, the surface reconstruction of the GaAs(001) surface may play a role in the roughening of the deposited Co film. Due to the 234 reconstruction,22 the surface might not provide a template as suitable as the (110) surface for driving Co atoms in the metastable bcc phase.23 In our samples, the Co films are deposited onto the GaAs fresh surface at room temperature and not annealed after the e-beam evaporation, so that the Co/GaAs interface reaction and interdiffusion should be negligible. We speculate that the existence of polycrystalline hcp structure may originate from the low deposition temperature, because at low temperature the Co atoms at GaAs do not have enough energy to reorganize themselves. Thus the surface roughens after the deposition of the first layer of Co atoms and the roughness develops with the proceeding of the deposition process, resulting in the polycrystalline bcc and hcp structures and differently oriented magnetic domains. The existence of different Co domains results in the degradation of magnetization of the Co films compared to the bulk Co material. V. CONCLUSION  We have deposited Co films on the GaAs(001) surface by using an e-beam evaporation method. The magnetic properties of the Co films have been measured by SQUID combined with RBS. The magnetization of the Co films is found to decrease with the increase of film thickness. The degradation of the magnetization of the Co film with the increase of film thickness is attributed to the roughness on the Co surface and/or Co/GaAs interface during the Co deposition. ACKNOWLEDGMENTS  This work is supported by the National Science Foundation Grant No. FRG DMR-0102755 and the State of New York Grant No. NYSTAR-FDP-C020095. References 1G. A. Prinz, Phys. Today 48, 58 (1995). 2G. Kirczenow, Phys. Rev. B 63, 054422/1 (2001). 3V. P. LaBella, D. W. Bullock, Z. Ding, C. Emery, A. Venkatesan, W. F. Oliver, G. J. Salamo, P. M. Thibado, and M. Mortazavi, Science 292, 1518 (2001). 4M. S. Lund, J. W. Dong, J. Lu, X. Y. Dong, C. J. Palmstrom, and C. Leighton, Appl. Phys. Lett. 80, 4798 (2002). 5A. T. Hanbicki, B. T. Jonker, G. Itskos, G. Kioseoglou, and A. Petrou, Appl. Phys. Lett. 80, 1240 (2002). 6P. M. Thibado, G. J. Salamo, and Y. Baharav, J. Vac. Sci. Technol. B 17, 253 (1999). 7J. B. Smathers, D. W. Bullock, Z. Ding, G. J. Salamo, P. M. Thibado, B. Gerace, and W. Wirth, J. Vac. Sci. Technol. B 16, 3112 (1998). 8V. P. LaBella, D. W. Bullock, C. Emery, Z. Ding, and P. M. Thibado, Appl. Phys. Lett. 79, 3065 (2001). 9C. Kittel, Introduction to Solid State Physics (Wiley, New York, 1953). 10N. W. Ashcroft and N. D. Mermin, Solid State Physics (Saunders College Press, Philadelphia, 1976). 11A. Y. Liu and D. J. Singh, J. Appl. Phys. 73, 6189 (1993). 12C. J. Palmstrom, C. C. Chang, A. Yu, G. J. Galvin, and J. W. Mayer, J. Appl. Phys. 62, 3755 (1987). 13E. J. Escorcia-Aparicio, H. J. Choi, R. K. Kawakami, and Z. Q. Qiu, Phys. Rev. B 58, 93 (1998). 14G. A. Prinz, Phys. Rev. Lett. 54, 1051 (1985). 15F. Xu, J. J. Joyce, M.W. Ruckman, H.W. Chen, F. Boscherini, D. M. Hill, S. A. Chambers, and J. H. Weaver, Phys. Rev. B 35, 2375 (1987). 16Y. U. Idzerda, W. T. Elam, B. T. Jonker, and G. A. Prinz, Phys. Rev. Lett. 62, 2480 (1989). 17Y. Z. Wu et al., Phys. Rev. B 57, 11935 (1998). 18E. Gu et al., Phys. Rev. B 52, 14704 (1995). 19M. A. Mangan, G. Spanos, T. Ambrose, and G. A. Prinz, Appl. Phys. Lett. 75, 346 (1999). 20K. G. Nath, F. Maeda, S. Suzuki, and Y. Watanabe, J. Appl. Phys. 90, 1222 (2001). 21B. T. Jonker, G. A. Prinz, and Y. U. Idzerda, J. Vac. Sci. Technol. B 9, 2437 (1991). 22V. P. LaBella, H. Yang, D. W. Bullock, P. M. Thibado, P. Kratzer, and M. Scheffler, Phys. Rev. Lett. 83, 2989 (1999). 23S. J. Blundell, M. Gester, J. A. C. Bland, C. Daboo, E. Gu, M. J. Baird, and A. J. R. Ives, J. Appl. Phys. 73, 5948 (1993).   FIG. 1. Illustration of the e-beam evaporation system. The quartz crystal monitor is used to measure the deposition rate of Co, and can be positioned in the same location as the sample. FIG. 2. X-ray reflectivity intensity as a function of glancing incident angle for a 50 nm cobalt film grown on GaAs(001). FIG. 3. Hysteresis loops of Co films obtained by SQUID measurement. The data were acquired at room temperature and the thicknesses of the films are 50 and 560 nm. FIG. 4. Magnetization of Co films with different thicknesses deposited by e-beam evaporation. The horizontal solid line shown near the top is the magnetization of bulk Co material.           Figure 1.              Figure 2.            Figure 3.                Figure 4.          

1506.08617

4

1506

Adiabatic response and quantum thermoelectrics for ac driven quantum systems

We generalize the theory of thermoelectrics to include coherent electron systems under adiabatic ac driving, accounting for quantum pumping of charge and heat as well as the associated work exchange between electron system and driving potentials. We derive the relevant response coefficients in the adiabatic regime and show that they obey Onsager reciprocity relations. We analyze the consequences of our generalized thermoelectric framework for quantum motors, generators, heat engines, and heat pumps, characterizing them in terms of efficiencies and figures of merit.

cond-mat.mes-hall

cond-mat

a Adiabatic response and quantum thermoelectrics for ac driven quantum systems Mar´ıa Florencia Ludovico,1, 2 Francesca Battista,1, 2 Felix von Oppen,3 and Liliana Arrachea1, 2 1Departamento de F´ısica, FCEyN, Universidad de Buenos Aires and IFIBA, Pabell´on I, Ciudad Universitaria, 1428 CABA Argentina 2International Center for Advanced Studies, UNSAM, Campus Miguelete, 25 de Mayo y Francia, 1650 Buenos Aires, Argentina 3Dahlem Center for Complex Quantum Systems and Fachbereich Physik, Freie Universitat Berlin, 14195 Berlin, Germany We generalize the theory of thermoelectrics to include coherent electron systems under adiabatic ac driving, accounting for quantum pumping of charge and heat as well as for the work exchanged between electron system and driving potentials. We derive the relevant response coefficients in the adiabatic regime and show that they obey generalized Onsager reciprocity relations. We analyze the consequences of our generalized thermoelectric framework for quantum motors, generators, heat engines, and heat pumps, characterizing them in terms of efficiencies and figures of merit. We illustrate these concepts in a model for a quantum pump. PACS numbers: 73.23.-b, 72.10.Bg, 73.63.Kv, 44.10.+i I. INTRODUCTION Describing the relation between particle and energy currents is at the heart of thermoelectrics.1–5 For dc driv- ing with small temperature gradients and bias voltages, linear-response relations between the currents and the applied forces constitute the basis to describe thermo- electric phenomena. When combined with the principles of thermodynamics, the resulting theory has the beauty of simplicity and the strength of high predictive power. Specifically, it allows for a successful characterization of the efficiency of various thermoelectric machines in terms of the figure of merit introduced by Ioffe in 1949.6 An important challenge is to incorporate genuine quan- tum effects associated with coherent transport in nano- devices into this theoretical framework for thermoelec- tric effects. Here, we address how to include adiabatic quantum pumping as a paradigm of coherent-transport effects into a suitably generalized thermoelectric frame- work and explore the fundamental relations of the cor- responding quantum machines. Quantum pumping gen- erates nonzero dc currents by locally applying purely ac drivings to a quantum coherent conductor7–9. It gen- erates both charge and energy currents11, enables heat pumping, and the exchange of work between different driving forces12,13. The aim of the present work is to ex- tend the linear-response theory of thermoelectric effects to systems under adiabatic driving. To this end, we need to include the energy flux between the electrons and the ac forces on an equal footing with the heat and particle fluxes. Figure 1 shows the setup that we have in mind. It consists of a central coherent conductor which is cou- pled to two reservoirs. In conventional thermoelectrics,2 the two reservoirs differ in both temperature and chem- ical potential. A thermal engine converts a temperature difference into electric power. As a consequence of the Second Law, the efficiency of this conversion process is limited by the Carnot efficiency ηC = (T2− T1)/T2 where T2 > T1 denote the temperatures of the reservoirs. The optimal efficiency that can be reached for a specific device is controlled by its figure of merit or ZT value,2 1 + ZT − 1 1 + ZT + 1 . (1) √ √ η = ηC The ZT value can be expressed in terms of the linear- response coefficients of the device, relating charge and heat currents to bias voltage and temperature gradient.2 The thermal engine can also be operated in reverse, realizing a refrigerator which invests electric power to FIG. 1. Sketch of the setup. A coherent quantum conductor is driven by time-periodic potentials and connected to two reservoirs biased by (a) a chemical-potential difference δµ or (b) a temperature gradient δT , or both. Charge N R, heat QR W are exchanged between the reservoirs and the and power ac sources. The solid (dashed) arrow indicate (a) the motor (generator) mode of the device, and (b) the heat engine (heat pump) mode. continuously extract heat from the colder reservoir. The maximal efficiency of this device is given by the appropri- ate Carnot efficiency ηC = T1/(T2 − T1) and in terms of this Carnot efficiency, the optimal efficiency for a specific device is again given by Eq. 1.2 In this paper, we consider setups in which the coherent conductor is subject to a set of ac potentials in addition. For definiteness, we will consider reservoirs which have ei- ther different chemical potentials [Fig. 1(a)] or different temperatures [Fig. 1(b)], although our theory could read- ily be applied to situations which combine ac potentials with both chemical-potential and temperature gradients. The physics of these setups can be understood by analogy to the Archimedes device, a pipe with a rotating screw, which can be used to pump water against gravity. This is a classical analog of an adiabatic quantum pump, where ac driving pumps a certain amount of electric charge per cycle. Specifically, this charge can be pumped against an applied dc bias voltage,21 in which case quantum pump realizes a generator. The Archimedes screw can also be operated in reverse, with water flowing between the reservoirs by gravity and setting the screw into rotational motion. An analogous effect can be used to turn an adiabatic quantum pump into an adiabatic quantum motor. This is most easily understood when imagining that the time dependence of the ac potentials derives from the (classical) dynamics of, say, one or more mechanical degrees of freedom.14 Then, a charge current pushed through the coherent conductor will set the mechanical degrees of freedom into motion. Generator and quantum motor are driven by a bias voltage and correspond to the setup sketched in Fig. 1(a). Alternatively, we can also consider devices involving tem- perature gradients instead of bias voltages, which realize heat pumps and heat engines. Such a device is depicted in Fig. 1(b). The devices in Fig. 1 are examples of nanomo- tors and nanoengines, which have received much atten- tion recently.14–19 We note that the effect of ac potentials on the conventional thermoelectric effects has been stud- ied in a number of recent papers.20–23 The aim of the present work is to extend the linear- response theory of thermoelectrics to such nanomotors and nanoengines, to understand their efficiencies, and to identify appropriate figures of merit. This program poses several conceptual questions: (i) We need to identify the current that complements the charge and heat currents and accounts for the effects of the ac potentials. Simi- larly, we need to identify the affinity that complements the (scaled) temperature difference and bias voltage. (ii) We need to develop the generalized linear response theory which includes these additional quantities. While this is a conventional linear-response theory for traditional ther- moelectrics, the ac potentials are not actually weak but only slowly varying. (iii) We finally need to identify ap- propriate efficiencies and figures of merit. We will see that the latter also differ in essential ways from those defined in conventional thermoelectrics. In Sec. II, we generalize linear-response theory to in- 2 clude the response to the adiabatically varying ac poten- tials in addition to the applied bias voltage. We do this by working to linear order in the rate of change (or ve- locity) of the ac potentials. We find that this can be done in a manner which closely resembles the deriva- tion of Kubo formulas in linear response theory. Conse- quently, we derive general Kubo-like expressions for the response of both the charge current and the generalized forces conjugate to the ac potentials. These expressions imply that the response coefficients satisfy Onsager-like relations and are thus not independent of one another. In Sec. III, we generalize the thermodynamical frame- work to include the time-averaged work per unit time performed by the ac forces as a third flux, along with the heat and particle fluxes. We also identify the scaled frequency ω/T of the driving as the appropriate third affinity, complementing the temperature and chemical- potential differences. In Sec. IV, we define and analyze efficiency and figure of merit for the various quantum machines sketched in Fig. 1. We find that the defini- tion of the appropriate figure of merit analogous to the ZT value differs in characteristic ways, reflecting the fact that the usual off-diagonal thermoelectric response coef- ficients, the off-diagonal coefficients involving the third flux or affinity do not enter into the entropy production. To illustrate these concepts, we apply our theory to an ex- ample device in Sec. V which is based on a simple model for a quantum pump. We summarize in Sec. VI. II. ADIABATIC RESPONSE AND ONSAGER RELATIONS We begin by evaluating the forces and currents induced by a set of time-periodic parameters in the adiabatic ap- proximation. We will see that this can be done in close analogy to linear-response theory, allowing us to derive Onsager-like relations. We collect the parameters Vi(t) of the Hamiltonian H into a vector V(t) = V(t+T ) = (V1(t), V2(t), . . .) so that H = H(V(t)), where T = 2π/ω is the driving period.24 Quite generally the Hamiltonian of the system can be expressed as H(V(t)) = H0 −(cid:88) FjVj(t), (2) j where H0 is the time-independent part of the Hamilto- nian and Fj are hermitian operators that play the role of generalized forces F(t) = − ∂ H(t) ∂V(t) . (3) The quantum expectation value tr{ρF} in terms of the electronic density matrix ρ are just conventional forces when the Vj denote regular cartesian coordinates of a classical system obeying Newtonian dynamics. (cid:90) t At lowest order in the adiabatic approximation, the system is described by the frozen density matrix ρt for the Hamiltonian Ht with t treated as a parameter. Ac- counting for the temporal variation of V(t) to lowest or- der, we can approximate the time evolution operator as U (t, t0) (cid:39) T exp{−i Ht(t − t0) − i dt(cid:48)(t − t(cid:48))F · V(t)}. (4) To linear order in the small "velocity" V(t), we can now follow the usual steps of linear response theory25 and express the expectation value O(t) of an observable O at time t as t0 dt(cid:48)(t − t(cid:48))(cid:104)(cid:104) O(t), F(t(cid:48)) (cid:105)(cid:105)t V(t) O(t) (cid:39) (cid:104) O(cid:105)t − i (cid:90) t t0 = (cid:104) O(cid:105)t + ΛOF · V(t). t (5) Here, the operators O(t) and F(t(cid:48)) are defined in the Heisenberg representation with respect to the frozen Hamiltonian Ht and (cid:104). . .(cid:105)t denotes the expectation value with respect to the frozen density matrix ρt. The re- sponse function ΛOF can be expressed through the re- (t − t(cid:48)) = −iθ(t − tarded adiabatic susceptibility χO,F t(cid:48))(cid:104)[ O(t), F(t(cid:48))](cid:105)t. We now expand the frozen average to linear order in an applied bias δµ, yielding (cid:104) O(cid:105)t (cid:39) ΛOc t δµ, where the linear-response coefficient ΛOc is given by the usual Kubo formula. Applying this procedure specifically to the charge current J c(t) and the forces F(t) (and post- poning the heat currents and temperature gradients for further below), we obtain t t t (cid:18) J c(t) (cid:19) F(t) (cid:19) (cid:18) J c t Ft = + (cid:18) Λcc t Λcf t Λf f t Λf c t (cid:19)(cid:18) δµ (cid:19) V(t) , (6) to linear order in δµ and V(t). The terms in Eq. (6) have clear physical interpreta- tions. The first term on the right hand side collects the currents and forces evaluated with the frozen den- sity matrix ρt in equilibrium (i.e., for δµ = 0). These terms have zero mean when averaged over one period of the ac fields. The forces can be thought of as con- servative Born-Oppenheimer forces and expressed as a gradient of the equilibrium energy of the system with respect to V(t). For several potentials this term may lead to exchange of work between the different forces Fj without dissipation. Such processes were considered in Refs. 12 and 13. Adiabatic quantum pumping of charge by the ac potentials is described by Λcf cap- tures the modification of the forces by the applied bias δµ. Both contributions are generally nonzero when av- eraged over a period, implying that this contribution to the force is nonconservative. This was discussed for non- interacting electrons coupled to adiabatic nanomechani- cal systems26,27 and nanomagnets.28 In the latter case, this corresponds to a spin-transfer torque. The diago- nal components describe the usual conductivity through t and the velocity-dependent force through Λf f Λcc In t , while Λf c . t t 3 time-reversal symmetric systems, the latter is symmetric and describes a frictional force. Without time-reversal symmetry, Λf f t may have an antisymmetric part which is analogous to the Lorenz force.27 The derivation of the response coefficients Λij t follows standard linear-reponse theory, including the "adiabatic" response to the ac potentials. Consequently, it is natural to expect that the response functions Λij t satisfy Onsager- like relations. In fact, these can be derived in the usual manner, as shown in detail in App. A. Thus, we find the generalized Onsager relations Λcc t (B) = Λcc t (−B) , Λf f Λcf j (B) = sjΛf c ij (B) = sisjΛf f j (−B), ji (−B) (7) where the sign sj = ± depends on the parity of the op- erators Fj under time reversal. As the derivation of On- sager relations is very general, these relations are valid at finite temperature T and in the presence of many-body interactions.29 The second line in Eq. (7) imposes a relation between the adiabatic quantum pumping of charge (as described by Λcf j ) and the nonconservative force (as described by Λf c j ). This relation which is valid in the adiabatic regime was previously found for noninteracting adiabatic quan- tum motors at zero temperature and B = 0.14 It has been pointed out that time-reversal symmetry – by way of Onsager-like arguments – does not imply symmetry of the pumped charge under magnetic-field reversal unless the system has additional spatial symmetries.30–35 The relation in Eq. (7) implies that there is still an Onsager relation associated with the pumped charge, but it does not relate the pumped charge to itself but rather to the nonconservative force in response to an applied bias. Before closing this section, we comment on how to in- clude heat currents and thermal gradients into this linear- response scheme. Within linear response, we can readily extend Eq. (6) into a 3 × 3 matrix equation  J c(t) J Q(t) F(t)  =  J c t J Q t Ft +  Λcc t t Λcq Λqc t Λqq Λf c t Λf q t Λcf t Λqf t Λf f t t  δµ δT V(t)  . (8) Here, we can identify the thermal conductance Λqq re- t lating the heat current J Q(t) to δT as well as the usual thermoelectric coefficients Λcq and Λqc. In addi- tion, our scheme includes the coefficients Λqf and Λf q t t which describe the generation of heat currents by a time- dependent driving (quantum pumping of heat) and the generation of a nonconservative force in response to a temperature gradient, respectively. The treatment of a temperature gradient within the Kubo approach is less straightforward, but has been ad- dressed numerous times in the literature.37–39 An alter- native route is to calculate the relevant observables with a non-equilibrium technique, such as the Keldysh or scat- tering matrix formalisms, and to perform the expansions in δT, δµ, and V a posteriori. (This is the route which we follow in Sec. V). Either approach yields the additional Onsager relations Λqq t (B) = Λqq t (−B) , Λcq Λqf j (B) = sjΛf q t (B) = Λqc j (−B) t (−B) (9) complementing Eq. (7). The first line corresponds to the usual thermoelectric Onsager relations. The second line contains the additional Onsager relations relating pump- ing of heat current and the force generated in response to an applied thermal gradient. III. GENERALIZED THERMOELECTRIC FRAMEWORK Conventional thermoelectrics considers particle and heat currents in response to chemical-potential and tem- perature differences. In the presence of ac driving as in the devices in Fig. 1, we have to take into account the pumping of particles and heat as well as the work performed by or on the ac potentials on the same foot- ing. To develop the corresponding generalized thermo- electrics, we first consider the entropy production of the system. After averaging over one period of the ac driv- ing, the net dissipation occurs only in the electrodes and we can write S = QL TL + QR TR , where the average heat flux in lead α is given by Qα = Eα − µα N α. (10) (11) The energies Eα and particle numbers Nα satisfy the conservation laws N R = − N L, EL + ER = W . (12) While particle-number conservation takes the same form as in standard thermoelectrics, energy conservation must account for the additional work W performed by the ac potentials on the electron system. The corresponding j Fj(t) Vj(t), yielding power can be expressed as W = −(cid:80) T 2 −(cid:88) Vj(t) T (13) the entropy production S = N R + QR δµ T Fj(t) δT j to linear order in the applied bias δµ = µL−µR and tem- perature difference δT = TL − TR. Note that after aver- aging over a period, the conservative Born-Oppenheimer forces in Eq. (6) do not contribute to entropy production. Then, the power can be expressed in linear response and for δT = 0 as (cid:33) (cid:32) W =−(cid:88) ( Λf c t )j Vj(t)δµ+ ( Λf f t )jl Vj(t) Vl(t) .(14) j l (cid:88) 4 Here, the first term on the right-hand side describes the work performed by the nonconservative force originating from the applied voltage δµ (δT would contribute a sim- ilar term) and the second term is the dissipated power due to a frictional force on the ac potentials. In conventional thermoelectrics, one defines the par- ticle and heat fluxes J1 = N R and J2 = QR as well as the corresponding affinities X1 = δµ/T and X2 = δT /T 2.41,42 To extend thermoelectrics to the present sit- uation, we need to identify appropriate fluxes and affini- ties for the ac driving terms. At first sight, Eq. (13) may suggest to define the −Fj as fluxes and the Vj/T as the associated affinities. However, Eq. (13) holds only after averaging over one period. Be- fore time averaging, the conservation laws involve addi- tional terms36 and the forces Fj(t) contain contributions that are conservative. We can identify an appropriate affinity by noting that after averaging, the first term in Eq. (14) is proportional to ω, while the second term is proportional to ω2. It is thus natural to define the affin- ity X3 = ω/T with associated flux J3 = W /(ω).41,42 Thus, Eq. (13) yields S = JjXj (15) for the rate of entropy production. j We complete our quantum thermoelectrics scheme by linear-response relations between fluxes and affinities, (cid:88) (cid:88) Ji = LikXk. (16) k The linear-response coefficients Lij are readily related to the coefficients which appeared in Eq. (8). Indeed, we have L11 = T Λcc L21 = T Λqc L31 = −T Λf c t t , L12 = T 2Λcq t , L22 = T 2Λqq t , L13 = T Λcf t , L23 = T Λqf t t · v, L32 = −T 2Λf q · v, L33 = −T vT · Λf f t t · v, · v, · v, (17) where we defined v through V = ωv and vT denotes the transpose of v. Thus, the coefficients Lij also obey Onsager relations, namely Lii(B) = Lii(−B) , Lij(B) = ±Lji(−B), (18) with i (cid:54)= j. The sign in the second relation depends on the behavior of the fluxes under time reversal. Assuming time reversal from now on (and thus B = 0), this yields the relation L12 = L21, which is well known from the usual theory of thermoelectrics, as well as L13 = −L31 and L23 = −L32. It is important to note that the off- diagonal response coefficients have the same sign in con- ventional thermoelectrics, while they have opposite signs when either J3 or X3 is involved. We will see that be- low this has significant consequences for the definition of figures of merit for the devices in Fig. 1. The transport coefficients Lij can be directly calcu- lated from the coefficients Λ, which are in turn given in terms of the susceptibilities χt(ω). Another possibility is to start from the expressions for the charge, heat, and work currents, to perform the expansions in ω, δµ, and δT , and to identify the coefficients L from the resulting expressions. For noninteracting systems, this procedure is rather straightforward. In Sec. V, we will illustrate our general theory for a general noninteracting model of a two-terminal conduc- tor and evaluate the various response coefficients explic- itly. This will rely on Green function12,35,40 and scatter- ing matrix11,12 expressions for the response coefficients which we derive by the procedure described in the pre- vious paragraph. Details of the calculations are given in App. C. The calculations start with the expressions for charge current [Eq. (B2)], heat current [Eq. (B4)], and work current [Eq. (B6)] for this model. Performing the expansions in ω, δµ, and δT , we find the explicit formu- las for the Lij given in App. D. One can also check that these expressions for the response coefficients satisfy the generalized Onsager relations Eq. (18), as they should. IV. EFFICIENCY AND FIGURE OF MERIT OF QUANTUM MACHINES A. Motors and generators Consider a situation with applied ac driving forces and a dc bias δµ, but uniform temperature T . The device in Fig. 1(a) can operate as a quantum motor or generator. When the ac potentials pump particles into the reservoir with lower chemical potential, the gain in electrical en- ergy can be used to perform work on the source of the ac potentials. This occurs for L31δµ/T < 0 and corresponds to a motor as the work performed on the ac potentials can be further transformed, say, into mechanical work.14 When reversing the sign of δµ and thus L31δµ/T > 0, the ac potentials pump particles into the reservoir with higher chemical potential and we have a generator. Using X2 = 0, the rate of entropy production becomes S = L11X 2 1 + L33X 2 3 + (L13 + L31) X1X3. (19) Interestingly, the last term on the right-hand side van- ishes due to the Onsager symmetry L13 = −L31, and the coefficients L13 and L31 do not affect the entropy production. As a consequence, the second law of ther- modynamics imposes L11 > 0 and L33 > 0. This is in contrast to conventional thermoelectric, where the off- diagonal response coefficients are symmetric, L12 = L21, and do contribute to entropy production. In the latter case, the second law imposes detL = L11L22 − L2 12 > 0 in addition. 5 We are now ready to characterize the performance of adiabatically-driven quantum motors or generators in terms of efficiencies and figures of merit. The efficiency ηmot of a motor is measured by the ratio of the work per unit time − W performed on the ac potentials and the power N Rδµ/e injected by the voltage source. Similarly, the efficiency of the generator ηgen is given by the inverse of this ratio, so that ηmot = 1 ηgen = − W N Rδµ/e . (20) Note that we have defined µL = µR + δµ (as well as TL = TR = T ), cf. Fig. 1. We first show that the second law of thermodynamics implies an upper limit for these efficiencies. Using Eqs. (10), (11), and (12), we find W = T S − δµ e N R. Substituting this into Eq. (20), we obtain ηmot = 1 ηgen = 1 − T S N Rδµ/e . (21) (22) Now, the second law of thermodynamics demands S > 0. Moreover, the current flows with the potential drop δµ N Rδµ > 0, but against the potential drop in the motor, N Rδµ < 0. Consequently, we find that for a generator, both ηmot and ηgen are upper bounded by unity. The efficiency in Eq. (20) can also be written as ηmot = 1 ηgen = − X3J3 X1J1 (23) and the currents expressed through their linear-response expressions (16). Still assuming time-reversal symmetry, so that L13 = −L31, we can then maximize the efficiency as a function of X1 at fixed X3. One finds that the efficiency is maximized for L11L33 ± √ X1 = L11L33detL L11L13 X3. (24) with +(−) for motors (generators). Alternatively, we can fix X1 is fixed and maximize the efficiency as a function of X3. In this case, one finds −L11L33 ± √ L11L33detL X3 = L33L13 X1, (25) again with +(−) for motors (generators). Substituting Eqs. (24) and (25) into Eq. (23), we find for the maximal efficiency √ √ ηmax = 1 + ζ − 1 1 + ζ + 1 (26) and identify the figure of merit as −L13L31 L11L33 ζ = . ficiencies. Using Eqs. (10), (11), and (12), we find (27) W = TR S + TL − TR TL QL. 6 (29) Note that ηmax and ζ are valid for both motors and gen- erators. Equations (26) and (27) should be contrasted with con- ventional thermoelectrics,2–6 where the optimal efficiency satisfies an analogous expression. In conventional ther- moelectrics, the efficiency of converting heat into electri- cal energy is limited by the Carnot efficiency ηC. How- ever, the maximal efficiency which can be reached given a set of linear response coefficients Lij is lower than the Carnot efficiency by a factor involving the figure of merit ZT = L2 12/detL, see Eq. (1). In contrast, Eq. (26) de- scribes the efficiency of converting electrical energy into other (e.g., mechanical) forms of energy. This process is not fundamentally limited and hence Eq. (26) does not contain an analog of the Carnot efficiency. However, it still contains an analog of the factor involving the ZT value, which contains an appropriate figure of merit ζ. Thus, the motor efficiency ηmot is bounded by unity, and reaches this limit when ζ → ∞, i.e., when one of the dissipative coefficients L11 or L33 approaches zero. The different form of the figure of merit, i.e., the absence of the coefficients L13 and L31 from the denominator, re- flects the fact that unlike L12 and L21, these coefficients do not affect entropy production. Inserting this into the definition (28) of the efficiencies, we obtain ηhe = 1 ηhp = TL − TR TL + TR S QL . (30) For heat engines, heat flows from the hot to the cold QL < 0, while for heat pumps, heat reservoir so that QL > 0. Thus, we flows in the opposite direction, find that the second law S > 0 implies that the ef- ficiencies are smaller than the familiar Carnot efficien- cies, i.e., ηC = (TL − TR)/TL for the heat engine and ηC = TL/(TL − TR) for the heat pump. The efficiencies for heat engine and heat pumps can alternatively be expressed as ηhe = 1 ηhp = − X3J3 X2J2 , (31) where the fluxes can be expressed through their linear- response expressions (16). Maximizing the efficiency as for motors and generators, we again find Eqs. (24, 25), but with the affinity X2 taking the place of X1. This leads to a maximal efficiency of (cid:113) (cid:113) 1 + ζ − 1 1 + ζ + 1 (32) (33) B. Heat engine and heat pump ηmax = ηc Analogous results are obtained when the device is driven by a temperature gradient δT at constant chem- ical potential (X1 = 0), see Fig. 1(b). When the device operates as a heat engine, i.e., for L32δT /T 2 < 0, heat flows to the cold reservoir and the system performs work on the ac potentials. Conversely, the device operates as a heat pump when L32δT /T 2 > 0, where heat is pumped to the hot reservoir by the ac potentials. As a result of the Onsager symmetry, we have L23 = −L32 for time- reversal symmetric systems, and we again find that the second law imposes L22 > 0 and L33 > 0. An appropriate measure of the efficiency of a heat en- gine ηhe is the ratio of the work per unit time performed by the electrons on the ac forces, − W , and the heat leav- ing the hot reservoir − QL. (We assume that the left reservoir with temperature TR = TL−δT is the hot reser- voir. The efficiency ηhp of a heat pump is characterized by the inverse ratio. Thus, we have ηhe = 1 ηhp = W QL (28) with the figure of merit ζ = −L23L32 L22L33 . Equation (32) holds for both heat engines and heat pumps, when the appropriate Carnot efficiency ηc is used. V. EXAMPLE To illustrate these concepts, we consider a quantum dot with a single level coupled to two reservoirs with chemical potentials µα and temperatures Tα, α = L, R as sketched in Fig. 2. We assume that the dot level and the barriers can be modulated periodically in time by ac gate potentials. This model can describe a single-electron source, similar to the GHz pump realized experimentally in Ref. 10. For noninteracting electrons, the model is described by the Hamiltonian H(t) = Hc(t) + Hres + HT . (34) for the efficiencies of heat engine and heat pump. We first show that the second law implies that these efficiencies are bounded by the corresponding Carnot ef- The first term describes the central conductor which is modeled as a discrete chain of N sites with local ener- gies εm, nearest-neighbor hopping w, and an ac potential 7 1 = V 0 3 = 4, V 0 FIG. 2. Sketch of the device. A single-level quantum dot (m = 2) is defined by two tunnel barriers (m = 1, 3). They are driven by periodic gate potentials Vm(t) = V 0 j cos(ωt + δm), with V 0 2 = 23, δ1 = 0, δ2 = π/2, and δ3 = π. The tunneling amplitudes between barriers and dot are w = 1 and wL = wR = 0.7 between barriers and reservoirs. The barriers and the dot are modeled by a discrete chain of N = 3 sites with local energies ε1 = ε3 = 3.3 and ε2 = −1 respectively. The reservoirs have µL = µ, µR = µ − δµ and temperature T . applied to each site, Hc(t) = (εm + Vm(t)) d† mdm + (cid:34) N(cid:88) m=1 (cid:35) FIG. 3. Maximum efficiency ηmax and transport coefficients at T = 0 for the motor (M ) or generator (G) modes. Inset: ηmax for generator/motor with µ = −0.7 (µ = 7.2). wd† mdm+1 +h.c. N−1(cid:88) m=1 (35) Specifically, we consider a setup with N = 3 sites, mod- eling the tunneling barriers (m = 1, 3) and the quantum dot (m = 2). The site energies εm (m = 1, 2, 3) are mod- ulated by three time-dependent gate voltages of the form Vm(t) = V 0 m cos(ωt + δm). The reservoirs are represented by free-electron Hamiltonians for free electrons, Hres = Ekαc † kα ckα, (36) (cid:88) α=L,R,kα HT = − (cid:88) and tunneling between reservoirs and central system is described by [wαd† nα ckα + h.c], (37) α,kα,n where nα denotes the site of the central conductor which is in contact with the reservoir α. The mean charge current N α and heat current Qα en- tering the reservoir α, as well as the mean power W de- veloped by the ac forces are calculated within a Floquet Green function formalism following Ref. 12, as reviewed in App. B. To derive the response coefficients Lij, we ex- pand the currents J1 = N R, J2 = QR, and J3 = W /(ω) to linear order in ω, see App. C. Explicit expressions for the coefficients Lij – in terms of Green functions12,35,40 or scattering matrices11 and valid for noninteracting sys- tems – can be found in App. D. These coefficients can also be calculated using an alternative procedure which does not rely on the Floquet decomposition, see Refs. 27 and 28. However, we prefer to use the Floquet approach be- cause this representation stresses that ω appears in the Fermi functions which enter the integrals for the currents on the same footing as the chemical potential µ. This provides an alternative argument for identifying ω/T as an affinity. For illustration, we consider an applied bias δµ at T = 0, i.e., the motor/generator regime, and calculate the coefficients listed in App. D for this case. In Fig. 3, we plot the transport coefficients and the maximum efficiency ηmax as functions of the chemical potential µ of the left reservoir. Large values of the figure of merit require a large charge pumping coefficient L13 along with a small value of L33L11, i.e., low friction or conductance. In the absence of driving at the central dot [V2(t) = 0], the conductance peaks near L11 = 1 when µ is in res- onance with the dot level. Driving the dot level with a phase lag relative to the barrier oscillations (δ2 − δm (cid:54)= 0 for m = 1, 3) favors charge pumping and decreases the conductance by dynamically tuning the dot off resonance. In this way, high efficiencies can be achieved despite large values of L33. As the dot the chemical potential passes level, the pumping coefficient changes sign, and the system switches from motor mode [L31δµ/T < 0; see region M in the Fig. 3] to generator mode [L31δµ/T > 0, see region G in the Fig. 3]. The efficiency becomes minimal when the chemical potential is resonant with the dot level, where the conductance is maximal and pumping vanishes by particle-hole symmetry. The device can also operate as a heat engine or pump when imposing a temperature gradient. As this requires finite T , quantum effects are less pronounced and effi- ciencies are lower than those shown in Fig. 3. However, we find that for appropriate parameters these may still !!!!""""be as high as ≈ 0.4ηc. VI. SUMMARY Motivated in part by Jarzynski's equality43 and Crook's theorem,44 there has been much interest in quan- tum thermodynamics, including fluctuation relations, work fluctuations, and the thermodynamic description of strongly coupled systems.45–47 Here, we provided a gener- alized thermoelectric framework to analyze the thermo- dynamics of ac-driven nanoscale systems which explic- itly accounts for the effects of quantum pumping and the related nonconservative forces. We identified the addi- tional flux and affinity through which these forces enter the theory and defined generalized Onsager relations for the associated response coefficients. This framework al- lowed us to define appropriate efficiencies and figures of merit which describe quantum motors, generators, heat engines, and heat pumps. We illustrated our results for a simple quantum-pump device. VII. ACKNOWLEDGMENTS We acknowledge useful discussions with A. Bruch, R. Bustos-Marun, S. Kusminskiy, and A. Nitzan. This work was supported by the Alexander-von-Humboldt Stiftung (L.A. and F.v.O.), CONICET (M.F.L. and F.B.), MIN- CyT and UBACyT (L.A.), as well as the Deutsche Forschungsgemeinschaft (F.v.O.). Appendix A: Response coefficients and microreversability The matrix elements entering Λcf t read dt(cid:48)(t − t(cid:48))θ(t − t(cid:48))(cid:104)(cid:104) J c(t), Fj(t(cid:48)) (cid:105)(cid:105)t (cid:90) ∞ (cid:90) ∞ j = −i Λcf −∞ ≡ dτ τ χJ,Fj t −∞ (τ ), (A1) (τ ) = −iθ(τ )(cid:104)(cid:104) J c(τ ), Fj(0) (cid:105)(cid:105)t. where we have defined the retarded susceptibility χJ,Fj In the latter step t we have stressed that for evolutions with the operator U (τ ) = e−iτ Ht, being Ht = H (V(t)) the frozen Hamil- tonian, the actual time argument of the integrand of (A1) is τ = t − t(cid:48). Representing the susceptibility in terms of the Fourier transform we can also write the previous ex- 8 (cid:21) (ω) = pression as Λcf j = Re = Re (cid:20)(cid:90) +∞ (cid:20) (cid:90) +∞ (cid:104) −∞ −i dτ −∞ χJ,Fj t Im (cid:90) +∞ −∞ (cid:105) (ω) e−iωτ ∂ωχJ,Fj dω 2πi t (cid:21) dω∂ωχJ,Fj t (ω)δ(ω) = can be written (A2) function = 0.25 (cid:105)(cid:105)t = , = lim ω→0 ω as (ω) (cid:104) (cid:104) χJ,Fj t ij = −i Λf f where we have used that Im the is odd in ω, hence Im (cid:105) (cid:90) ∞ (cid:20) spectral χJ,Fj (0) t Analogously, the matrix elements of Λf f t (cid:105) dt(cid:48)(t − t(cid:48))θ(t − t(cid:48))(cid:104)(cid:104) Fi(t), Fj(t(cid:48)) (cid:90) +∞ (cid:104) (cid:105)(cid:105)t. −iθ(τ )(cid:104)(cid:104) Fi(τ ), Fj(0) −∞ χFi,Fj t where χFi,Fj dω∂ωχFi,Fj −∞ −i = lim ω→0 (ω)δ(ω) = Re (ω) is the Fourier transform of χFi,Fj (cid:21) (cid:105) (ω) Im = ω , t t t (A3) (τ ) = The calculation of the conductivity follows the usual procedure of the Kubo formula presented in text books.25 We start by considering an extra perturbation due to the coupling to an electric field E(t) = ∂tA(t). In the Fourier domain the extra perturbation is H(cid:48)(ω) = J · E(ω)/(iω), which leads to the definition of the dc conductance (cid:104) (cid:104) (cid:105) (cid:105) Im χJ,J t (ω) Λcc = lim ω→0 ω , (A4) (ω) is the Fourier transform of χJ,J (τ ) = t (cid:105)(cid:105)t. −iθ(τ )(cid:104)(cid:104) J(τ ), J(0) where χJ,J t respect to δµ leads to Similarly, evaluating the forces in linear response with Λf c j = lim ω→0 Im (ω) χFj ,J t ω , (A5) (ω) is the Fourier transform of χFj ,J (τ ) = t −iθ(τ )(cid:104)(cid:104) Fj(τ ), J(0) where χFj ,J t (cid:105)(cid:105)t. t The above definitions indicate that the susceptibili- ties χOi,Oj , with Oi a generic operator, satisfy microre- versibility with respect to τ . It can be directly verified that (−τ ) = −iθ(−τ )(cid:104)(cid:104) Oi(−τ ), Oj(0) (cid:105)(cid:105)t = (cid:105)(cid:105)t = = iθ(−τ )(cid:104)(cid:104) Oj(τ ), Oi(0) χOi,Oj t (A6) = −iθ(−τ ) Im[χOj ,Oi t (ω)]e−iωτ . dω π (cid:90) +∞ −∞ Hence, under a transformation τ → −τ the coefficient Λij transforms to ΛOi,Oj ij = Re i Im[χOj ,Oi t (ω)] dτ τ θ(−τ )e−iωτ (cid:21) (cid:90) +∞ −∞ (cid:20) (cid:90) +∞ (cid:104) −∞ Im dω π χOj ,Oi t (cid:105) In the last step we have used (cid:82) 0 = lim ω→0 ω = ΛOj ,Oi ji (ω) . (A7) −∞ dτ τ e−iωτ = 1 ω2 + iπδ(cid:48)(ω). In the presence of a magnetic field B, a time-reversal transformation implies changing B → −B in the Hamil- tonian Ht defining the frozen density matrix ρt used to evaluate the expectation values. This property leads to the following Onsager relations for the usual susceptibilities in the presence of B, χOi,Oj (B, ω) = (−B, ω), where the signs si, sj = ± depend on sisjχOj ,Oi the parity of the operators Oi, Oj under a time-reversal transformation. t t Appendix B: Green function formalism In Ref. 12 it was shown that the averaged charge N α Qα currents entering the reservoir α, as well and heat as the mean power W developed by the ac forces can be written in terms of the retarded Green function of the central structure connected to the reservoirs expanded in the Floquet-Fourier transform as GR(t, t(cid:48)) = e−i E (t−t(cid:48)) G(n, E). (B1) (cid:90) ∞ ∞(cid:88) e−inωt dE 2π −∞ This function is calculated from the Hamiltonian H(t) by solving the Dyson equation (see Refs. 12 and 40). The resulting expression for the charge current is n=−∞ (cid:90) (cid:88) dE [fβ(E)−fα(E + nω)]Tαβ(n, E),(B2) N α = e h with n,β Tαβ(n, E) = Gαβ(n, E)2 Γβ Γα. Qα = Eα − µα N α e , The heat current reads with (cid:90) Eα = (cid:88) n dE [fβ(E)−fα(E + nω)]Tαβ(n, E).(B5) E h (B3) (B4) Similarly, the work performed by the ac potentials can be written as W = − 1 h ×Im (cid:90) +∞ (cid:104) V (n) G(n + l, E)Γα G†(l, E) (cid:105)(cid:111) (cid:88) (cid:110) dEnωfα(E) , (B6) −∞ α,l,n Tr (cid:32) where A=  2 9 n (cid:80) contact with the reservoir equal to wα22π(cid:80) where V (n) are the Fourier components of V (t) = V (n)einωt, being V (t) a matrix with diagonal ele- ments Vm(t). In the above expressions we introduced the hybridization matrix Γα which has a single element at the δ(E − Ekα ). For practical uses it can be considered in the wide band limit, thus, independent of E. The Fermi-Dirac distribution fα(E) = [1 + e(E−µα)/Tα ]−1 characterizes the thermal occupation of the electrons in the reservoirs (from now on we set the Boltzmann constant kB = 1). kα The other possible approach is the Floquet scatter- ing matrix formalism used in Ref. 11. The elements sij(Em, En) of the Floquet scattering matrix s(E), with En = E +nω, are the amplitudes for an electron to scat- ter from lead j to lead i after acquiring m − n Floquet quanta ω. The general relation between the Floquet scattering matrix elements and the Fourier coefficients for the Green's function is the generalized Fisher-Lee formula40 sij(Em, En) = δijδmn − i(cid:112)ΓiΓjGij(m − n, En). (B7) Appendix C: Linear response In order to calculate the currents Jl, l = 1, 2, 3 up to linear order in ω, δµ and δT we perform the following expansion of the Fermi function entering the integrands of Eqs. (B2), (B4) and Eq. (B6) fα(E + nω) ∼ fα(E) + nω∂Efα(E) (C1) (Tα − T ). We also evaluate G(n, E) up to linear order in ω by expanding the Dyson equation in powers of ω (see 35 and 36). Up to the first order in ω it reads G(t, E) ∼ Gf (t, E) + i G(1)(t, E), (µα − µ) − ∂f (E) ∂E − ∂f (E) ∂E (E − µ) T (C2) n=−∞ G(n, E)e−inωt. The first term is with G(t, E) =(cid:80)∞ the frozen Green function (cid:34) (cid:35)−1 Gf (t, E) = 1 E − Ht c + i Γ 2 , (C3) corresponding to the frozen Hamiltonian at time t, Ht c = Hc(t) (Γ collects the hybridization functions of the reser- voirs). The next term is first order in ω. It reads G(1)(t, E) = Gf (t, E) + A(t, E), ∂E∂t (C4)  2 ∂E Gf (t, E) d V dt Gf (t, E)− Gf (t, E) d V dt ∂E (cid:33) Gf (t, E) . (C5) The expansion of the Floquet scattering matrix up to the first order in the driving frequency ω cast sij(E, En) = + dt e−inωt[sij(t, E) + (C6) ∂Esij(t, E) + ωAij(t, E)]. (cid:90) τ 0 1 τ nω 2 Here sij(t, E) is the frozen scattering matrix. The matrix elements Aij(t, E) define a first order correction to the adiabatic scattering matrix. The frozen scattering matrix sij(t, E) as well as Aij(t, E) do not change significantly on the energy scale ω and T and depend on the specific realization of the scatterer. Anyway, it can be shown that, due to the unitarity of the Floquet scattering matrix 10 (cid:19) (cid:18) ∂s† ∂t and of the frozen scattering matrix they satisfy11 ω[s† A + A†s] = i 2 − ∂s† ∂E ∂s ∂t ∂s ∂E . (C7) Equation (B7) defines an explicit relation between A and A.40 Appendix D: Transport coefficients Substituting the expansions for the Fermi function, Eq. (C1), and for the Green function, Eq. (C2), into Eqs. (B2, B4) and Eq. (B6) and collecting terms up to first ω order in the affinities X1 = δµ T we obtain: T , X2 = δT T 2 and X3 = (cid:90) T 0 dt L11 = − T hT L12 = L21 = − T hT L13 = −L31 = − T 2πh (cid:90) T 0 dt L22 = − T hT L23 = −L32 = − T 2πh L33 = − TT (cid:90) T dt 8π2h 0 0 dt −∞ (cid:90) +∞ (cid:90) T (cid:90) T (cid:90) +∞ (cid:90) T (cid:90) +∞ −∞ 0 0 dE Gf df dE (cid:90) +∞ (cid:90) +∞ −∞ RL(t, E)2 ΓL ΓR (cid:40)(cid:34) df dE Gf dE(E − µ) dt −∞ dE df dE RL(t, E)2 ΓL ΓR Im Gf (t, E)Γ ∂ Gf†(t, E) ∂t (cid:41) (cid:35) Γ RR dE (E − µ)2 df dE (cid:90) +∞ Gf (cid:40)(cid:34) RL(t, E)2 ΓL ΓR dt −∞ dE (E − µ) df dE (cid:40) (cid:34) dE df dE −∞ Re Tr ∂ Gf (t, E) ∂t Γ ∂ Gf†(t, E) ∂t Im Gf (t, E)Γ (cid:41) (cid:35) Γ RR ∂ Gf†(t, E) (cid:35)(cid:41) ∂t Γ . (D1) Within the scattering matrix formalism the coefficients read sRL(t, E)2 L11 = − T hT L12 = L21 = − T hT L13 = −L31 = − T 2πh (cid:90) T 0 dt (cid:90) T 0 dt L22 = − T hτ L23 = −L32 = − T 2πh L33 = − TT (cid:90) T dt 8π2h 0 0 dt −∞ (cid:90) +∞ (cid:90) T (cid:90) T (cid:90) +∞ (cid:90) T (cid:90) +∞ −∞ 0 0 −∞ dE df dE (cid:90) +∞ (cid:90) +∞ −∞ dt dE −∞ df dE dE(E − µ)2 df dE (cid:90) +∞ dt −∞ df dE dE Tr dE(E − µ) sRL(t, E)2 df dE (cid:26)(cid:20) (cid:21) (cid:27) ∂s†(t, E) ∂t RR Im s(t, E) sRL(t, E)2 (cid:26)(cid:20) dE(E − µ) (cid:20) ∂s(t, E) Im df dE ∂s†(t, E) (cid:21) s(t, E) ∂t ∂t (cid:21) (cid:27) ∂s†(t, E) ∂t RR . (D2) The matrices A in the Green-function language, and A in the scattering matrix version, in principle seem to con- tribute to the coefficient L33. In particular, they appear in an integrand of the form (cid:26) (cid:88) ij 2Im Aij(t, E) (cid:27) . ∂s∗ ij(t, E) ∂t (D3) However, as shown in Ref. 27, due to the unitary con- dition of the frozen scattering matrix ss† = 1 and the property (C7) such term vanishes. In fact, it can be also written as = 11 = (D4) (cid:105)(cid:111) (cid:110) (cid:104) (cid:105) (cid:104)(cid:16) Tr(cid:2)(cid:0)∂ts†∂E s − s†∂E s∂ts†s(cid:1) ∂ts†s(cid:3) = 0. ∂ts† A s† A + A†s ∂ts† A − A†∂ts (cid:17) = −iTr ∂ts†s (cid:104) (cid:105) = 2Im Tr = −iTr 1 2ω 1 L. Onsager, Phys. Rev. 38, 2265 (1931); M. Buttiker, Phys. Rev. Lett. 57, 1761 (1986); P. N. Butcher, J. Phys.: Con- dens. Matter 2, 4869 (1990). 2 G. Benenti, G. Casati, T. Prosen, K. Saito, arXiv:1311.4430. 24 We assume for definiteness that all parameters Vj(t) trans- form in the same way under time reversal. 25 H. Bruus and K. Flensberg, Many-body quantum field the- ory in condensed matter physics, (Oxford, 2005). 26 J. T. Lu, M. Brandbyge, and P. Hedegard, Nano Lett. 10, 3 G. Benenti, K. Saito, and G. Casati, Phys. Rev. Lett. 106, 1657 (2010). 230602 (2011). 27 N. Bode, S. Viola Kusminskiy, R. Egger, and F. von Op- 4 K. Brandner, K. Saito, and U. Seifert, Phys. Rev. Lett. pen, Phys. Rev. Lett. 107, 036804 (2011). 110, 070603 (2013). 28 N. Bode, L. Arrachea, G. S. Lozano, T. S. Nunner, and F. 5 O. Entin-Wohlman, Y. Imry, and A. Aharony, Phys. Rev. von Oppen, Phys. Rev. B 85, 115440 (2012) . B 91, 054302 (2015). 6 M. V. Vedernikov and E. K. Iordanishvili, 17th Int. Conf. on Thermoelectrics vol 1, pp. 37-42 (1998) 7 D. J. Thouless, Phys. Rev. B 27, 6083 (1983). 8 M. Buttiker, H. Thomas, and A. Pretre, Z. Phys. B 94, 133 (1994). 9 P.W. Brouwer, Phys. Rev. B 58, 10135(R) (1998). 10 M. D. Blumenthal, B. Kaestner, L. Li, S. Giblin, T. J. B. M. Janssen, M. Pepper, D. Anderson, G. Jones, and D. A. Ritchie, Nature Phys. 3, 343 (2007). 11 M. Moskalets and M. Buttiker, Phys. Rev. B 69, 205316 (2004). 29 Similar relations were found for closed systems (i.e., not coupled to reservoirs) without applied dc bias voltages in: D. Cohen, Phys. Rev. B 68, 155303 (2003). 30 T. A. Shutenko, I. L. Aleiner, and B. L. Altshuler, Phys. Rev. B 61, 10366 (2000). 31 I. L. Aleiner, B. L. Altshuler, and A. Kamenev, Phys. Rev. B 62, 10373 (2000). 32 M.G. Vavilov and I. L. Aleiner Phys. Rev. B 64, 085115 (2001). 33 L. DiCarlo, C. M. Marcus, and J. S. Harris Jr, Phys. Rev. Lett. 91, 246804 (2003). 34 M. Moskalets and M. Buttiker, Phys. Rev. B 72, 035324 12 L. Arrachea, M. Moskalets, and L. Martin-Moreno, Phys. (2005). Rev. B 75, 245420 (2007). 35 M. F. Ludovico and L. Arrachea, Phys. Rev. B 87, 115408 13 M. Moskalets and M. Buttiker, Phys. Rev. B 80, 081302(R) (2013). (2009). 14 R. Bustos-Marun, G. Refael, and F. von Oppen, Phys. Rev. Lett. 111, 060802 (2013). 15 A. M. Fennimore, T. D. Yuzvinsky, W. Q. Han, M. S. Fuhrer, J. Cumings, and A. Zettl, Nature 58, 408410, (2003). 16 A. del Campo, J. Goold, and M. Paternostro, Scientific 36 M. F. Ludovico, J. S. Lim, M. Moskalets, L. Arrachea, and D. Sanchez, Phys. Rev. B 89, 161306(R) (2014). 37 G. D. Mahan Many Particle Systems, New York: Plenum, (1990). 38 J. M. Luttinger, Phys. Rev. 135 A1505 (1964); Phys. Rev. 136 A1481 (1964). 39 For a review, see B. S. Shastry, Rep. Prog. Phys. 72, Reports 4, 6208 (2014). 016501 (2009). 17 F. Mazza, R. Bosisio, G. Benenti, V. Giovannetti, R. Fazio, 40 L. Arrachea and M. Moskalets, Phys. Rev. B 74, 245322 and F. Taddei, New J. Phys. 16, 085001 (2014). 18 T. E. Humphrey, R. Newbury, R. P. Taylor, and H. Linke, Phys. Rev. Lett. 89, 116801 (2002). (2006). 41 H. Callen, Phys. Rev. 73, 1349 (1948). 42 L. D. Landau and E. M. Lifshitz, Statistical Physics, Vol.5, 19 T. E. Humphrey and H. Linke, Phys. Rev. Lett. 94, 096601 Pergamon Press, London-Paris (1959). (2005). 20 A. Cr´epieux, F. Simkovic, B. Cambon, F. Michelini, Phys. Rev. B 83, 153417 (2011). 21 S. Juergens, F. Haupt, M. Moskalets, and J. Splettstoesser, 43 C. Jarzynski, Phys. Rev. Lett. 78, 2690 (1997). 44 G. E. Crooks, Phys. Rev. E 60, 2721 (1999). 45 J. Kurchan, J. Stat. Mech. P07005 (2007). 46 M. Esposito, U. Harbola, and S. Mukamel, Rev. Mod. Phys. Rev. B 87, 245423 (2013). Phys. 81, 1665 (2009). 22 J. S. Lim, R. L´opez, and D. S´anchez, Phys. Rev. B 88, 47 M. Campisi, P. Hanggi, and P. Talkner, Rev. Mod. Phys. 201304(R) (2013). 23 A.-M. Dar´e and P. Lombardo, arXiv:1509.05606. 83, 771 (2011). 

1807.11550

2

1807

Duplication, collapse and escape of magnetic skyrmions revealed using a systematic saddle point search method

Various transitions that a magnetic skyrmion can undergo are found in calculations using a method for climbing up the energy surface and converging onto first order saddle points. In addition to collapse and escape through a boundary, the method identifies a transition where the skyrmion divides and forms two skyrmions. The activation energy for this duplication process can be similar to that of collapse and escape. A tilting of the external magnetic field for a certain time interval is found to induce the duplication process in a dynamical simulation. Such a process could turn out to be an important avenue for the creation of skyrmions in future magnetic devices.

cond-mat.mes-hall

cond-mat

Duplication, collapse and escape of magnetic skyrmions revealed using a systematic saddle point search method Gideon P. Müller,1, 2, 3 Pavel F. Bessarab,1, 4 Sergei M. Vlasov,1, 4 Fabian Lux,2, 3 Nikolai S. Kiselev,2 Stefan Blügel,2 Valery M. Uzdin,4, 5 and Hannes Jónsson1, 6, ∗ 1Science Institute and Faculty of Physical Sciences, University of Iceland, VR-III, 107 Reykjavík, Iceland 2Peter Grünberg Institut and Institute for Advanced Simulation, Forschungszentrum Jülich and JARA, 52425 Jülich, Germany 3RWTH Aachen University, D-52056 Aachen, Germany 4ITMO University, 197101, St. Petersburg, Russia 5Dpt. of Physics, St. Petersburg State University, St. Petersburg 198504, Russia 6Dpt. of Applied Physics, Aalto University, FIN-00076 Espoo, Finland Various transitions that a magnetic skyrmion can undergo are found in calculations using a method for climbing up the energy surface and converging onto first order saddle points. In addition to collapse and escape through a boundary, the method identifies a transition where the skyrmion divides and forms two skyrmions. The activation energy for this duplication process can be similar to that of collapse and escape. A tilting of the external magnetic field for a certain time interval is found to induce the duplication process in a dynamical simulation. Such a process could turn out to be an important avenue for the creation of skyrmions in future magnetic devices. Localized, non-collinear magnetic states are receiving a great deal of attention, where skyrmions have come under special focus. Along with interesting transport properties, skyrmions exhibit particle-like behaviour and carry a topological charge enhancing their stability with respect to uniform ferromagnetic background. In addi- tion to the interest in their intriguing properties, they have been suggested as a basis for technological applica- tions e.g. data storage or even data processing devices [1, 2]. A racetrack design of a memory device has been outlined where a spin polarized current drives a chain of skyrmions past a reading device [3, 4]. The effect of tem- perature and external magnetic field on the stability of the skyrmions need to be studied, as well as ways to gen- erate and manipulate them. The effect of defects is also an important consideration [5, 6]. Two mechanisms for the annihilation of skyrmions have been characterized by theoretical calculations of atomic scale systems: Collapse of a skyrmion to form ferromagnetic state [7 -- 11] and es- cape of a skyrmion through the boundary of the magnetic domain [9 -- 12]. The effect of a non-magnetic impurity has also been calculated [11]. By using harmonic transition state theory for magnetic systems [14, 15], the lifetime of skyrmions has been estimated [11, 12]. Parameter values obtained from density functional theory [13] are found to give results that are consistent with experimental ob- servations [16, 17]. The challenge is to design materials where magnetic skyrmions are small enough while being sufficiently stable at ambient temperature, and to de- velop methods for manipulating them. Theoretical calculations can help accelerate this devel- opment by identifying the various possible transforma- tions that a skyrmion can undergo at a finite temperature on a laboratory time scale. This can be achieved by the use of rate theory where the major challenge is to find the relevant transition mechanisms. If the final state of a transition is specified, in addition to the initial state, the geodesic nudged elastic band (GNEB) method [7, 18] can be used to find the minimum energy path of the tran- sition and, thereby, the activation energy which is the highest rise in energy along the path. However, the final states of possible transitions are not always known. An- other category of methods for identifying possible transi- tion mechanisms where the final states are not specified, only the initial state, has turned out to be highly valu- able in a different context, namely in studies of atomic rearrangements such as chemical reactions and diffusion events [19, 20]. Unexpected transition mechanisms have in many cases turned out to be preferred over mecha- nisms that seem a priori most likely [21]. Here, we describe a method that can be applied to identify transition mechanisms in magnetic systems with- out specifying final states. It represents an adaptation of a method that has been used extensively in studies of atomic rearrangements. A complication arises from the fact that magnetic systems are characterized by the ori- entation of the magnetic moments while the length of the magnetic moments is either fixed or obtained from self- consistency calculations [22]. The configuration space is therefore curved. The method is first described briefly, with more detailed information in Supplemental Material [23], and then an application to a magnetic skyrmion is presented where, in addition to collapse and escape, the method gives a mechanism and activation energy that has not been reported before: Duplication of a skyrmion. Fi- nally, a dynamical simulation is described where a time dependent external field is used to induce such an event. Within harmonic transition state theory [14, 15, 19], the mechanism and rate of a thermally induced transi- tion is characterized by the first order saddle point [24] representing the bottleneck for the transition. Given an initial state corresponding to a local energy minimum, the various possible transitions the system may undergo can be identified by climbing up the energy surface and converging on the various first order saddle points on the energy ridge surrounding the minimum. A version of this approach, referred to as minimum mode following [26, 27], is illustrated for a single-spin test system in Fig. 1. The method is based on the evaluation of eigenvalues and corresponding eigenvectors of the Hessian, H. First, the region near the minimum, the convex region where all eigenvalues of the Hessian are positive, is escaped by following, for example, the gradient of the energy. Alter- natively, a random vector or one of the eigenvectors of H can be followed to escape the convex region. Once an eigenvalue turns negative, an effective force F eff = F − 2(λ · F ) λ, (1) is followed, where F = −∇H is the negative gradient of the energy and λ is the normalized eigenvector cor- responding to the negative eigenvalue. Note that these vectors are 3N-dimensional for a system with N spins. As the system is displaced in the direction of the effec- tive force, it moves to higher energy along λ but to lower energy along the orthogonal degrees of freedom. Eventu- ally, this brings the system to a first order saddle point on the energy surface. The final state of the transition can be obtained by a slight displacement further along λ, followed by energy minimization. This approach has not, to the best of our knowledge, been applied previously to a magnetic system. Here, the configuration space Mphys is given by the direct prod- uct of N spheres. In order to apply the mode follow- ing method, knowledge of second order derivatives is re- quired but, as is well-known from the theory of Rieman- nian manifolds, they need to be treated with special care. The Hessian can be calculated by application of covariant derivatives, but their evaluation in spherical coordinates is usually cumbersome and suffers from singularities at the poles. A more convenient approach is offered by viewing the configuration space as being embedded in a surrounding euclidean space E ⊃ Mphys. In this larger space, sec- ond order derivatives are easily performed. The Hessian in the physical subspace can then be reconstructed by a projection operator approach [28]. For any scalar func- tion f on the manifold Mphys, this true Hessian is defined as x ∂ ¯f ), Hess f (x)[z] = Px∂2 ¯f (x)z + Wx(z, P ⊥ (2) where z is a vector tangent to the manifold, Px and P ⊥ x are the projectors onto the tangent space and onto the normal space, respectively, to the surface of the manifold at a point x on the sphere, ¯f is the smooth extension of f to the Euclidean space, and Wx is the Weingarten map at x. The Weingarten map, sometimes also referred to as the shape operator, describes the curvature of 2D surfaces in terms of an embedding space. For a Hamiltonian H of a spin system, the matrix rep- resentation of the Hessian is obtained by its action on the 2 FIG. 1. An illustration of a method for climbing up the energy surface from an initial state minimum to a first order saddle point. The system contains only a single spin (see [23]) and because only the orientation of the magnetic vector can change, the energy surface can be mapped onto a sphere. The local energy minima (blue) are separated by an energy ridge where points of low energy correspond to first order saddle points (black ×). The red curves illustrate saddle point search paths starting from various points. In this illustration, the system is made to follow the gradient of the energy until the lowest eigenvalue of the Hessian becomes negative (at the white dashed line). Beyond that point, the system is displaced along the effective force given by Eq. (1). This brings the system to a first order saddle point on the energy surface. basis vectors (see [23]), i.e., ¯HijTj − T T i I(nj · ∇j ¯H)Tj, Hij = T T i (3) where the indices i and j denote spins, ¯H = ∂2 ¯H(x), I is the 3 × 3 unit matrix and Ti is the 3 × 2 matrix that transforms into the tangent space of spin i. As the Hessian matrix given by Eq. (3) is represented in 2N, the evaluation of an eigenmode in the 3N-representation requires a transformation, i.e. λ3N = T λ2N (see [23]). The formulation of the constrained Hessian given by Eq. (3) and the saddle point search method described above have been implemented in the Spirit [29] software and used here to analyse transitions from a magnetic skyrmion state. The energy of the system is described by an extended Heisenberg model H = −µS H· ni − J ni · nj − D dij · (ni × nj), N(cid:88) i=1 (cid:88) (cid:104)ij(cid:105) (cid:88) (cid:104)ij(cid:105) (4) where H is a uniform external magnetic field, ni is the magnetic moment of spin i, J is the exchange coupling between nearest neighbor spins, the parameter D and unit vector dij give the Dzyaloshinskii-Moriya vector in the plane of the lattice parallel to the vector connecting the two nearest neighbors i and j. The sums include only distinct nearest neighbor pairs. The system consists of 40×40 spins on a square lattice with free boundary con- ditions and the parameter values and field strength are chosen to be the same as in a recent theoretical study [30] (values are given in [23]). The Bloch skyrmion, shown in Fig. 2, is metastable with respect to the ferromagnetic phase. Fig. 2 shows an illustration of the eigenvectors corre- sponding to the three lowest eigenvalues. They corre- spond to translation, breathing and elliptical distortion [31]. In order to escape from the convex region, we have chosen here to follow each one of these three modes un- til the corresponding eigenvalue becomes negative. After that, the system is displaced in the direction of the ef- fective force given by Eq. (1), where λ is the eigenvector of the initially selected mode rather than the one with lowest eigenvalue. FIG. 2. Top panel: (Left) Minimum energy configuration of the skyrmion. (Right) Minimum energy paths for the three types of transitions found: Duplication, collapse and escape. The reaction coordinate is the scaled total displacement along the path. The energy is given in units of the exchange cou- pling constant, J. Middle panel: Saddle point configurations found for escape through a boundary (left), radial collapse (middle) and duplication (right). Bottom panel: The transla- tional, breathing and elliptical eigenmodes of the skyrmion in the minimum energy configuration, each leading to the saddle point displayed directly above. By following the translational mode, the mode that has lowest eigenvalue initially, the skyrmion moves towards the edge of the system and eventually, as it is pushed along the effective force, converges on a saddle point cor- 3 responding to an escape through the edge. Relaxation of the system after a slight displacement along the unstable mode at the saddle point brings the system to the fer- romagnetic state. Similarly, by following the breathing mode, the second lowest eigenvalue mode, the skyrmion shrinks and eventually, converges to a saddle point cor- responding to collapse. Again the final state of the tran- sition is the ferromagnetic state. However, by following the elliptical mode, the skyrmion becomes stretched and converges on a saddle point corresponding to a division of the skyrmion and formation of two skyrmions. While the initial state contains one skyrmion, the final state contains two. The energy along the minimum energy paths for the three transitions is shown in Fig. 2. These were calcu- lated using the geodesic nudged elastic band method [7] using an initial path formed by linear interpolation be- tween the intial state and the saddle point, as well as between the saddle point and the final state. The activa- tion energy for the collapse and escape is quite similar, but the activation energy for the duplication is higher in this case. The relative height of the energy barriers for the three processes depends on the parameter values. A calculation using a 10% smaller field and 60% larger value of D/J gives a lower activation energy for duplica- tion than the other two transitions [23]. The shape of the three curves is quite different. The total displacement along the paths from the initial state to the saddle point is shortest for the duplication, while the displacement of the skyrmion to the boundary and to the saddle point for escape involves the largest displace- ment. The energy profile for the collapse is similar to what has been presented before, a gradual shrinkage of the skyrmion to the saddle point followed by abrupt drop of the energy to that of the ferromagnetic state [7, 8]. The duplication of the skyrmion leads to an increase in the energy of the system because the skyrmion is only metastable with respect to the ferromagnetic state for this set of parameters, while the other two transitions lead to a decrease in the energy. The energy along the minimum energy path past the saddle point for duplica- tion contains important information about skyrmion in- teraction. It shows how the repulsive interaction between skyrmions varies with the distance between them. The minimum energy path going from right to left along the mininum energy path in Fig. 2 shows how two skyrmions can merge to form one skyrmion. The duplication transition identified here has not been described previously, but may turn out to be an impor- tant mechanism for generating skyrmions. The first order saddle point on the energy surface corresponding to this mechanism can be found for a wide range of parameter values, as will be described in detail in a later publica- tion. For example, we have found the duplication saddle point for a Pd/Fe bilayer on an Ir(111) substrate, a sys- tem that has been studied extensively, using parameter 4 FIG. 3. Snapshots from a dynamical simulation where the duplication of a skyrmion is induced by applying a magnetic pulse over 200 psec, giving a total field that is tilted with respect to the normal to the plane. (see [23] for parameter values). The labels on top of the frames give the time and magnetic field of the pulse (which adds to a constant field in the direction of the normal to the plane). The pulse is applied at time t=0 and lasts until t=200 ps. At the end of the pulse, an elongated defect is formed which later splits up (at t=209 ps) to form two skyrmions. values that give close agreement with experimental re- sults [12, 13]. There, the energy barrier for duplication turns out to be slightly lower than that of collapse, 78 meV vs. 80 meV. The question now arises whether it is possible to in- duce the duplication event dynamically. We address this by performing a dynamical simulation where a short mag- netic pulse is applied in addition to the stationary mag- netic field. The pulse is represented by a uniform mag- netic field which has both in-plane and out-of-plane com- ponents. For the duration of the pulse, 200 ps, the total magnetic field is tilted (see [23]). This is known to be an efficient way of exciting nonlinear skyrmion dynam- ics [32]. The saddle point searches show that an excita- tion of the elliptical mode can induce duplication. Fig. 3 shows shapshots from the simulation. During the pulse, an elongated structure is formed. After the pulse has been turned off, it splits up and eventually two skyrmions remain in the system. In addition to the elliptic elonga- tion, the duplication also requires the deformed structure to be bent in order for it to divide up. The bending corre- sponds to another eigenmode of the elongated skyrmion. More efficient and/or reliable ways of achieving the dupli- cation process could be devised, for example by pinning the skyrmion, introducing defects or by other more elab- orate external stimuli. The simulation described here is merely meant to show that it is possible to induce skyrmion duplication in a rather simple way. We have presented here a method that can be used to search for transition mechanisms and determine the activation energy for transitions in magnetic systems. It represents an adaptation of mode following methods that have been used for several years in studies of atomic rear- rangements. The fact that magnetic transitions involve rotations of the magnetic vectors leads to constraints that need to be incorporated into the Hessian matrix in order to determine the low lying eigenvalues and corresponding eigenvectors. Saddle point searches where only the ini- tial state of the transition is specified, such as the ones carried out here, are more challenging than a minimum energy path calculation where both the initial and final states are specified, but they have the advantage of being able to reveal unexpected transitions and unknown final states. In the application calculations presented here, we have chosen to drive the system out of the convex region by following a particular mode. We note that, eigenval- ues may cross and modes change direction as the system is pushed along a specific mode, so care should be taken to remain on the same mode throughout the saddle point search (see [23]). Applications of the saddle point search method pre- sented here to three-dimensional systems can be expected to yield an even larger variety of mechanisms than for two-dimensional systems since more possibilities open up with the additional dimension. For example, cylindrical skyrmions can contract and form bobbers [30] and pos- sibly other previously unknown magnetic structures. A saddle point search method that only requires the initial state as input (unlike the GNEB method where the final state also needs to be specified), can be used to sample an energy surface in a systematic way and search for possible states of the system, each state corresponding to a local energy minimum. The search takes the system from one local minimum to another via first order saddle points. This can be used as the basis of a simulation of the long time scale evolution of a system that undergoes thermally activated transitions that may be enhanced by an external field. For each state visited, multiple saddle point searches are carried out to identify the most rele- vant, low energy saddle points. The transition rate cor- responding to each saddle point can be estimated using harmonic transition state theory, and a random number used to choose which transition will occur next based on the normalized rates. This is referred to as adaptive ki- netic Monte Carlo algorithm [33]. The simulated time can be estimated from the sum of rates. This approach has been used as the basis for simulations of long time scale evolution of complex atomic systems where atomic rearrangement events can involve non-intuitive displace- ments of multiple atoms [34, 35]. The adaptation pre- sented here of the saddle point search method to spins opens the possibility of carrying out such long time scale t=0 ps, 𝐻)=0 𝐻*t=100 ps, 𝐻)=0.63 𝐻*t=200 ps, 𝐻)=0 𝐻*t=210 ps, 𝐻)=0 𝐻*t=400 ps, 𝐻)=0 𝐻*simulations of magnetic systems. This work was funded by Icelandic Research Fund (grants 185405-051 & 184949-051), Academy of Fin- land (grant 278260), Russian Foundation of Basic Re- search (grant RFBR 18-02-00267 A), EU-H2020 MAG- icSky (grant 665095), DARPA TEE (#HR0011831554) from DOI, and DFG CRC 1238 (project C01). ∗ hj@hi.is [1] N. S. Kiselev, A. N. Bogdanov, R. Schäfer, and U. K. Rössler, Chiral skyrmions in thin magnetic films: new ob- jects for magnetic storage technologies? J. Phys. D: Appl. Phys. 44, 392001 (2011). [2] A. Fert, N. Reyren and V. Cros, Magnetic skyrmions: ad- vances in physics and potential applications, Nature Re- views Materials 2, 17031 (2017). [3] A. Fert, V. Cros, and J. Sampaio, Skyrmions on the track, Nature Nanotechnology 8, 152 (2013). [4] J. Müller, Magnetic skyrmions on a two-lane racetrack. New J. Phys. 19, 025002 (2017). [5] J. Müller, and A. Rosch, Capturing of a magnetic skyrmion with a hole, Phys. Rev. B 91, 054410 (2015). [6] C. Hanneken, A. Kubetzka, K. von Bergmann, R. Wiesen- danger, Pinning and movement of individual nanoscale magnetic skyrmions via defects, New J. Phys. 18, 055009 (2016). [7] P. F. Bessarab, V. M. Uzdin, and H. Jónsson, Method for Finding Mechanism and Activation Energy of Magnetic Transitions, Applied to Skyrmion and Antivortex Annihi- lation. Comp. Phys. Commun. 196, 335 (2015). [8] I. Lobanov, H. Jónsson, and V. M. Uzdin, Mechanism and activation energy of magnetic skyrmion annihilation obtained from minimum energy path calculations, Phys. Rev. B 94, 174418 (2016). [9] D. Stosic, J. Mulkers, B. Van Waeyenberge, T. Luder- mir, and M.V. Milosević, Paths to collapse for isolated skyrmions in few-monolayer ferromagnetic films. Phys. Rev. B 95, 214418 (2017). [10] D. Cortés-Ortuno et al., Thermal stability and topolog- ical protection of skyrmions in nanotracks, Scientific Re- ports 7, 4060 (2017). [11] V.M. Uzdin, M.N. Potkina, I.S. Lobanov, P.F. Bessarab and H. Jónsson, Energy surface and lifetime of magnetic skyrmions, Journal of Magnetism and Magnetic Materials 459, 236 (2018). [12] P. F. Bessarab, G. P. Müller, I. S. Lobanov, F. N. Rybakov, N. S. Kiselev, H. Jónsson, V. M. Uzdin, S. Blügel, L. Bergqvist, and A. Delin, Lifetime of Racetrack Skyrmions, Sci. Rep. 8, 3433 (2018). [13] S. von Malottki, B. Dupé, P. F. Bessarab, A. Delin, and S. Heinze, Enhanced skyrmion stability due to exchange frustration. Sci. Rep. 7, 12299 (2017). [14] P. F. Bessarab, V. M. Uzdin, and H. Jónsson, Har- monic Transition-State Theory of Thermal Spin Transi- tions, Phys. Rev. B 85, 184409 (2012). [15] P. F. Bessarab, V. M. Uzdin, and H. Jónsson, Potential Energy Surfaces and Rates of Spin Transitions. Zeitschrift Für Physikalische Chemie 227, 1543 (2013). [16] N. Romming, C. Hanneken, M. Menzel, J. E. Bickel, B. 5 Wolter, K. von Bergmann, A. Kubetzka, and R. Wiesen- danger, Writing and deleting single magnetic skyrmions, Science 341, 636 (2013). [17] A. Kubetzka, C. Hanneken, R. Wiesendanger, and K. von Bergmann, Impact of the skyrmion spin texture on magnetoresistance, Phys. Rev. B 95, 104433 (2017). [18] P. F. Bessarab, Comment on "Path to collapse for an isolated Néel skyrmion", Phys. Rev. B 95, 136401 (2017). [19] B. Peters, Reaction Rate Theory and Rare Events (Else- vier Science & Technology, 2017). [20] H. Jónsson, Simulation of Surface Processes, Proceedings of the National Academy of Sciences 108, 944 (2011). [21] G.H. Jóhannesson and H. Jónsson, Optimization of Hy- perplanar Transition States. J. Chem. Phys. 115, 9644 (2001). [22] P. F. Bessarab, V. M. Uzdin, and H. Jónsson, Calcu- lations of magnetic states and minimum energy paths of transitions using a non-collinear extension of the Alexander-Anderson model and a magnetic force theorem, Phys. Rev. B 89, 214424 (2014). [23] See Supplemental Material for details on the methodol- ogy, description of the 2-dimensional test problem, param- eter values used in the calculation described in the main text, as well as another set of values and the corresponding minimum energy paths. [24] At a first order saddle point the gradient of the energy vanishes and the Hessian, the matrix of second derivatives of the energy, has one and only one negative eigenvalue. [25] A. Pedersen and H. Jónsson, Simulations of Hydrogen Diffusion at Grain Boundaries in Aluminum, Acta Mate- rialia 57, 4036 (2009). [26] G. Henkelman, and H. Jónsson. A Dimer Method for Finding Saddle Points on High Dimensional Potential Sur- faces Using Only First Derivatives, J. Chem. Phys. 111, 7010 (1999). [27] M. P. Gutiérrez, C. Argáez, and H. Jónsson. Improved Minimum Mode Following Method for Finding First Order Saddle Points. J. Chem. Theo. Comput. 13, 125 (2017). [28] P.-A. Absil, R. Mahony and J. Trumpf. An Extrinsic Look at the Riemannian Hessian. In Geometric Science of Information 361-368. Lecture Notes in Computer Sci- ence. Springer, Berlin, Heidelberg (2013). [29] Spirit - a spin simulation framework (see https:// spirit-code.github.io ) [30] Rybakov, F. N., Borisov, A. B., Blügel, S., Kiselev, N. S. New type of particlelike state in chiral magnets. Phys. Rev. Lett. 115 117201 (2015). [31] S-Z. Lin, C. D. Batista, and A. Saxena, Internal Modes of a Skyrmion in the Ferromagnetic State of Chiral Magnets Phys. Rev. B 89, 24415 (2014). [32] C. Heo, N. S. Kiselev, A. K. Nandy, S. Blügel, and T. Rasing, Switching of chiral magnetic skyrmions by picosec- ond magnetic field pulses via transient topological states, Sci. Rep. 6, 27146 (2016). [33] G. Henkelman and H. Jónsson, Long time scale kinetic Monte Carlo simulations without lattice approximation and predefined event table, J. Chem. Phys. 115, 9657 (2001). [34] S. T. Chill, M. Welborn, R. Terrell, L. Zhang, J-C. Berthet, A. Pedersen, H. Jónsson and G. Henkelman, EON: Software for long time simulations of atomic scale systems, Materials Science and Engineering 22, 055002 (2014). [35] A. Pedersen and H. Jónsson, Simulations of Hydrogen Diffusion at Grain Boundaries in Aluminum, Acta Mate- rialia 57, 4036 (2009). 6 Supplemental Material for "Duplication, collapse and escape of magnetic skyrmions revealed using a systematic saddle point search method" Gideon P. Müller,1,2,3 Pavel F. Bessarab,1,4 Sergei M. Vlasov,1,4 Fabian Lux2,3 Nikolai S. Kiselev,2 Stefan Blügel,2,3 Valery M. Uzdin,4,5 and Hannes Jónsson1,6 7 1Science Institute and Faculty of Physical Sciences, University of Iceland, VR-III, 107 Reykjavík, Iceland 2Peter Grünberg Institut and Institute for Advanced Simulation, Forschungszentrum Jülich and JARA, 52425 Jülich, Germany 3RWTH Aachen University, D-52056 Aachen, Germany 4ITMO University, 197101, St. Petersburg, Russia 5Dpt. of Physics, St. Petersburg State University, St. Petersburg 198504, Russia 6Dpt. of Applied Physics, Aalto University, FIN-00076 Espoo, Finland EVALUATION OF THE HESSIAN FOR SPIN SYSTEMS Eigenvectors of the Hessian matrix are needed in the saddle point search method. However, the second derivatives do not have a direct geometrical meaning and need to be treated within the theory of Riemannian manifolds [1]. Below, a representation is derived for the Hessian of spin systems where the spin length is fixed, i.e where the manifold Mphys of physical states is composed of the direct product of N spheres N(cid:79) Mphys = S2 ⊂ R3N . (S1.1) i=1 Therefore, Mphys is a submanifold of the embedding euclidean space E = R3N. Due to the singularities that can arise when using spherical coordinates, we choose to use a 3N cartesian represen- tation of the spin coordinates. As the Hamiltonian H can be extended to a function ¯H which is defined on E, the derivatives can be calculated easily. We denote the gradient taken in the embedding space E as ∂ ¯H and the gradient taken on the manifold Mphys as Px∂ ¯H, i.e., the projection of the gradient onto the tangent space. The matrix of second derivatives in the embedding space is denoted ∂2 ¯H. In order to calculate the second derivatives on Mphys while keeping the description in Cartesian coordinates, the projector approach described in Ref. [2] is used. For any scalar function f on the manifold Mphys, this true, covariant Hessian is defined as Hess f (x)[z] = Px∂2 ¯f (x)z + Wx(z, P ⊥ x ∂ ¯f ). Here, Wx denotes the Weingarten map which for any vector v at a point x of a spherical manifold is given by Wx(z, v) = −zxT v, (S1.2) (S1.3) where z is a tangent vector to the sphere at x. In order to calculate the Hessian, v = P ⊥ x ∂ ¯H = −zxT xxT ∂ ¯H = −zxT ∂ ¯H, x ∂ ¯H) = −zxT P ⊥ Wx(z, P ⊥ x ∂ ¯H is inserted to give where xT ∂ ¯H is the scalar product of the spin with the gradient. In Spirit [3], the Hessian is implemented in matrix representation, so we switch notation and drop the subscript x. For spin indices i and j, we write the gradient ∂ ¯H as ∇i ¯H (now to be understood as a three-dimensional object) and the second derivative ∂2 ¯H as ¯H. We reformulate Eq. (S1.2) within the Euclidean representation as a 3N × 3N matrix H3N = (Hij3N ) = H113N H123N ··· H213N H223N ··· ... ... ...  (S1.4) (S1.5) which consists of N 2 blocks that correspond to the different spin-spin subspaces. This matrix representation of the Hessian is obtained by acting with Eq. (S1.2) on the euclidean basis vectors of the extended space. Here, each of the 3 × 3 subspace matrices is defined as Hij3N = Pi ¯Hij − δijInj · ∇j ¯H, (S1.6) 8 where I denotes the 3 × 3 unit matrix and nj the normalized direction of spin j. The resulting matrix H3N will, however, have N eigenvectors orthogonal to the tangent space of Mphys, representing unphysical degrees of freedom in the embedding space E. In order to remove these unphysical degrees of freedom, the matrix is transformed using the tangent basis to Mphys and written as Hij = T T i Hij3N Tj, where Ti is the basis transformation matrix of spin i fulfilling T T P = T T and T T T = I2N. Thereby, the true Hessian of Eq. (S1.2), in the 2N × 2N matrix representation, becomes H = (Hij) with the spin-spin blocks defined as ¯HijTj − T T i I(nj · ∇j ¯H)Tj, Hij = T T i (S1.7) and which now contains only the physical degrees of freedom. From this matrix, the eigenmodes λ2N are calculated. The 3N representation is obtained as λ3N = T λ2N. The transformation matrix Ti can be a 3 × 2 matrix of two tangent vectors to spin i, that can be obtained from any random vector and another vector found by orthogonalization. Here, we have chosen the unit vectors of spherical coordinates θ and ϕ, which gives cos θ cos ϕ − sin ϕ  = cos θ sin ϕ cos ϕ − sin θ 0 zx/rxy −y/rxy zy/rxy x/rxy −rxy 0  T = {eθ, eϕ} = √ where rxy = sin θ = orthogonalize them with respect to the spin vector. 1 − z2. Note that the poles need to be excluded, but one may simply choose ex and ey and Finally, the Hessian matrix in the embedding space E = R3N is needed, denoted Hij3N. For a Hamiltonian of the form N(cid:88) H · ni − N(cid:88) Ki( K · ni)2 −(cid:88) Jij ni · nj −(cid:88) H = −µS i=1 i=1 (cid:104)ij(cid:105) (cid:104)ij(cid:105) Dij · (ni × nj), (S1.9) where H is the external magnetic field, K are uniaxial anisotropies (set to zero in the calculations presented here), Jij are the exchange constants and Dij the Dzyaloshinskii-Moriya vectors of the unique interaction pairs denoted by (cid:104)ij(cid:105), the spin-spin matrix blocks of ¯H = ∂2 ¯H become (S1.8) diagonal blocks: ¯Hi=j = −2Ki KxKx KxKy KxKz KyKx KyKy KyKz  KzKx KzKy KzKz off diagonal blocks: ¯Hi(cid:54)=j =  −(Jij + Jji) (−Dij + Dji)z −(−Dij + Dji)y (−Dij + Dji)x −(−Dij + Dji)z (−Dij + Dji)y −(−Dij + Dji)x −Jij − Jji −Jij − Jji (S1.10) (S1.11)  . SADDLE POINT SEARCH METHOD The saddle point search method is an iterative process where the spin configuration is modified according to a force acting on the spins. Each iteration involves finding the Hessian and its eigenvalues and eigenvectors. An important aspect of the adaptation of this method to spin systems is the evaluation of the Hessian for the curved configuration space. The optimization method involves rotating the spins in the direction of the force using the velocity projection method described in [4] until the force is zero. Outside the convex region, where one or more of the eigenvalues of the Hessian is negative, the effective force, Feff, is given by Eq. (4) in the main text. There are various approaches for escaping from the convex region around the initial state minimum: • Follow a (constant) random direction. • Make an initial (small) random displacement and then follow the gradient. 9 • Pick one of the eigenvectors of the Hessian and follow it. In the illustrative test problem shown in Fig. 1 in the main text, the second choice was made (due to the reduced dimensionality of the configuration space). In the saddle point searches for the skyrmion the third choice was made. The iterative optimization using Feff may lead back into the convex region. Also, eigenvalues may become degenerate and the lowest eigenvalue mode may change significantly throughout a calculation. When following a given mode, the scalar product of the current and previous mode is evaluated and, if a significant change in direction has occurred, the search is carried out for the mode that has the largest scalar product with the previous mode. This approach works due to the fact that eigenmodes are mutually orthogonal and should change continuously throughout configuration space. In special cases the gradient may become orthogonal to the mode to within a given tolerance. This case needs to be treated specially. The following algorithm was used in the present studies: • When the mode is not orthogonal to the gradient -- if the mode has a negative eigenvalue: follow Feff -- if the mode has a zero or positive eigenvalue: follow the mode. • When the mode is orthogonal to the gradient -- if the eigenvalue is zero (to within a tolerance): follow the gradient -- if the eigenvalue is positive: follow Feff. 1. For single spin illustration PARAMETER VALUES The energy surface of the single-spin system shown in Fig. 1 in the main text to illustrate the saddle point search method is a sum of 7 Gaussians of the form (cid:88) H = (cid:88) (cid:19) (cid:18) − l2 i (n) 2σ2 i Hi = ai exp , (S3.12) with parameters given in Table 1. i i TABLE I. Parameters of the Gaussians in the energy surface of the single-spin system shown in Fig. 1 in the main text. a −1.1 0.8 −0.9 9 · 10 0.13 −0.9 −0.9 σ 6 · 10 0.15 0.1 −2 3 · 10 3 · 10 0.1 0.1 px py pz −2 −0.2 0 −0.9 0.2 −0.2 −1 1 −0.2 −0.1 −2 0.8 0.5 −0.8 −2 0.8 −0.5 −0.7 1.2 −0.4 0.5 0.2 −0.9 −0.4 2. For Hamiltonian in the skyrmion calculations The values of the parameters in the Heisenberg type Hamiltonian, Eq. (4) in the main text, were chosen to be the same as in a previous theoretical study [5], D = 0.45 J and H = 0.8 HD, where HD = D2/(µSJ) is the critical field strength. For the parameter values used here and setting J=1 meV, gives HD = 3.5 T. 10 FIG. S3.1. Minimum energy paths for the three types of transitions using parameter values of H = 0.71 HD, D = 0.7 J. The reaction coordinate is the total rotation of all spins, normalized by the full length of the path. MINIMUM ENERGY PATHS In addition to the minimum energy paths shown in Fig. 2 in the main text, a calculation was carried out with another set of parameters, D = 0.7 J and H = 0.71 HD. In this case, the activation enegy is quite similar for the three transitions, the duplication barrier being lowest, as can be seen from Fig. S3.1 in this Supplemental Material. Here, the skyrmion state is lower in energy than the ferromagnetic state. DYNAMICAL SIMULATION In the dynamical simulation of the duplication process, a magnetic pulse is applied over a time period of 200 ps. The pulse has a magnetic field directed along the vector {x,z}={0.8,-0.61} while the static magnetic field is directed along the normal along the vector {x,z}={0,1} so the pulse partly cancels out the static field and leads to a tilt. The static field has magnitude H = 0.8 HD but during the pulse the total magnetic field drops to Hp = 0.63 HD. D = 0.45 J as in the calculations shown in Fig. 2 in the main text. The time evolution was carried out using the semi-implicit algorithm of Mentink et al. [6] with a time step of 0.01 ps. ∗ hj@hi.is [1] M. Nakahra. Geometry, topology and physics. CRC Press (2003). [2] P.-A. Absil, R. Mahony and J. Trumpf. An Extrinsic Look at the Riemannian Hessian. In Geometric Science of Information 361-68. Lecture Notes in Computer Science. Springer, Berlin, Heidelberg (2013). [3] Spirit - spin simulation framework (see https://spirit-code.github.io). [4] P. F. Bessarab, V. M. Uzdin, and H. Jónsson, Method for Finding Mechanism and Activation Energy of Magnetic Transitions, Applied to Skyrmion and Antivortex Annihilation. Comp. Phys. Commun. 196, 335 (2015). [5] Rybakov, F. N., Borisov, A. B., Blügel, S., Kiselev, N. S. New type of particlelike state in chiral magnets. Phys. Rev. Lett. 115 117201 (2015). [6] J.H. Mentink, M.V. Tretyakov, A. Fasolino, M.I. Katsnelson and Th. Rasing, Stable and fast semi-implicit integration of the stochastic Landau-Lifshitz equation, J. Phys.: Condens. Matter 22 176001 (2010). 00.20.40.60.8102468ReactioncoordinateE−E0[J]EscapeCollapseDuplication

1311.2514

1

1311

Tight-Binding Model for Adatoms on Graphene: Analytical Density of States, Spectral Function, and Induced Magnetic Moment

In the limit of low adatom concentration, we obtain exact analytic expressions for the local and total density of states (LDOS, TDOS) for a tight-binding model of adatoms on graphene. The model is not limited to nearest-neighbor hopping but can include hopping between carbon atoms at any separation. We also find an analytical expression for the spectral function $A({\bf k}, E)$ of an electron of Bloch vector ${\bf k}$ and energy E on the graphene lattice, to first order in the adatom concentration. We treat the electron-electron interaction by including a Hubbard term on the adatom, which we solve within a mean-field approximation. For finite Hubbard $U$, we find the spin-polarized LDOS, TDOS, and spectral function self-consistently. For any choice of parameters of the tight-binding model within mean field theory, we find a critical value of $U$ above which a moment develops on the adatom. For most choices of parameters, we find a substantial charge transfer from the adatom to the graphene host.

cond-mat.mes-hall

cond-mat

Tight-Binding Model for Adatoms on Graphene: Analytical Density of States, Spectral Function, and Induced Magnetic Moment N. A. Pike and D. Stroud Department of Physics, The Ohio State University, Columbus, OH 43210 (Dated: November 12, 2013) In the limit of low adatom concentration, we obtain exact analytic expressions for the local and total density of states (LDOS, TDOS) for a tight-binding model of adatoms on graphene. The model is not limited to nearest-neighbor hopping but can include hopping between carbon atoms at any separation. We also find an analytical expression for the spectral function A(k, E ) of an electron of Bloch vector k and energy E on the graphene lattice, to first order in the adatom concentration. We treat the electron-electron interaction by including a Hubbard term on the adatom, which we solve within a mean-field approximation. For finite Hubbard U , we find the spin-polarized LDOS, TDOS, and spectral function self-consistently. For any choice of parameters of the tight-binding model within mean field theory, we find a critical value of U above which a moment develops on the adatom. For most choices of parameters, we find a substantial charge transfer from the adatom to the graphene host. PACS numbers: 73.20.At, 73.20.Fz, 73.22.Pr, 75.70.Ak I. INTRODUCTION Graphene is a well known allotrope of carbon in which the carbons bond in a planar sp2 configuration[1, 2]. As a result, the single graphene sheet is effectively two- dimensional[3, 4]. Of the four n = 2 electrons which occupy the outer shell of a carbon atom, three are in sp2 orbitals and form in-plane π bonds between the nearest- neighbor carbon atoms, while the fourth occupies a 2pz orbital. These 2pz orbitals form a band of states which is responsible for many of the characteristic electronic properties of graphene[3]. Among these properties are a zero band gap at the so-called Dirac point, an elec- tronic dispersion relation that, near the Dirac point, is equivalent to that of massless Dirac fermions, and spin- orbit coupling which is believed to be small because of the low atomic number of carbon[3, 4]. Graphene has a vast number of potential applications, including photo- voltaic cells[5], ultracapacitors[6, 7], and spin-transport electronics[8–10]. Recently, a number of researchers have carried out experimental and theoretical investigations into the ef- fects of adatoms and impurities on both the band structure and localized magnetic moments in graphene. Among these are theoretical studies of carbon vacan- cies in graphene[11–13], hydrogen atoms on the sur- face of graphene[14], and several other types of disor- der in graphene[15, 16]. In several of these cases and in other work[16–18], impurity effects have been treated us- ing a tight-binding model for the electronic structure of graphene and impurities, vacancies, or adatoms. These calculations have, however, either been carried out nu- merically, or in the limit of energies close to the Dirac point, where the graphene density of states can be ap- proximated as linear[19]. Density-functional on adatoms for calculations graphene have also been carried out. They have shown that the introduction of an adatom bonded to the surface of graphene can lead to a quasi-localized state with an energy near the Fermi energy and that the wave function of this quasi-localized state includes contributions from the orbitals of neighboring carbon atoms[12, 14]. In some cases it has been found that, even if the introduced defect or adatom is non-magnetic, a localized magnetic moment can form at the defect site[12]. In this paper, we extend the previous work on graphene with adatoms in two ways. First, we show that the tight-binding model for adatoms on graphene can, in the limit of low concentrations, be solved analytically in the absence of electron-electron interactions. Specifi- cally, we obtain analytical expressions for the local den- sity of states (LDOS) on the adatom, the total density of states (TDOS) of the adatom-graphene system, and the spectral function A(k, E ) for an electron with Bloch vector k and energy E in graphene in the presence of the adatom[20]. All these results are expressed as a func- tion of the graphene density of states, which itself is known analytically for a nearest-neighbor tight binding model[18, 21]. Secondly, we calculate the magnetic properties of the system using the Hubbard model for the electron-electron interaction, which we treat using a standard mean-field approximation. This treatment leads to a transition be- tween a non-magnetic and magnetic state above a critical value of U which depends on the parameters of the tight- binding model. In the presence of a finite U our model is basically a special case of the well-known Anderson model[22], but with a linear rather than a constant den- sity of states near the Fermi energy. Our results include not only the magnetic moment on the adatom, but also that on the graphene sheet and the charge transfer from the adatom to the sheet, all as functions of the model parameters. The remainder of the article is arranged as follows: In Sec. II, we describe the model tight-binding Hamilto- nian for the graphene-adatom system. We also describe its generalization to include electron-electron interaction on the adatom via a Hubbard U term and the mean- field treatment of this term. In Sec. III, we describe the Green’s function method used to analytically calculate the local density of states on the adatom, the total den- sity of states on the graphene lattice in the presence of the adatom, and the spectral function. In Sec. IV, we present numerical results for the local densities of states, total density of states, and the spectral function for the graphene-adatom system. We also give the local den- sities of states in the presence of a finite Hubbard U within mean-field theory, and give the magnetic moment induced by the adatom, as well as the charge transfer from the adatom to the graphene, as a function of the adatom parameters. In Sec. V, we give a concluding dis- cussion. II. THE MODEL HAMILTONIAN Graphene is composed of two inter-penetrating trian- gular lattices, which we will label α and β , and thus two carbon atoms per primitive cell. In the present work, we are interested in a system consisting of a perfect lattice of graphene plus a single adatom, which we will assume has one atomic orbital. We also assume that the adatom lies at the so-called T site, above one of the carbon atoms. It has been found, using ab initio electronic structure calculations, that a several species of adatoms, including hydrogen, fluorine, and gold, do occupy a location above one of the carbon atoms[23–25]. 2 FIG. 1: (Color Online) Left: graphene crystal structure show- ing the two interpenetrating lattices labeled as α and β and the set of nearest neighbor vectors d1 , d2 , and d3 . Right: first Brillouin zone for graphene showing the high symmetry points. This figure is a modified version of one shown in Ref. [3]. We treat the graphene-adatom system using a tight- binding Hamiltonian that can, in principle, include hop- ping between any two carbon atoms. For pure graphene, the Hamiltonian can be written in terms of the creation and annihilation operators for electrons of spin σ on a site in the nth primitive cell of the α and β sublattices. We denote the creation (annihilation) operators for the α and β sublattices by a†nσ (anσ ) and b†nσ (bnσ ). The cor- responding tight-binding Hamiltonian H0 for graphene may be written in real space as tαα,δ (a†n,σ an+δ,σ + b†n,σ bn+δ,σ ). H0 = − Xn,δ,σ (tαβ ,δ a†n,σ bn+δ,σ + h. c. ) − Xn,δ 6=0,σ Here tαβ ,δ and tαα,δ (tαβ , tαα > 0) are hopping integrals which take an electron of spin σ (σ = ±1/2) from a lattice site to a neighboring lattice site, and δ denotes a Bravais lattice vector of the triangular lattice. The first sum represents hopping between sublattices and therefore all possible Bravais lattice vectors are summed over, whereas the second sum only includes hopping between sites on the same sublattice and thus, δ = 0 is not allowed. We have assumed that the hopping integrals for hopping on the same sublattice are identical for the α and β sublattices and therefore set tαα = tββ . Eq. (1) can be Fourier transformed as H0 = Xk,σ H0,k,σ = −tαβ (k)a†k,σ bk,σ − t∗αβ (k)ak,σ b†k,σ − tαα (k)[a†k,σ ak,σ + b†k,σ bk,σ ], H0,k,σ ; (1) (2) tαβ (k) = Pδ eik·δ tαβ ,δ and tαα (k) = where Pδ 6=0 eik·δ tαα,δ . In the limit of only nearest neighbor hopping, tαα,δ = 0, and tαβ ,δ = 0 except for the three nearest neighbors. In this limit, we write tαβ (k) = t(k). The three nearest neighbor vectors are shown in Fig. 1, where they are denoted by d1 , d2 , and d3 . In this limit, H0,k,σ = −t(k)a†k,σ bk,σ − t∗ (k)ak,σ b†k,σ . (3) Here the operator a†k,σ = 1√N Pn an,σ exp(ik · δn), where N is the number of primitive cells in the graphene lattice and δn is the nth Bravais lattice vector of the triangular lattice; an analogous definition holds for b†k,σ . The sum over k is confined to the first Brillouin zone and t(k) is given by [17] 2 (cid:19) cos √3ky a0 2 !# . t(k) = t "1 + 2 exp (cid:18) 3ikxa0 In Eq. (4), t is the hopping energy between nearest neigh- bor carbon atoms (t = 2.8 eV for graphene[17]), and a0 = 1.42A is the nearest-neighbor bond length[12]. (4) 0.6 0.4 0.2 ) 1 - V e ( ) E ( 0 ρ 0 -9 -6 -3 0 Energy (eV) 3 6 9 FIG. 2: (Color Online) Density of states per spin and per primitive cell ρ0 (E ) for the tight-binding model defined by Eqs. (3) and (4) for the pz orbitals of graphene, as calculated from (22)[18, 21]. Following Ref. [17], we assume t = 2.8 eV for graphene. We wish to investigate what happens to the density of states when when an isolated adatom is adsorbed onto the host graphene at a T site[23, 24]. The extra piece of the tight-binding Hamiltonian, HI , due to the adatom may be written in real space as h†0,σ h0,σ − t′ Xσ (cid:16)h†0,σ a0,σ + h0,σ a†0,σ (cid:17) . (5) HI = ǫ0 Xσ Here h†0,σ and h0,σ are creation and annihilation opera- tors for an electron of spin σ (σ = ±1/2) at the site of the adatom, which we assume is located at the site 0 of the α sublattice, ǫ0 is the on-site energy of an electron on 3 that site (relative to the Dirac point of the pure graphene band structure), and t′ , (t′ > 0), is the energy for an elec- tron to hop between the adatom and the carbon atom at the site 0 of the α sub-lattice. We now wish to express HI in terms of Bloch eigen- states of H0 . These eigenstates may be written as two- component column vectors with components ψ1 (k) and ψ2 (k) satisfying the eigenvalue equation (hereafter we suppress the spin subscript until needed) ǫk + tαα (k)(cid:19) (cid:18)ψ1,k ψ2,k(cid:19) = 0. (cid:18)ǫk + tαα (k) tαβ (k) t∗αβ (k) The solution to this eigenvalue problem is (6) ǫk = −tαα (k) ± tαβ (k), which gives the tight-binding band structure of pure graphene[17], and the corresponding eigenvectors satisfy (7) ψ1,k = ∓e−iφk ψ2,k , where the phase factor e−iφk is given by (8) e−iφk = tαβ (k) tαβ (k) We can then write the destruction operator for a Bloch electron in the upper band as (9) . (10) γk,1 = 1 √2 (cid:0)eiφk ak + bk(cid:1) , and in the lower band as 1 √2 (cid:0)eiφk ak − bk(cid:1) , where we have defined γk,1 and γk,2 to be properly normalized, so that, for example, the anticommutator {γk,1 , γ †k,1} = 1. With these definitions, we can now use the inverse Fourier transform of the γk,1 and γk,2 to obtain γk,2 = (11) an = e−ik·δn e−iφk (γk,1 + γk,2 ). 1 √2N Xk Thus, we can rewrite the impurity Hamiltonian (5) in momentum space as (12) √2N "h†0 Xk e−iφk (γk,1 + γk,2 ) + h.c.# . t′ Thus, in HI , the creation and annihilation operators of the adatom are connected to every eigenstate of the graphene band structure by matrix elements of equal magnitude (though different phase). For a hydrogen adatom, we take ǫ0 = 0.4 eV and t′ = 5.8 eV , as found in [17]. The one-electron Hamiltonian, H0 + HI , is a special case of the HI = ǫ0h†0h0 − (13) Anderson impurity model[22], where the impurity state is coupled to all the band electron states by matrix elements of equal magnitude. We also include in our calculation the effects of an on-site electron-electron interaction of the Hubbard form, 4 HU = U n0↑n0,↓ , where n0,σ = h†0,σ h0,σ is the number of electrons with spin σ on the hydrogen site. For a hydrogen adatom we take U to be the difference between the ionization potential and the electron infinity providing us with a numberical value of U ∼ 12.85eV = 4.59t[26, 27]. The Hubbard term given in Eq. (14) is quartic in the creation and annihilation operators. Therefore, in order to calculate the properties of the Hamiltonian including this term, we use a standard mean field theory to rewrite this term (see, e. g., Ref. [28]) in the form HU ∼ U hh†0↑ h0↓ hn0↑ i − hn0↑ ihn0↓ ii . h0↑ hn0↓ i + h†0↓ (15) (14) With this approximation, the total Hamiltonian, con- sisting of the sum of Eqs. (3), (13), and (15), becomes quadratic in electron creation and annihilation operators, and can be diagonalized. The Fermi energy, total en- ergy, and magnetic properties of the system can then be obtained by an iterative process as described below. The electronic density of states corresponding to the one- electron Hamiltonian, H0 + HI , can be obtained analyti- cally, as we describe below, which makes the calculation of the total energy and the magnetic properties quite simple. III. GREEN’S FUNCTION, DENSITY OF STATES, AND SPECTRAL FUNCTION OF GRAPHENE-ADATOM SYSTEM A. Green’s Function We use the single particle Green’s function approach to calculate the local and total density of states of the graphene-adatom system, initially omitting the Hubbard-U term. We continue to suppress the spin de- gree of freedom since, in the absence of the Hubbard term, spin just gives an extra factor of 2. To that end, we first introduce the resolvent operator , (16) G(z ) = 1 z − H where z = E + iη , (η → 0+), and H = H0 + HI . If there are 2N carbon atoms and 1 adatom, G(z ) can be ex- pressed as an (2N + 1) × (2N + 1)-dimensional matrix. It is convenient to use the 2N Bloch states (corresponding to N k values) created by the operators γ †k,1 and γ †k,2 as the basis for this matrix, plus the adatom orbital corre- sponding to h†0 . If we let the adatom orbital correspond to the first of the (2N + 1) states, then one can easily write out the matrix z − H of which G(z ) is the inverse. B. Density of States We denote the local electronic density of states per spin on the adatom site by ρ00 (E ). ρ00 (E ) is related to G(z ) by 1 π 1 π (17) Imh0 ρ00 (E ) = − ImG00 (z ) = − 1 z − H 0i. Here z = E + iη , (η → 0+ ), and h01/(z − H )0i denotes the matrix element of 1/(z − H ) evaluated at the location of the adatom, which we take to be above the atom 0 on the α sub-lattice. This corresponds to an element in the first row and first column of the matrix (z − H )−1 . We can obtain this matrix element as , G00 (z ) = cof 00 (z − H ) det(z − H ) where cof00 (z − H ) denotes the cofactor of the element in the first row and first column of the matrix z − H , and the denominator is the determinant of z − H . Both quantities are readily evaluated, and the result for ρ00 (E ) is (18) ρ00 (E ) = − 1 π Im 2N G0 (z ) !z=E+i0+ 1 z − ǫ0 − t′2 , (19) where G0 (z ) = 2 Xk,λ=1 (cid:18) z − ǫk,λ (cid:19) ≡ Tr (cid:18) 1 z − H0 (cid:19) 1 and ǫk,2 = −ǫk,1 is given by Eq. (7). Im G0 (z ) is related to the (unperturbed) graphene den- sity of states per graphene unit cell (per spin), which we denote ρ0 (E ), by (20) 1 π − Im G0 (E + iη) = N ρ0(E ), (21) 5 2E t2 π2 2E t2 π2 1 < x < 3, 0 < x < 1 (22) where the integral runs over the range where ρ0 (E ′ ) 6= 0. The density of states on the carbon sites (per spin) in the presence of an adatom may be written as ρg (E ) = 1 − 1 z−H k, λi, where z = E + iη , (η → 0+ ). π Im Pk,λ hk, λ Each of the elements of this sum can be computed using the analog of Eq. (18), with the result with η → 0+ [16, 22]. For the form of H0 which includes only nearest neighbor hopping, ρ0 (E ) is[18, 21] ρ0 (E ) =  f (x) (cid:19) K (cid:18) 4x 1 pf (x) K (cid:18) f (x) 4x (cid:19) 1  √4x where x = E /t, f (x) = (1 + x)2 − (x2−1)2 and K (m) is 4 the elliptic integral of the first kind. We plot Eq. (22) in Fig. (2) normalized such that R 0 −3t ρ0 (E )dE = 1. Re G0 (E ) is related to Eq. (22) via the principal value integral[16, 22] Re G0 (E ) = N P (cid:18)Z 3t dE ′(cid:19) , −3t The total density of states per spin is is the sum of the expressions in Eqs. (19) and (24), and can be rearranged to have the form t′2/2N z − ǫ0 − t′2 2N G0 (z ) dz !z=E+i0+ dG0 (24) ρ0 (E ′ ) E − E ′ ρg (E ) = N ρ0 (E )+ Im (23) 1 π . ρtot (E ) = ρg (E ) + ρ00 (E ) = N ρ0 (E ) − 1 π Im (cid:18) d dz ln[z − ǫ0 − t′2 2N G0 (z )](cid:19)z=E+i0+ . (25) C. Spectral Function We can use an analogous approach to calculate the spectral function A(k, E ). A(k, E ) represents the probability density that an electron with Bloch wave-vector k has energy E , and is given by z − H (cid:19) k, λi. hk, λ (cid:18) 1 Im Xλ These matrix elements can be evaluated using the methods of the preceding section, with the result A(k, E ) = − (26) 1 π + Im 1 π (27) A(k, E ) = − 2 2N (z − ǫk,λ )2 #z=E+i0+ Xλ=1 " 2N G0 (z ) ! t′2 1 1 1 z − ǫ0 − t′2 z − ǫk,λ It is convenient to write this spectral function in terms of a self-energy Σλ (k, E ) as Im 2 z − ǫk,λ − Σλ (k, z ) !z=E+i0+ 1 Xλ=1 where Σλ (k, E ) is readily computed by equating Eqs. (27) and (28). To first order in 1/N , Σλ (k, z ) is found to be independent of both λ and k and to take the form 2N G0 (z ) !z=E+i0+ 2N t′2 1 z − ǫ0 − t′2 All these equations [(19), (20), (21), (23)-(25), (28), and (29)] remain valid for non-nearest-neighbor hopping; only the form of the graphene density of states has to be changed. A(k, E ) = − Σλ (k, z ) = (29) (28) 1 π . , . D. Effects of Electron-Electron Interaction; Spin-Polarized Density of States and Magnetic Moment treat HU by mean-field theory [Eq. (15)], then the den- sities of states for spin-up and spin-down electrons may Finally, we discuss the effects of including a non-zero Hubbard term HU [Eq. (14)] in the Hamiltonian. If we be different. We can calculate these partial densities of states self-consistently as follows. First, we make an ini- tial assumption for the value of hn0↑ i and hn0↓ i. Then the effective on- site energy for an up-spin electron on the hydrogen adatom is obtained by making the replacement calculation). Typically, about twenty iterations of the self-consistent equations are needed to attain this degree of convergence, as discussed further below. Once EF has been found, the total magnetic moment µT of the system is obtained from 6 (30) ǫ0,↑ → ǫ0 + U hn↓ i, with a corresponding expression for ǫ0,↓ . Given ǫ0,↑ and ǫ0,↓ , we can compute the local densities of states ρ00,↑ and ρ00,↓ using the appropriate generalizations of Eq. (19); we can also obtain the total densities of states ρtot,↑ and ρtot,↓ using the corresponding generalizations of Eq. (25). The Fermi energy, EF , is then obtained from the condition (31) ρtot (E )dE , 2N + 1 = Z EF −3t where we assume one adatom, 2N carbon sites, and ρtot (E ) = ρtot,↑ (E ) + ρtot,↓ (E ). Given EF , we then recal- culate hn0,↑ i and hn0,↓ i. The procedure is repeated until successive iterations do not lead to a significant change in hn0,↑ i and hn0,↓ i. In practice, we require that these quantities change by no more than ±0.001na on succes- sive iterations (here na is the number of adatoms in the (32) µT µB [ρtot,↑ (E ) − ρtot,↓ (E )] dE , = Z EF −3t where µB is the Bohr magneton. In the limit U → ∞, the mean-field version of the Hub- bard model can be done without iteration. In this limit, only one of the quantities hn↑ i or hn↓ i is non-zero. The reason is that if, say, hn↑ i is non-zero, then the energy to put a spin-down electron on the adatom becomes infi- nite, and hence the number of spin-down electrons must be zero. To be definite, we assume that hn↓ i = 0. In that case, we just have ǫ0,↑ = ǫ0 , and ǫ0,↓ → ∞. The total density of states for the up spins will then be given by Eq. (25), while that for the down spins is just that of unperturbed graphene: ρtot,↓ = N ρ0 (E ). In the limit U → ∞ the Fermi energy, EF , is obtained from Eq. (31) and may be simplified to Z EF 0 2N ρ0 (E )dE − 1 π Im ln " EF − ǫ0 − t′2 2N G0 (−3t) # = 1. 2N G0 (EF ) −3t − ǫ0 − t′2 (33) Once EF has been obtained, the magnetic moment µT can be again found using Eq. (32). Since both ρ↑ (E ) and ρ↓ (E ) are available analytically, using Eqs. (32) and (33), µT is easily computed in closed form. IV. NUMERICAL RESULTS In Fig. 3(a), we plot the local density of states ρ00 (E ) for parameters appropriate to a hydrogen adatom on graphene with U = 0, as calculated from Eq. (19). We use the parameters t = 2.8 eV , t′ = 5.8 eV , and ǫ0 = 0.4 eV , as given by Ref. [17] for a hydrogen adatom. In Fig. 3(b), we plot the change in the total density of states produced by a single hydrogen atom, i. e., the quantity ρtot (E ) − N ρ0 (E ) for the three cases of Fig. 3(a), calculated using Eqs. (17) and (25). Next, we calculate both the spectral function A(k, E ) for U = 0 and the spin polarized spectral function Aσ (k, E ) for U = ∞ as functions of E for several val- ues of k, using the parameters of Fig. 2. Aσ (k, E ) is obtained using a generalization of Eq. (28) in the limit U → ∞ as discussed in subsection III D. The resulting spectral functions are shown in Fig. 4 through first order in 1/N . The contribution from the adatom appears as the sharp spike near EF = 0.173 eV , while the contribu- tion from the graphene sheet is shown as broadened peaks near the values of ǫk,i (i = 1, 2) for the three choices of k. The self energy term [Eq. (29)] controls the width of the graphene resonances. The integral of the graphene sheet’s contribution to the spectral function will be of order N times larger then that of the adatom. Further- more, the width of the graphene peaks in the spectral function is proportional to the density of adatoms.  15 10 5 ) 1 - V e ( ) E ( 0 0 ρ (a) 10 0 -1 0 1 150 100 50 ) 1 - V e ( ) E ( 0 ρ N - ) E ( t o t ρ 7 (b) 100 0 -1 0 1 0 -3 0 Energy (eV) 3 0 -3 0 Energy (eV) 3 FIG. 3: (Color Online) (a). Local density of states per spin on the adatom site, ρ00 (E), with U = 0 for a graphene sample with N = 500 graphene unit cells (1000 C atoms) and one adatom. We assume the model described in the text [Eqs. (3), (4), and (13)]. Black curve: t′ = 5.8 eV , ǫ0 = 0.4 eV ; blue curve: t′ = 1.0 eV , ǫ0 = 0.4 eV ; and red curve: t′ = 1.0 eV , ǫ0 = −0.4 eV . In all three cases, t = 2.8 eV . The Fermi energy is calculated using Eq. (31) and gives EF = 0.236 eV . (b). The change in the density of states per spin due to the adatom for the three cases shown in (a). In both (a) and (b), the insets are enlargements of the region between -1 eV and +1 eV. (34) ρ00,σ (E )dE , Using the mean-field methods described in section II we can calculate a variety of other spin-independent and spin-dependent properties of the adatom-graphene sys- tem. These include ρ00,↑ (E ) and ρ00,↓ (E ), the local den- sity of states of up and down spin on the adatom; the induced magnetic moment on the adatom (µa ) and in the entire system of graphene sheet plus adatom (µT ); and the net charge transfer from the adatom to the sheet, all as functions of the parameters U , ǫ0 , and t′ . The mag- netic moment on the adatom site is µa = (hn↑ i− hn↓ i)µB . hnσ i is obtained from hnσ i = Z EF −∞ where ρ00,σ (σ =↑ or ↓) is defined using the appropriate generalization of Eq. (19). The total magnetic moment is given in Eq. (32). The net charge transfer from adatom to the graphene lattice is obtained by first integrating ρ00,↑ + ρ00,↓ up to the Fermi energy, to obtain the net number of electrons on the adatom, then subtracting this quantity from the adatom valence Z (i. e., for hydrogen, Z = 1) to obtain the net charge transfer. We have carried out these calculations for various val- ues of the adatom on-site energy ǫ0 , Hubbard parame- ter U , and hopping energy t′ . In Table I we summarize the results above for parameters appropriate to a hydro- gen adatom and summarize the trends when the various adatom parameters are varied. Additional results are shown in Figs. 3 and 5. As can be seen in Table I and Fig. 5, the parameter values thought to be appropriate to an H adatom (U = 4.59t, t′ = 5.8 eV , and ǫ0 = 0.4 eV ), lead to a very small magnetic moment on the adatom though there is an increase in both the LDOS and TDOS close to the Fermi energy (EF = 0.173 eV ; see Fig. 5(b)). U Summary of Numerical Results t′ µa (µB ) µT ǫ0 Charge EF (µB ) Transfer (eV) per (eV) (eV) (e) adatom 0.0ta 0.0 0.0 0.372 5.8 0.4 4.59tb 0.4 0.695 5.8 2.67E-4 1.13E-3 0.758 5.8 3.11E-3 7.67E-3 10.0t 0.4 0.699 0.871 0.300 5.8 ∞ 0.4 0.738 0.927 0.260 ∞ 0.0 5.8 ∞ −0.4 5.8 0.338 0.958 0.662 0.639 0.990 0.360 ∞ −1.0 5.8 0.641 0.506 0.358 1.8 ∞ 0.4 ∞ 0.4 7.8 0.219 -0.11 0.780 0.372 0.173 0.236 0.111 0.050 0.050 -0.01 0.236 0.236 TABLE I: Magnetic moment on the adatom (µa ), total mag- netic moment on the graphene-adatom system (µT ) (both in units of µB ), and charge transferred from the adatom to the graphene lattice (in units of an electron charge), for various choices of U , on-site energy ǫ0 , and hopping energy t′ . Note that U = 4.59t, ǫ0 = 0.4 eV , and t′ = 5.8 eV corresponds to the expected parameters of a hydrogen adatom. When U → ∞, we find a spin polarized state near the Fermi energy. The magnetic moment calculated on the adatom is done us- ing a combination of Eqs. (15) and (19) and the magnetic moment on the sheet is calculated using Eq. (32). (a)- Spin polarized LDOS plotted in Fig. (3). (b)- Spin polarized LDOS plotted in Fig. (5b). In general, as seen in Table I, for sufficiently large U , a nonzero magnetic moment develops on both the adatom and the graphene sheet. The moment on the adatom is 8 of order 0.3 µB in this limit, for the given parameters, while the sum of the moments on the adatom and the sheet approaches µB in this limit. We also find in all of our calculations that a large fraction (typically 0.6 − 0.7 of the electron) is transferred from the adatom to the graphene sheet for the parameters we consider. For the parameters appropriate to hydrogen adatoms, the model predicts no, or only a very small, induced magnetic mo- ment. A possible explanation is that our model assumes no lattice distortion due to the adatom. But DFT calcu- lations have shown that the surface of graphene is warped due to the addition of an adatom. This warping could change the distance between the adatom and the neigh- boring carbon atoms, and hence possibly the value of the Coulomb integral. In Fig. 6, we show the total magnetic moment of the system as given by Eq. (32) plotted as a function of U , for various values of t′ . In each case shown, ǫ0 = 0.4 eV and t = 2.8 eV . In all the plots, there is an apparent threshold behavior: the moment becomes nonzero only if U exceeds a threshold value which depends on t′ as well as on U . While these calculations are done using a sim- ple mean-field approximation, they seem to be consistent with other work on related models[22, 32]. V. DISCUSSION Using a tight-binding model we have calculated the local and total density of states and the spectral func- tion for a system consisting of a single adatom in a T site on graphene. Because the hopping integral from the adatom to a graphene Bloch state has the same magni- tude for any k, we have shown that these quantities can be calculated analytically. This simplification holds even if we do not make the oft-used linear approximation[16] for the graphene density of states near the Dirac point. It is also valid even if we include non-nearest-neighbor hop- ping in the tight-binding graphene Hamiltonian. Since our numerical results give both the local and total den- sity of states, we can compute the charge transfer from the adatom to the graphene. Our numerical results show that, for most parameters we consider, this charge trans- fer is a substantial fraction of an electron (approximately 70% for parameters appropriate to hydrogen). Because the calculations are at low adatom concen- tration, the adatom-induced density of states is lin- ear in concentration. Other work has treated the same system at higher adatom concentration, but only numerically[17]. In future work, it might be possible to treat the present model analytically at higher concen- trations, at least approximately. It would also be of in- terest to include effects of lattice distortions, which are known to exist when adatoms bind to graphene [29], and which can lead to a large increase in spin-orbit interac- tions [30, 31]. Such spin-orbit interactions would likely have a large effect on the magnetic properties arising from the adatom. (a) M (b) Γ (c) K (d) M 8.0 4.0 0 8.0 4.0 0 8.0 4.0 0 8.0 4.0 ) V e / 0 0 0 1 ( ) E , k ( A ) V e / 0 0 0 1 ( ) E , k ( A ) V e / 0 0 0 1 ( ) E , k ( A ) V e / 0 0 0 1 ( ) E , k ( σ A 0 -9 -6 0 -3 Energy (eV) 3 6 9 FIG. 4: (Color Online) (a)-(c). Spectral function A(k, E ) at U = 0 for the graphene-adatom system, as calculated us- ing Eqs. (28) and (29) for three values of k corresponding to M , K , and Γ respectively, assuming a nearest-neighbor tight- binding band. We use t = 2.8 eV , t′ = 5.8 eV , ǫ0 = 0.4 eV , and N = 500. (d). The spin polarized spectral function for the point M and the case U → ∞ in the mean-field approxi- mation, using the same parameters as in (a)-(c). Aσ (k, E ) for the ma jority spin component is shown in black and that of the minority component in red. For the minority spin compo- nent, Aσ (k, E ) is just that of the unperturbed graphene sheet, i. e., delta functions at the unperturbed pure graphene energy eigenvalues for all values of k. For the case U = 0, and for the ma jority spin at U = ∞, the adatom contribution occurs near EF , while the graphene sheet contribution corresponds to broadened peaks near the unperturbed energy eigenvalues ǫk given in Eq. (7). For points Γ and M , the integral of the peak near E ∼ 0 is of order 1/(2N) times those of the main peaks in the spectral function, and thus vanishes as N → ∞. ) 1 - V e ( n w o d , 0 0 ρ , p u , 0 0 ρ ) 1 - V e ( n w o d , 0 0 ρ , p u , 0 0 ρ 1 0.8 0.6 0.4 0.2 0 0.8 0.6 0.4 0.2 0 (a) (b) (c) 9 (d) (e) (f) (g) (h) -9 -6 -3 0 3 Energy (eV) 6 9 -6 -3 0 3 Energy (eV) 6 9 -6 -3 0 3 Energy (eV) 6 9 -6 -3 0 3 Energy (eV) 6 9 FIG. 5: (Color Online) Spin-polarized local density of states (sLDOS) per adatom for both the ma jority spin (black solid line) and minority spin (red solid line), as obtained by substituting Eq. (30) into Eq. (19). We use t = 2.8 eV and the following values of the on-site energy ǫ0 , hopping energy t′ , and Hubbard energy U : (a). (t′ , ǫ0 , U ) = (5.8, 0.4, 4.59t); (b). (5.8, 0.4, 10t); (c)(5.8, 0.4, ∞) (d). (5.8, 0.0, ∞); (e). (5.8, −0.4, ∞); (f ). (5.8, −1.0, ∞); (g). (1.8, 0.4, ∞); (h). (7.8, 0.4, ∞). We have also calculated the magnetic properties in- duced by the adatom, using a Hubbard model treated within mean-field theory. For all choices of the Hamilto- nian parameters, we find that there is a critical value of the Hubbard U above which the density of states near the Fermi energy is spin-polarized and a net induced mag- netic moment is formed. The appearance of this mag- netic moment was predicted long ago to occur within mean-field theory for models with a slowly varying den- sity of states near the Fermi energy[22]. Here, it is also found to occur in a system with a roughly linear density of states near EF . While the mean-field approximation is probably unreliable for this model, we note that a similar threshold for moment formation was also found, within the Kondo Hamiltonian, for a system with a linear den- sity of states[32]. Since, at large U , it is known that the Hubbard model can be approximately transformed into the Kondo model[33], it seems plausible that there could really be a threshold behavior in the Hubbard model with a linear density of state such as is found here, even though we use a mean field theory to obtain it. A somewhat counterintuitive result of our calculations is that, as t′ increases, the value of U needed to induce a magnetic moment becomes smaller. Since a larger t′ suggests that it is easier for the electron to hop from the impurity to the graphene, one might expect that a mo- ment on the impurity atom would be less likely to form. A possible explanation is that the larger t′ also causes the peak in the impurity density of states to shift closer to the Dirac point, where the graphene density of states is smaller. Thus, there are fewer final states available for an electron to hop into, and hence, the electron is less likely to hop, thus increasing the likelihood of moment formation on the impurity. In our approximation we also calculate the spectral function of our graphene-adatom system to first order in 1/N . The main effect of the adatom is, as expected, sim- ply to broaden the delta-function peaks that the spectral function would exhibit in an ideal graphene lattice. In our approach, this broadening, and the shape of the spec- tral line, are computed analytically. To the same order, we find that both the adatom contribution to the spectral function at E ∼ 0 and the broadening of the graphene spectral lines will vanish as N → ∞.  1 0.8 0.6 0.4 0.2 ) B µ ( T µ t’ = 2.0t t’ = 1.0t 10 In summary, we have, using a single-particle Green’s function approach together with a tight binding Hamil- tonian in the limit of no electron-electron correlations, obtained analytical equations for the LDOS, TDOS, and spectral function for adatoms on the surface of graphene. Using the same model with a finite Hubbard energy U , we find that a magnetic moment is induced both on the adatom and nearby on the graphene sheet above a crit- ical value of U which depends on the other model pa- rameters. These results are not only of intrinsic interest but also may be useful in understanding the behavior of a variety of adatoms on graphene. 0 1 10 100 1000 U (eV) 10000 100000 FIG. 6: Total magnetic moment on the adatom and the graphene (µT ), in units of µB , versus U for two values of t′/t (Red curve: t′ = 1.0t, Black curve t′ = 2.0t). In this plot, we hold ǫ0 = 0.4eV and t = 2.8eV as we vary U for two different choices of t′ . In both cases, the moment seems to become nonzero at a characteristic value of U , which depends on t′ , t, and ǫ0 . Lines connect calculated points. VI. ACKNOWLEDGMENTS This work was supported by the Center for Emerging Materials at The Ohio State University, an NSF MR- SEC (Grant No. DMR0820414) . We thank Prof. Roland Kawakami for valuable discussions. [1] P. R. Wallace, Phys. Rev. 71, 662 (1947). [2] K. S. Novoselov, A. K. Geim, S. V. Morozov, D. Jiang, Y. Zhang, S. V. Dubonos, I. V. Grigorieva, and A. A. Firsov, Science 306, 666 (2004). [3] A. H. Castro Neto, F. Guinea, N. M. R. Peres, K. S. Novoselov, and A. K. Geim, Rev. Mod. Phys. 81 109 (2009). [4] R. Saito, G. Dresselhaus, and M. S. Dresselhaus, Phys- ical Properties of Carbon Nanotubes (Imperial College Press, London, 1998). [5] X. Wang, L. Zhi, and K. Mullen, Nano Letters 8, 323 (2008). [6] W. Liu, T. Dang, Z. Xiao, Z. Li, C. Zhu, and X. Wang, Carbon. 49, 884 (2011). [7] M. D. Stoller, S. Park, Y. Zhu, J. An, and R. Ruoff, Nano Letters 8, 3498 (2008). [8] O. V. Yazyev and M. I. Katsnelson, Phys. Rev. Lett. 100, 047209 (2008). [9] N. Tombros, C. Jozsa, M. Popinciuc, H. T. Jonkman, and B. J. van Wees, Nature 448, 06037 (2007). [10] K. M. McCreary, A. G. Swartz, W. Han, J. Fabian, and R. K. Kawakami, Phys. Rev. Lett. 109, 186604 (2012). [11] P. O. Lehtinen, A. S. Foster, Yuchen Ma, A. V. Krasheninnikov, and R. M. Nieminen. Phys. Rev. Lett. 93, 187202 (2004). [12] O. V. Yazyev and L. Helm, Phys. Rev. B 75, 125408 (2007). [13] Y. V. Skrypnyk and V. M. Loktev, Phys. Rev. B 73, 241402(R) (2006). [14] J. O. Sofo, G. Usa j, P. S. Cornaglia, A. M. Suarez, A. D. Hern´andez-Nieves, and C. A. Balseiro. Phys. Rev. B 85, 115405 (2012). [15] V. M. Pereira, J. M. B. Lopes dos Santos, and A. H. Cas- tro Neto, Phys. Rev. B 77, 115109 (2008). [16] T. O. Wehling, M. I. Katsnelson, and A. I. Lichtenstein, Chem. Phys. Lett. 476 125 (2009). [17] A. L. Rakhmanov, A. V. Rozhkov, A. O. Sboychakov, and F. Nori. Phys. Rev. B 85, 035408 (2012). [18] S. Yuan, H. De Raedt, and M. I. Katsnelson, Phys. Rev. B 82, 115448 (2010). [19] T. O. Wehling, A. V. Balatsky, M. I. Katsnelson, A. I. Lichtenstein, K. Scharnberg, and R. Wiesendanger, Phys. Rev. B75, 125425 (2007). [20] C. Bena and G. Montambauz, New J. Phys. 11, 0950003 (2009). [21] J. Hobson and W. A. Nierenberg, Phys. Rev. 86, 662 (1953). [22] P. W. Anderson, Phys. Rev. 124, 41 (1961). [23] K. Nakada and A. Ishii, Solid State Commun. 151, 13 (2011). [24] K. T. Chan, J. B. Neaton, and M. L. Cohen, Phys. Rev. B 77, 235430 (2008). [25] A. Ishii, M. Yamamoto, H. Asano, and K. Fujiwara. J. Phys. Conf. Ser. 100, 052087 (2008). [26] R. Pariser and R. Parr, J. Chem. Phys. 21, 767-775 (1953). [27] K. R. Lykke, K. K. Murray, and W. C. Lineberger, Phys. Rev. A 43, 6104-6107 (1991). [28] P. Fazekas, Lecture Notes on Electron Correlation and Magnetism, (World Scientific, Singapore, 1999). [29] D. W. Boukhvalov, M. I. Katsnelson, and A. I. Lichten- stein, Phys. Rev. B 77, 035427 (2008). [30] A. H. Castro Neto and F. Guinea. Phys. Rev. Lett. 103, 026804 (2009). [31] Martin Gmitra, Denis Kochan, and Jaroslav Fabian, Phys. Rev. Lett. 110 246602 (2013). [32] D. Withoff and E. Fradkin, Phys. Rev. Lett. 64, 1835 (1990). [33] J. R. Schrieffer and P. A. Wolff, Phys. Rev. 149, 491 (1966). 11 

1302.3782

1

1302

Super-Planckian Near-Field Thermal Emission with Phonon-Polaritonic Hyperbolic Metamaterials

We study super-Planckian near-field heat exchanges for multilayer hyperbolic metamaterials using exact S-matrix calculations. We investigate heat exchanges between two multilayer hyperbolic metamaterial structures. We show that the super- Planckian emission of such metamaterials can either come from the presence of surface phonon-polaritons modes or from a continuum of hyperbolic modes depending on the choice of composite materials as well as the structural configuration.

cond-mat.mes-hall

cond-mat

Super-Planckian Near-Field Thermal Emission with Phonon-Polaritonic Hyperbolic Metamaterials S.-A. Biehs and M. Tschikin Institut fur Physik, Carl von Ossietzky Universitat, D-26111 Oldenburg, Germany. R. Messina and P. Ben-Abdallah Laboratoire Charles Fabry, UMR 8501, Institut d'Optique, CNRS, Universit´e Paris-Sud 11, 2, Avenue Augustin Fresnel, 91127 Palaiseau Cedex, France. (Dated: June 29, 2018) We study super-Planckian near-field heat exchanges for multilayer hyperbolic metamaterials using exact S-matrix calculations. We investigate heat exchanges between two multilayer hyperbolic metamaterial structures. We show that the super- Planckian emission of such metamaterials can either come from the presence of sur- face phonon-polaritons modes or from a continuum of hyperbolic modes depending on the choice of composite materials as well as the structural configuration. In the last few years several fascinating experiments have demonstrated that for small separation distances compared with the thermal wavelength the thermal radiation exchanged between two hot bodies out of thermal equilibrium increases dramatically compared with what we observe at large distances and can even exceed the well-known Stefan-Boltzmann law by orders of magnitude1 -- 6. Accordingly, thermal emission is in that case also called super-Planckian emission emphasizing the possibility to go beyond the classical black-body theory. There are several promising applications of super-Planckian emitters ranging from thermal imaging7 -- 9 and thermal rectification/management10 -- 12 to near-field thermophoto- voltaics13 -- 17. This has triggered many studies on the possibilities of tailoring and controlling the super-Planckian radiation spectrum by means of designing the material properties18 -- 23, using phase-change materials24 or 2D systems such as graphene25,26, for instance. Recently, it was shown that hyperbolic metamaterials can lead to broad-band photonic thermal conductance inside the material itself27 and between two hyperbolic materials only separated by a vacuum gap28. Further Nefedov et al. considered nanorod-like structures 3 1 0 2 b e F 5 1 ] l l a h - s e m . t a m - d n o c [ 1 v 2 8 7 3 . 2 0 3 1 : v i X r a 2 made of nanotubes which are interlocked and highlighted a giant radiative heat flux which could be utilized for near-field thermophotovoltaics energy conversion29. Finally, Guo et al. have studied the energy density produced by the thermally fluctuating fields close to a hyperbolic structure and found a broadband near-field contribution from which they have concluded that super-Planckian emission will be broad-band for hyperbolic materials30. This is in accordance with the findings in Ref.28 for the energy exchange between two hyperbolic nanowire structures. The aim of this letter is to show that the surface modes supported by the topmost layers of phonon-polaritonic metamaterials can give the dominant contribution to the super-Planckian emission. As was shown in Ref.31 materials which have a broad hyperbolic frequency band as predicted from effective medium theory can support surface modes inside these frequency bands as well which will compete with the hyperbolic modes31,32. In particular, we will show that for the realization of a hyperbolic metamaterial as studied in Ref.30 the main contribution to super-Planckian radiation is not necessarily due to hyperbolic modes but can be due to surface modes depending on the choice of the topmost layer. We will show that in order to allow for broad-band super-Planckian emission by hyperbolic modes, mainly, it is important to use a material for that topmost layer which does not support surface modes in the thermal freqeuency range. Before we start to study the super-Planckian thermal radiation let us first recall the concept of indefinite or hyperbolic materials33 -- 35. Such materials are first of all a special class of uni-axial anisotropic materials. For uni-axial materials the permittivity ǫk parallel to the optical axis is different from the permittivity ǫ⊥ perpendicular to the optical axis. For hyperbolic materials one can find frequency bands where ǫk and ǫ⊥ have different signs, i.e. ǫ⊥ǫk < 0. Thus the dispersion relation of the photons in such a material36 κ2 ǫ⊥ + k2 z ǫk = ω2 c2 (1) describes a hyperbolic function rather than an ellipse as for usual anisotropic materials; here κ (kz) is the wavevector inside the hyperbolic medium perpendicular (parallel) to the optical axis which is assumed to point in z-direction. Such metamaterials can for example be designed by multilayer structures, since in the long wavelength regime such structures can be described as homogeneous anisotropic media with the effective permittivities ǫ⊥ = ǫ1f + ǫ2(1 − f ), ǫk = 1 f /ǫ1 + (1 − f )/ǫ2 , 3 (2) (3) where ǫ1 and ǫ2 are the permittivities of the two layer-materials and f is the filling fraction of the topmost material 1, i.e. f = l1/(l1 + l2). The effective permittivities allow for a calculation of the hyperbolic frequency bands of the multilayer structure where ǫ⊥ǫk < 0. There are in general two different kinds of bands: a frequency band ∆1 where ǫk < 0 and ǫ⊥ > 0 and a frequency band ∆2 where ǫk > 0 and ǫ⊥ < 0. As will become clear in the following the such calculated frequency bands ∆1 and ∆2 can also support surface modes, which are not taken into account in the effective description31. Figure 1: Sketch of the geometry of two hyperbolic multilayer materials separated by a vacuum gap. In order to study super-Planckian radiation we consider the geometry depicted in Fig. 1. The heat transfer coefficient h(d) between the two metamaterials which are assumed to be at local thermal equilibrium can be determined by37 h(d) = Z ∞ dω 2π 0 f (ω, T ) Xj=s,p (2π)2 Tj(ω, κ; d) = Z ∞ Z d2κ dω 2π 0 f (ω, T )H(ω, d) (4) where f (ω, T ) = (ω)2/(kBT 2)eω/kBT /(eω/kBT − 1)2. Ts(ω, κ; d) and Tp(ω, κ; d) are the energy transmission coefficients for the s- and p-polarized modes which can be easily de- termined for semi-infinite materials, anisotropic materials and multilayer structures19,38 -- 47. Here we use the standard S-matrix approach as in Refs.42,45 to calculate the amplitude re- flection coefficients rj of our multilayer structures from which we can easily determine the energy transmission coefficients Tj(ω, κ; d) =   (1 − rj2)2/Dj2, 4[Im(rj)]2e−2kz0d/Dj2, κ > ω/c 4 (5) κ < ω/c including the contributions of the propagating modes with κ < ω/c and the evanescent modes with κ > ω/c. Here Dj = 1 − rjrje2ikz0d is a Fabry-P´erot-like denominator with k2 z0 = k2 0 − κ2; k0 = ω/c. Now, let us consider a concrete example of a hyperbolic structure which is composed by layers of polar materials. Because these structures can support surface phonon-polaritons as well, they are also called phonon-polaritonic hyperbolic structures. We choose to consider the structure in Ref.30 which is made of layers of SiC and SiO2. In general amorphous SiO2 supports surface modes in the infrared as well as SiC, but to get results which are comparable with the calculations done in Ref.30 we assume that ǫSiO2 = 3.9 adding a vanishingly small absorption. The optical properties of SiC are taken from Ref.48. The layer thicknesses are (a) l1 = 50 nm for the SiC layers and l2 = 150 nm for the silica layers so that the filling fraction is f = 0.25 and (b) l1 = l2 = 100 nm so that f = 0.5. For our exact S-matrix calculations we use N = 50 where the last layer is a semi-infinite layer with the material properties of the topmost layer. The hyperbolic frequency bands calculated from Eqs. (2) and (3) are (a) ∆1 = 1.495 − 1.623 · 1014 rad/s and ∆2 = 1.778 − 1.826 · 1014 rad/s and (b) ∆1 = 1.495 − 1.712 · 1014 rad/s and ∆2 = 1.712 − 1.827 · 1014 rad/s. In order to see the structure of contributing modes we have plotted the transmission coefficient Tp(ω, κ; d) in Fig. 2. The horizontal dashed white lines mark the hyperbolic bands as determined from effective medium theory36. The solid white lines are the borders of the Bloch bands as determined from Bloch mode dispersion relation for p polarization36 cos(kz,B(l1 + l2)) = − 1 2(cid:16)ǫ2kz1 ǫ1kz2 + ǫ1kz2 ǫ2kz1(cid:17) sin(kz1l1) sin(kz2l2) (6) with the permittivities ǫi (i = 1, 2) of the two layer materials and the wavevector along + cos(kz1l1) cos(kz2l2), the optical axis in z direction kzi = pk2 0ǫi − κ2. Note that kz,B is the Bloch wavevector inside the multilayer structure which can be approximated by its homogenized version kz in Eq. (1) together with Eqs. (2) and (3) in the regime where the effective description is valid. Only inside these Bloch bands one can find modes which are propagating modes inside the 5 (a) 1.9 ) s / d a r 4 1 0 1 ( ω 1.65 1.4 0 (b) 1.9 ) s / d a r 4 1 0 1 ( ω 1.65 1.4 0 ∆2 ∆1 25 50 κ/k0 75 100 ∆2 ∆1 25 50 κ/k0 75 100 1 0.8 0.6 0.4 0.2 0 1 0.8 0.6 0.4 0.2 0 Figure 2: Transmission coefficient Tp(ω, κ; d) from Eq. (5) for both SiC-SiO2 multilayer structures (a) l1 = 50 nm and l2 = 150 nm, and (b) l1 = l2 = 100 nm for the interplate distance d = 100 nm. hyperbolic material. It can be seen that there are also very dominant modes outside the Bloch bands contributing significantly to the energy transmission. These modes are the coupled surface modes of the topmost SiC layers of each hyperbolic material which means that they are evanescent modes inside and outside the hyperbolic structure. The respective contribution of the modes inside and outside the Bloch bands to the spectral heat transfer coefficient H(ω, d) is plotted in Fig. 3 for a distance of d = 100 nm. From that figure it becomes apparent that within the hyperbolic frequency bands one has quite large contributions steming from modes outside the Bloch bands which are mainly the coupled surface modes of the topmost layers. Hence, for the chosen structure the broadband super-Planckian radiation from the hyperbolic frequency band is not due to hyperbolic modes only. The relative contribution of surface modes and all the other modes is plotted in Fig. 4 where we show the heat transfer coefficient as a function of distance. From that figure it becomes obvious that for distances about 100 nm and smaller the heat flux is dominated solely by the coupled surface modes of the topmost layers showing a typical 1/d2 dependence37,49. Whereas for larger distances the heat flux is dominated by the contributions 6 inside the Bloch bands. These contributions are on the one hand hyperbolic modes steming from frequencies inside the hyperbolic bands ∆1 and ∆2. On the other, for frequencies outside the frequency bands ∆1 and ∆2 the modes are usual propagating or frustrated total internal reflection modes. Note that for distances of the order of max(l1, l2)/π the Bloch- mode contribution reaches a maximum. This can be attributed to the large wavevector cutoff by the edge of the Bloch bands which can be understood as the inset of nonlocal effects since for such distances the main wavevector contributions to the thermal emission are of the order π/d. To quantify the heat flux mediated by the hyperbolic modes we plot in Fig. 5 the different contributions of the modes inside and outside the Bloch bands, and the contribution from the hyperbolic modes separately. The separate contributions hB(d), hN B(d), and hhm(d) to the heat transfer coefficient are normalized to the total heat transfer coefficient htot(d) = hB(d)+hN B(d). Apparently, in both configurations the contribution of the hyperbolic modes is for all chosen distances smaller than 35%. This is a rather small value for a hyperbolic structure which is constructed for the purpose of enhancing the thermal radiation by the hyperbolic-mode contribution. 1500 1000 500 ) ω ( B B H / ) d , ω ( H total Bloch Non-Bloch ∆ 1 ∆ 2→ total Bloch Non-Bloch (a) 2000 1500 1000 500 ) ω ( B B H / ) d , ω ( H ∆ 1 ∆ 2 (b) 0 1.4 1.5 1.7 1.6 ω (1014 rad/s) 1.8 1.9 0 1.4 1.5 1.7 1.6 ω (1014 rad/s) 1.8 1.9 Figure 3: Spectral heat transfer coefficient H(ω, d) defined in Eq. (4) normalized to the black-body result HBB(ω) = ω2/(2πc2) for both SiC-SiO2 multilayer structures (a) l1 = 50 nm and l2 = 150 nm, and (b) l1 = l2 = 100 nm for the interplate distance d = 100 nm. Here we choose T = 300 K. The vertical dashed lines mark the borders of the hyperbolic frequency bands ∆1 and ∆2. Now, let us see if the dominant surface-mode contribution vanishes when choosing the passive SiO2-layer as the topmost layer. Here, it is important to keep in mind that SiO2 supports surface modes in the infrared. We assume here that it can be described by a (a) 1000 B B h / ) d ( h 100 10 1 0.1 10 (b) 1000 total Bloch Non-Bloch 100 1000 10000 d (nm) B B h / ) d ( h 100 10 1 0.1 10 7 total Bloch Non-Bloch 100 1000 10000 d (nm) Figure 4: Heat transfer coefficient h(d) from Eq. (4) as a function of interplatedistance d using T = 300 K for both SiC-SiO2 multilayer structures (a) l1 = 50 nm and l2 = 150 nm, and (b) l1 = l2 = 100 nm. The heat transfer coefficient is normalized to the black-body value hBB = 6.1 Wm−2K−1. The contributions from the Bloch bands and from regions outside the Bloch bands are shown separately. constant permittivity in the frequency band of interest (it is in this sense "passive") in order to compare our results to existing results in the literature. Hence, we repeat the same calculations for the same structure as before but with the difference that for both hyperbolic structures the topmost layer is SiO2 followed by SiC, etc. The results for the spectral heat transfer coefficient are shown in Fig. 6 (a). There is still a surface-mode contribution, but it is very small compared to the Bloch-mode contributions. Finally, from the distance dependent results in Fig. 6 (b) and (c) it can be seen that the super-Planckian radiation is mainly due to Bloch modes, i.e. frustrated total internal reflection modes and hyperbolic modes. In particular, the contribution of the hyperbolic modes can be larger than 50% in the strong near-field regime for distances of about 10 nm. As is obvious from Fig. 6 (b) the overall heat flux is only one order of magnitude larger than that of a black body so that the hyperbolic material considered here and in Ref.30 is a poor near-field emitter compared to the previous structures with SiC as topmost layer. But there is a simple method for increasing the hyperbolic contribution by just making the thickness of the layers smaller. Then, the border of the Bloch bands will shift to larger wavevectors which results in a broadband contribution to the transmission coefficient for larger wavevectors and hence to a larger thermal radiation. In Fig. 7 we show the heat (a) 1 ) d ( t o t h / ) d ( i h 0.8 0.6 0.4 0.2 0 10 Bloch Non-Bloch Hyperbolic 100 1000 10000 d (nm) (b) 1 ) d ( t o t h / ) d ( i h 0.8 0.6 0.4 0.2 0 10 8 Bloch Non-Bloch Hyperbolic 100 1000 10000 d (nm) Figure 5: Heat transfer coefficients hB, hN B, and hhm of the Bloch modes, the modes outside the Bloch bands and the hyperbolic modes normalized to the total heat transfer coefficient htot(d) = hB(d)+hN B(d) as a function of distance. Again we show both cases (a) l1 = 50 nm and l2 = 150 nm, and (b) l1 = l2 = 100 nm, and set T = 300 K. flux for hyperbolic structures with a filling factor of 0.5 but layer thicknesses of 100 nm, 50 nm and 5 nm. It can be seen that the heat flux increases by orders of magnitude in the strong near-field regimes, i.e., for distances smaller than 100 nm, when making the layers thinner. We have checked that the main contribution is due to hyperbolic modes in that regime (not shown here). Further studies have to find an optimized design and optimal composite materials in order to further improve the thermal radiation properties of hyperbolic materials to attain thermal heat fluxes wich are as large as the heat flux by surface modes or even larger. Note that in Ref.28 such a structure was proposed on the basis of an effective description. In conclusion, we have studied the super-Planckian emission of hyperbolic structures by using the fluctuational electrodynamics theory combined with the S-matrix method. It has been shown that to properly describe the energy exchanges it is of crucial importance not only to choose a good combination of material composites for having broad-band super- Planckian radiation but also to use a passive material as topmost layer, i.e. a material which does not support surface mode resonances within the thermally accessible spectrum. Also, we have shown for multilayer structures that the thickness of layers determines the wavevector cutoff of the Bloch band so that it appears clearly advantageous to use thin layers with elementary thicknesses l1, l2 ≪ d to observe a large super-Planckian emission at a given distance d from the surface. These findings provide the basis for realizing an optimized design of hyperbolic thermal emitters with broad-band super-Planckian spectra. 9 (a) ) ω ( B B H / ) d , ω ( H 70 60 50 40 30 20 10 0 1.4 (b) 100 B B h / ) d ( h 10 1 0.1 10 (c) 1 ) d ( t o t h / ) d ( i h 0.8 0.6 0.4 0.2 0 10 ∆ 2 total Bloch Non-Bloch ∆ 1 1.5 1.6 1.7 1.8 1.9 ω (1014 rad/s) total Bloch Non-Bloch 100 1000 10000 d (nm) Bloch Non-Bloch Hyperbolic 100 1000 10000 d (nm) Figure 6: (a) Spectral heat transfer coefficients H(ω, d) between two hyperbolic materials with SiO2 as topmost layer choosing d = 100 nm. The vertical dashed lines mark the borders of the hyperbolic frequency bands ∆1 and ∆2. (b) Heat transfer coefficients h(d) for the same materials setting T = 300 K normalized to the black-body value hBB = 6.1 Wm−2K−1. Finally in (c) we plot the relative contributions of the Bloch modes, Non-Bloch modes and the hyperbolic modes. 10 1000 100 10 1 B B h / ) d ( h 0.1 10 5 nm 50 nm 100 nm 100 1000 10000 d (nm) Figure 7: The heat transfer coefficient for the structure with the passive material as topmost layer for different layer thicknesses l1 = l2 (f = 0.5) of 100 nm, 50 nm, and 5 nm normalized to the black-body value hBB = 6.1 Wm−2K−1. Acknowledgments M.T. gratefully acknowledges support from the Stiftung der Metallindustrie im Nord- Westen. The authors acknowledge financial support by the DAAD and Partenariat Hubert Curien Procope Program (project 55923991). This work has been partially supported by the Agence Nationale de la Recherche through the Source-TPV project ANR 2010 BLANC 0928 01. 1 L. Hu, A. Narayanaswamy, X. Chen, and G. Chen, Appl. Phys. Lett. 92, 133106 (2008). 2 A. Narayanaswamy, S. Shen, and G. Chen, Phys. Rev. B 78, 115303 (2008). 3 S. Shen, A. Narayanaswamy, and G. Chen, Nano Lett. 9, 2909 (2009). 4 E. Rousseau, A. Siria, G. Jourdan, S. Volz, F. Comin, J. Chevrier and J.-J. Greffet, Nature Photonics 3, 514 (2009). 5 R. S. Ottens, V. Quetschke, S. Wise, A. A. Alemi, R. Lundock, G. Mueller, D. H. Reitze, D. B. Tanner,and B. F. Whiting, Phys. Rev. Lett. 107, 014301 (2011). 6 T. Kralik, P. Hanzelka, M. Zobac, V. Musilova, T. Fort, and M. Horak, Phys. Rev. Lett. 109, 224302 (2012). 7 Y. De Wilde, F. Formanek, R. Carminati, B. Gralak, P.A. Lemoine, K. Joulain, J.P. Mulet, Y. Chen, and J.J. Greffet, Nature 444, 740 (2006). 11 8 A. Kittel , U. Wischnath , J. Welker , O. Huth , F. Ruting, and S.-A. Biehs, Appl. Phys. Lett. 93, 193109 (2008). 9 F. Huth, M. Schnell, J. Wittborn, N. Ocelic and R. Hillenbrand, Nature Materials 10, 352 (2011). 10 C. R. Otey, W. T. Lau, and S. Fan, Phys. Rev. Lett. 104 154301 (2010). 11 S. Basu and M. Francoeur, Appl. Phys. Lett. 98, 113106 (2011). 12 S.-A. Biehs, F. S. S. Rosa, and P. Ben-Abdallah, Appl. Phys. Lett. 98, 243102 (2011). 13 R. S. DiMatteo, P. Greiff, S. L. Finberg, K. A. Young-Waithe, H. K. Choy, M. M. Masaki, and C. G. Fonstad, Appl. Phys. Lett. 79, 1894 (2001). 14 A. Narayanaswamy and G. Chen, Appl. Phys. Lett. 82, 3544 (2003). 15 M. Laroche, R. Carminati, and J.-J. Greffet, J. Appl. Phys. 100, 063704 (2006). 16 K. Park, S. Basu, W. P. King, and Z. M. Zhang, J. Quant. Spect. Rad. Transf. 109, 305 (2008). 17 S. Basu, Z. M. Zhang, and C. J. Fu, International Journal of Energy Research 33, 1203 (2009). 18 C.J. Fu and Z. M. Zhang, Int. J. Heat Mass Transfer 49, 1703 (2006). 19 S.-A. Biehs, P. Ben-Abdallah, F. S. S. Rosa, K. Joulain, and J.-J. Greffet, Opt. Expr. 19, A1088 (2011). 20 K. Joulain, J. Drevillon, and P. Ben-Abdallah, Phys. Rev. B 81, 165119 (2010). 21 R. Gu´erout, J. Lussange, F. S. S. Rosa, J.-P. Hugonin, D. A. R. Dalvit, J.-J. Greffet, A. Lambrecht, and S. Reynaud, Phys. Rev. B 85 (R), 180301 (2012). 22 J. Lussange, R. Gu´erout, F. S. S. Rosa, J.-J. Greffet, A. Lambrecht, and S. Reynaud, Phys. Rev. B 86, 085432 (2012). 23 L. Cui, Y. Huang, J. Wang, and K.-Y. Zhu, Appl. Phys. Lett. 102, 053106 (2013). 24 P. J. van Zwol, K. Joulain, P. Ben-Abdallah, and J. Chevrier, Phys. Rev. B 84, 161413 (2011). 25 V. B. Svetovoy, P. J. van Zwol, and J. Chevrier, Phys. Rev. B 85, 155418 (2012). 26 O. Ilic, M. Jablan, J. D. Joannopoulos, I. Celanovic, H. Buljan, and M. Solja¸ci´c, Phys. Rev. B 85, 155422 (2012). 27 E. E. Narimanov and I. I. Smolyaninov, arXiv:1109.5444v1. 28 S.-A. Biehs, M. Tschikin, and P. Ben-Abdallah, Phys. Rev. Lett. 109, 104301 (2012). 29 I. S. Nefedov and C. R. Simovski, Phys. Rev. B 84, 195459 (2011). 30 Y. Guo, C. L. Cortes, S. Molesky, and Z. Jacob, Appl. Phys. Lett. 101, 131106 (2012). 31 M. Tschikin, S.-A. Biehs, R. Messina, and P. Ben-Abdallah, submitted (2012). 12 32 G. Rosenblatt and M. Orenstein, Opt. Exp. 19, 20372 (2011). 33 D. R. Smith, Willie J. Padilla, D. C. Vier, S. C. Nemat-Nasser and S. Schultz, Phys. Rev. Lett. 84, 4184 (2000). 34 D. R. Smith and D. Schurig, Phys. Rev. Lett. 90, 077405 (2003). 35 L. Hu and S. T. Chui, Phys. Rev. B 66, 085108 (2002). 36 P. Yeh, Optical Waves in Layered Media, (Wiley, Hoboken, 2005). 37 K. Joulain, J.-P. Mulet, F. Marquier, R. Carminati, and J.-J. Greffet, Surf. Sci. Rep. 57, 59-112 (2005). 38 S.-A. Biehs, Eur. Phys. J. B 58, 423 (2007). 39 M. Francoeur, P. Mengu¸c, and R. Vaillon, Appl. Phys. Lett. 93, 043109 (2008). 40 W. T. Lau, J.-T. Shen, G. Veronis, S. Fan, and P. V. Braun, Appl. Phys. Lett. 92, 103106 (2008); W. T. Lau, J.-T. Shen, and S. Fan, Phys. Rev. B 80, 155135 (2009). 41 P. Ben-Abdallah, K. Joulain, J. Drevillon, and G. Domingues, J. Appl. Phys. 106, 044306 (2009). 42 M. Francoeur, P. Mengu¸c, R. Vaillon, J. Quant. Spect. Rad. Transf. 110, 2002 (2009). 43 P. Ben-Abdallah, K. Joulain, and A. Pryamikov., Appl. Phys. Lett 96, 143117 (2010). 44 A. Pryamikov, K. Joulain, P. Ben-Abdallah, J. Drevillon, J. Quant. Spect. Rad. 112, 1314 (2011). 45 M. Tschikin, P. Ben-Abdallah and Svend-Age Biehs, Phys. Lett. A 376, 3462 (2012). 46 R. Messina, M. Antezza, P. Ben-Abdallah, Phys. Rev. Lett. 109, 244302 (2012). 47 S. I. Maslovski, C. R. Simovski, S. A. Tretyakov, arXiv:1210.6569v1. 48 E. D. Palik, Handbook of Optical Constants of Solids, (Academic Press, Florida, 1985). 49 A. I. Volokitin and B. N. J. Persson, Rev. Mod. Phys. 79, 1291 (2007). (b) 1.9 ) s / d a r 4 1 0 1 ( ω 1.65 1.4 0 ∆2 ∆1 25 50 κ/k0 75 100 1 0.8 0.6 0.4 0.2 0 

1901.09310

1

1901

Absence of dissipationless transport in clean 2D superconductors

Dissipationless charge transport is one of the defining properties of superconductors (SC). The interplay between dimensionality and disorder in determining the onset of dissipation in SCs remains an open theoretical and experimental problem. In this work, we present measurements of the dissipation phase diagrams of SCs in the two dimensional (2D) limit, layer by layer, down to a monolayer in the presence of temperature (T), magnetic field (B), and current (I) in 2H-NbSe2. Our results show that the phase-diagram strongly depends on the SC thickness even in the 2D limit. At four layers we can define a finite region in the I-B phase diagram where dissipationless transport exists at T=0. At even smaller thicknesses, this region shrinks in area. In a monolayer, we find that the region of dissipationless transport shrinks towards a single point, defined by T=B=I=0. In applied field, we show that time-dependent-Ginzburg-Landau (TDGL) simulations that describe dissipation by vortex motion, qualitatively reproduce our experimental I-B phase diagram. Last, we show that by using non-local transport and TDGL calculations that we can engineer charge flow and create phase boundaries between dissipative and dissipationless transport regions in a single sample, demonstrating control over non-equilibrium states of matter.

cond-mat.mes-hall

cond-mat

Absence of dissipationless transport in clean 2D superconductors A. Benyamini1†*, E.J. Telford2*, D.M. Kennes3, D. Wang2, A. Williams2, K. Watanabe4, T. Taniguchi4, J. Hone1, C.R. Dean2, A.J. Millis2, A.N. Pasupathy2 1Department of Mechanical Engineering, Columbia University, New York, NY, USA 2Department of Physics, Columbia University, New York, NY, USA 3Dahlem Center for Complex Quantum Systems and Fachbereich Physik, Freie Universitat Berlin, 14195 Berlin, Germany 4National Institute for Materials Science, 1-1 Namiki, Tsukuba, 305-0044 Japan *Equally contributing authors †Corresponding author: ab4442@columbia.edu Dissipationless charge transport is one of the defining properties of superconductors (SC). The interplay between dimensionality and disorder in determining the onset of dissipation in SCs remains an open theoretical and experimental problem. In this work, we present measurements of the dissipation phase diagrams of SCs in the two dimensional (2D) limit, layer by layer, down to a monolayer in the presence of temperature (𝑻), magnetic field (𝑩), and current (𝑰) in 2H-NbSe2. Our results show that the phase-diagram strongly depends on the SC thickness even in the 2D limit. At four layers we can define a finite region in the I- B phase diagram where dissipationless transport exists at 𝑻 = 𝟎. At even smaller thicknesses, this region shrinks in area. In a monolayer, we find that the region of dissipationless transport shrinks towards a single point, defined by 𝑻 = 𝑩 = 𝑰 = 𝟎. In applied field, we show that time-dependent-Ginzburg-Landau (TDGL) simulations that describe dissipation by vortex motion, qualitatively reproduce our experimental I-B phase diagram. Last, we show that by using non-local transport and TDGL calculations that we can engineer charge flow 1 and create phase boundaries between dissipative and dissipationless transport regions in a single sample, demonstrating control over non-equilibrium states of matter. A uniform superconducting condensate can transmit kinetic energy from one end to the other via electrical currents without any dissipation1. The dissipationless energy transfer is disrupted by the motion of vortices which can arise due to an applied magnetic field, thermal or quantum fluctuations, or electrical current2 -- 4. Experimentally the magnetic field and electrical current also act as two control knobs to probe and alter the superconducting state5. Magnetic field tunes the vortex density, 𝑛𝑉 = 𝐵 Φ0 (Φ0 = ℎ 2𝑒 quantum of flux attached to a single vortex, ℎ planck's constant, 𝑒 electron charge) and electrical current exerts a Lorentz force on each vortex, 𝐹𝐿𝑜𝑟𝑒𝑛𝑧 = 𝐽Φ0𝑡 (𝑡 sample thickness and vortex length, 𝐽 current density and the force is perpendicular to the magnetic field and the current). In the classical 2D mean-field picture at 𝑇 = 0 and in the absence of pinning, vortices will form a solid due to vortex-vortex interactions. In the presence of a current the Lorenz force will move the solid, causing energy dissipation2, once the force overcomes the vortex solid confining potential at the sample boundaries. Defects inevitably present in real materials may pin vortices, so that no dissipation would be measured until the Lorenz force overcomes the pinning potential. At low vortex density vortices will be pinned individually while at high densities, where vortex-vortex interactions dominate, vortices will be pinned collectively in a solid or glass like states4,6,7. Beyond mean-field physics, thermal and perhaps quantal fluctuations can melt the solid and a non-zero current can dislodge vortices from the pinned solid/glass, and create vortex-antivortex pairs, leading to current-dependent dissipation at zero and nonzero applied field. Above the melting temperature pinned vortices will be thermally activated, contributing to dissipation4,6. Figure 1b summarizes the expected B-T phase diagram in 2D in the linear-regime in the presence of pinning6. Understanding how vortices, the topological defects of 2 a superconductor, are created and proliferate in a flowing current beyond the linear regime or even the existence of a linear regime in 2D are important problems in fundamental physics. At the same time, dissipation in 2D SCs is crucial to applications that depend on zero resistance and phase coherence in superconductors. The many-body vortex state is complex due to the interplay of long ranged vortex-vortex interactions, pinning, thermal and quantum fluctuations. In 2D, one observes interesting phenomena such as the SC-insulator transition8,9 and metallic-like behavior in the SC state10. Experimentally, in thin film SCs produced by evaporation or sputtering, crystal imperfections are observed to increase with reduced thickness11 -- 13. Ultra-thin 2D superconductors produced in this way are therefore typically in a disorder-dominated limit. Van der Waals materials are 2D in nature and can be exfoliated to a single layer with the same level of crystal imperfection as in bulk14. These material offer both a unique opportunity to study ultra-thin SCs beyond the disorder/pinning dominated limit and new perspectives into dissipation mechanisms in 2D SCs. The equilibrium B-T phase diagrams of thin crystalline SCs (the limit of small current drive) has been measured in a variety of van der Waals SCs10,14. It was recently demonstrated that at high values of current drive and moderate magnetic fields, a dissipative state with nonzero resistance emerges as 𝑇 → 015. A natural question arises: at what magnetic fields and currents does one make a transition from a dissipative to a dissipationless state at T=0? Here, we answer this question by investigating transport in 2H-NbSe2 at thickness ranging from one to four layers, and complement the measurements with theoretical and numerical analyses based on TDGL. The devices are fabricated by the dry transfer16 and via contact technique17 where an insulating hBN with embedded metallic contacts is used to pick-up a few layer 2H-NbSe2 and then placed on a second hBN, all done in an inert nitrogen environment in a glovebox. This allows one 3 to simultaneously contact the air-sensitive 2H-NbSe2, while preserving it from oxidation. As a result our samples are in the low disorder limit, with a mean free path 𝑙 > 𝜉∥ (𝑙 = ℎ 𝑒2 ⋅ 1 √2𝜋𝑛1𝑅1 is the 2D mean free path, 𝑅1 and 𝑛1 are the resistance and electron density per layer, and 𝜉∥ is the in- plane Ginzburg-Landau coherence length). The sample parameters are summarized in S1 for the main devices in the paper. Illustrations of the device geometry are shown in figure 1a. All samples are in the 2D limit with a thickness smaller than the c-axis SC coherence length, 𝑡 < 𝜉⊥~2.7nm18. We measure 4-probe voltages in a Hall bar-like geometry. We source and drain alternating current (AC) and direct current (DC) far from the measurement probes to measure the response to spatially uniform currents. We denote 𝑅𝐴𝐶 = 𝑑𝑉 𝑑𝐼 𝐼𝐷𝐶 and 𝑅𝐷𝐶 = 𝑉𝐷𝐶 𝐼𝐷𝐶 for the inferred resistances. In supplementary S2 we show that a simple heating picture due to the finite DC currents in our measurements does not explain our results. We begin our discussion by presenting data obtained in the limit of very weak applied currents. The measurements are limited by the experimental noise floor and in most regimes reasonably extrapolate to the linear response limit of current tending to zero for the AC excitation used in the experiments. Figure 1c shows the temperature dependence of the linear response resistivity at 𝐵 = 0. The traces show a temperature above which the resistivity takes the normal state value, which we identify as the mean field transition temperature 𝑇𝑐 and a temperature at which the resistivity becomes indistinguishable from zero, which we identify as the Berezinskii- Kosterlitz-Thouless (BKT) transition temperature 𝑇𝐵𝐾𝑇 = 𝜋 2 𝜌𝑆(𝑇 = 𝑇𝐵𝐾𝑇) where 𝜌𝑆(𝑇) is the temperature dependent superfluid stiffness of the 2D system19. The difference between 𝑇𝐵𝐾𝑇 and 𝑇𝑐 is a measure of the strength of beyond-mean-field fluctuations which we express in terms of 𝜂, proportional to the ratio of 𝑇𝑐 and the zero temperature superfluid stiffness: 𝜂 = 2𝑇𝑐 𝜋𝜌𝑆0 . In bulk 2D 4 materials the superfluid stiffness is much larger than the transition temperature (fluctuation parameter 𝜂 ≪ 1). Using the standard mean field temperature dependence 𝜌𝑆(𝑇) = 𝜌𝑆0 (1 − 𝑇 𝑇𝑐 ) and the data in Fig. 1b yields 𝜂 ≈ 0.11, 0.32, 0.66 for the quadrilayer, bilayer and monolayer devices respectively. Use of the transition temperatures given in the supplemental material then gives for the superfluid stiffness per layer 𝜌𝑠0,𝑙𝑎𝑦𝑒𝑟 ≈ 8.9, 5, 3.3𝐾 for the quadrilayer, bilayer and monolayer devices respectively. The large values found here for the fluctuation parameter and small values of the superfluid stiffness per layer reflect the unique fragility of the superconducting state, particularly for the mono and bilayer systems. We now turn to the full field and temperature dependence of the linear response resistivity, shown as a color map for the quadrilayer sample in Fig 1d. Consistent with the relatively small value of the fluctuation parameter, the results are consistent with mean field theory and with previous measurements on bulk crystals20. At each field, a reasonably sharp crossover separates a normal state with a resistivity that is very weakly dependent on field and temperature from a state with a resistance which is very low, and strongly field and temperature dependent, which we identify as the SC state. The crossover defines the upper critical field 𝐻𝑐2(𝑇) (defined here as the field at which the resistivity is 90% of the normal state value) shown by the white dashed line. Monolayer and bilayer devices display similar behavior, but with lower transition temperatures and correspondingly lower critical fields, and much broader crossover regimes as seen in figure 1e for the monolayer device. Below 𝐻𝑐2(𝑇) the resistance is thermally activated down to our noise floor (figure S2a). The activation energies, 𝑈, are found (S3) to vary with the magnetic field as 𝑈(𝐵) = 𝑈0 ⋅ log ( 𝐵0 𝐵 ) as expected from the logarithmic vortex-vortex interactions. The prefactor 𝑈0 is expected to arise 5 from vortex-vortex interactions primarily mediated by the superfluid stiffness; the theoretical result is 𝑈0/𝑘𝐵 = 𝜋𝜌𝑆. Results of fitting measured resistivities are shown in the inset of Fig 1d and are consistent with the results found from the analysis of the 𝐵 = 0 resistivity and further confirm the small values of the superfluid stiffness and the approximate linearity in layer number. We now turn to the current dependence of the dissipation. Figure 2a shows for a monolayer and quadrilayer device the 𝐼𝐷𝐶 dependence of the differential resistance 𝑅𝐴𝐶 in log-scale obtained at 𝐵 = 0 and the lowest temperature in this study, 250𝑚𝐾. The 𝑅𝐴𝐶 vs 𝐼𝐷𝐶 curves are independent of temperature below ~1𝐾 and we take the result as representative of the 𝑇 = 0 behavior. The differential resistance curves indicate two characteristic drive currents. The lower drive current, 𝐼𝑐, is the drive at which the measured differential resistance becomes larger than the noise floor. The larger drive current, 𝐼0, is the drive at which the differential resistance goes over to the normal state value. We interpret 𝐼0 as the `microscopic' critical current marking the destruction of superconductivity. There are two physical origins of 𝐼0: the `depairing current' for which the current excites quasiparticles over the gap, and the Ginzburg-Landau critical current related to current-induced gradients of the superconducting phase. In strongly type II materials such as the ones studied here the Ginzburg-Landau critical current is typically lower and controls the behavior. In the clean, low-T limit the Ginzburg-Landau critical current per layer is proportional to the transition temperature and to the square root of the superfluid stiffness per layer. 𝐼𝑐 is the current needed either to create a measurable number of vortex-antivortex pairs out of the condensate or to detach a measurable number of vortices from the pinned vortex lattice. Both phenomena are controlled by the superfluid stiffness of the device. Consistent with the relatively small value of the fluctuation parameter, the 𝐵 = 0 𝑅𝐴𝐶 vs 𝐼𝐷𝐶 trace shown in figure 2a for the quadrilayer device is as expected by mean field behavior: 𝐼𝑐 and 𝐼0 coincide, indicating a discontinuous change of 6 differential resistance from the noise floor to the normal state value. In contrast, the monolayer device shows a continuous onset of resistance from the noise floor consistent with the much lower superfluid stiffness. A summary of the critical currents with layer number is shown in terms of their densities in S4. We now consider the magnetic field dependence of the dissipation. A peak in the differential resistance typically occurs at the Ginzburg-Landau critical current; this appears as a red region in figures 2b-2d. We see that 𝐼0 decreases with 𝐵, initially linearly and then with some curvature. Smooth evolution of the 𝐵 = 0 behavior is observed to finite magnetic field for all samples. Traces of the magnetic field dependence of 𝑅𝐴𝐶 at 𝐼𝐷𝐶 = 0 are shown in the insets. Similarly to the DC current dependence of the resistance it is again observed that as the layer thickness is decreased the on-set of resistance occurs at a lower magnetic field, 𝐵𝑐 (noted by a white arrow in the insets), as 𝐵𝑐/𝐻𝑐2~0.7,0.2,0.03 for the quadrilayer, bilayer and monolayer respectively. This is also consistent with the much lower superfluid stiffness for lower layer number, as the on-set of resistance has to do with shaking vortices loose from the vortex lattice that is held by a force proportional to the superfluid stiffness. For the quadrilayer, at 𝐵 < 𝐻𝑐2, the resistance versus current is characterized by a sharp onset from the noise floor near the critical current 𝐼0. It is reasonable to assume that this sharp drop is to a dissipationless state, indicating that a substantial region of the I-B map corresponds to a dissipationless SC. For the bi and monolayer we observe that the noise-floor region in the I-B map shrinks quickly. In the monolayer, no sharp drops are seen in the resistance versus current down to the noise floor at ~0.1 ⋅ 𝐼0 at 𝐵 = 0. This result, that dissipationless transport in clean monolayer 2H-NbSe2 with low superfluid stiffness exists only in the limit of 𝐼 = 𝐵 = 𝑇 = 0 is the main finding of this paper. 7 For all samples, we observe 𝐼𝑐 to smoothly evolve from large applied fields to 𝐵 = 0. In the limit of 𝐵 = 0, no field-induced vortices exist in the system. At 𝐵 = 0 the dissipation arises from the creation of vortex-antivortex pairs out of the condensate21. The creation process relies on the fact that a current pushes vortices and antivortices apart in opposite directions with a force that is independent of vortex-vortex separation; this force competes with the vortex-vortex attraction ∝ 𝜌𝑆 𝑟 . The forces are equal at a distance 𝑟 ∝ 𝜌𝑆 𝐽 defining an energy barrier ∝ 𝜌𝑆𝑙𝑛 𝜌𝑆 𝐽 ; over which the vortex-antivortex pairs must be activated. The rapid decrease of superfluid stiffness with layer number then shows why the dissipation effects are much more evident in the monolayer. The dissipative state at intermediate currents, below 𝐼0, is accompanied by a saturation of the resistance at the limit of 𝑇 = 0 indicative of metallic-like behavior, figure 3a. At nonzero magnetic fields, we can gain physical intuition about the nature of the non-equilibrium metallic- like state, regime of intermediate resistance in figure 3b, using TDGL simulations (see S5-S6). Figure 3d shows the full non-equilibrium phase-diagram for a theoretical device geometry very similar to the experimental one; 40𝜉 × 40𝜉 simulated compared to ~120𝜉 × 120𝜉 in the experiment. It exhibits qualitatively similar features to the experimental phase-diagram of a quadrilayer, figure 3b with a fluctuation regime which is negligible at 𝐵 = 0 and broadens rapidly as field increases. The 90% line defining the `microscopic' 𝐻𝑐2 is very similar between experiment and theory. We see that at higher fields, a regime of nonvanishing dissipation is observed even at low drive currents. The non-vanishing dissipation comes from vortices which are detached from the vortex lattice and can move freely. This increased dissipation at lower current and higher field is more similar to the monolayer phase-diagram indicating that it is easier to detach vortices from the lattice at the monolayer limit. 8 We now add disorder (see S7). Figure 3c show the two simulated traces of resistance versus inverse temperature which show the same metallic-like behavior observed experimentally, figure 3a. The agreement to experiment solidifies the vortex dissipation picture at high field, which consists of a mixed vortex state of thermally activated pinned vortices and unpinned freely moving vortices6. In the simulations we observe the two vortex states, as well as how the freely moving vortices interchange with pinned vortices even without thermal fluctuations due to a strong enough Lorentz force, see supplemental movies M4-M11. The activated region in the temperature-dependent resistance at high field can also be used to gain new insight into the depinning mechanisms at play in 2D SCs from the dependence of the activation energy on current and field. Many theoretical works have studied depinning in 2D, predicting a power-law dependence in the weak collective pinning6,22. To the best of our knowledge these theoretical predictions have never been explored experimentally in the clean 2D limit. Shown in Figure 4a is the measured activation energy dependence on 𝐼𝐷𝐶 in a log-log scale for several magnetic fields for the quadrilayer device (for individual traces of resistance versus temperature, see supplementary information S8). We clearly see two power-law regimes across samples, separated by a current we denote by 𝐼1 (or current density 𝐽1). The magnetic field dependence of the fitted exponents for both regimes are summarized in figure 4b showing a logarithmic dependence on 𝐵. Following the theories6,22, if an increase in the exponent indicates an increase in the bundle size, we can deduce that at low drive currents it is favorable to activate single vortices for any field, while at higher currents we cross to a regime where it is favorable to activate bundles with a size that increase with magnetic field. We can collect our observations at finite drive current to construct an I-B phase diagram for dissipation in 2D NbSe2 shown in figure 4c. In dark and light blue are the two activated regimes of the pinned vortex state (dark for weak 9 dependence on current and light for strong), in yellow the metallic-like regime of the unpinned vortex state and in red the normal state. Our observations show that as the thickness is reduced to the monolayer limit, the pinned vortex state regime shrinks until it eventually disappears at the monolayer thickness and any finite current at finite fields will detach vortices from the lattice creating dissipation. Looking at the non-equilibrium phase diagram in figure 4c we recognize that if we park at a nonzero magnetic field and vary the current density in space, regions of different vortex states will be established. Realizing this will demonstrate non-equilibrium control over quantum matter23,24. In figure 5 we demonstrate how we stabilize different non-equilibrium steady-states of the 2D SC along the sample by sourcing non-uniform currents. Panels a-d show the experimental non-local response for increasing DC source current. For low DC current all non-local probes show activated behavior. As the DC current is increased, the probes closest to the source-drain contacts show saturated behavior while the furthest still show activated behavior. At the highest DC current, the source-drain area is in the normal state while the other regions are saturated. To gain intuition on the way non-uniform currents affect the SC and vortices non-locally, we simulated this scenario with TDGL in the absence of pinning, see supplemental movies M12- M15. Pictures from the movies are shown in figure 5e for a finite field and different currents, the source and drain are noted by brown rectangles, color represents the size of the SC gap with blue being zero, and the white arrow's direction and length indicate the vortex velocity direction and size correspondingly. Following figure 5a-d the panels in figure 5e were generated with increasing DC current from left to right. An area where SC is destroyed next to the contacts is observed which increases in size as the current is increased. As no pinning sites are present vortices move freely with an overall vortex velocity in a direction that is perpendicular to the local current density and 10 proportional in size to the local Lorentz force, 𝑣⃗ = 𝐹⃗𝐿𝑜𝑟𝑒𝑛𝑡𝑧 𝜂 (𝜂 is the vortex viscosity). The non- uniform current density makes vortices move faster where the current density is higher and slower where the current density is lower, noted by the white arrows size. In the presence of pinning, vortices will get pinned if the combined Lorentz force and the force from other vortices is smaller than the pinning force. To map the simulation with no pinning to the case with pinning we need to imagine that the slower vortices will get pinned and be thermally activated, located further from the source drain, while the faster ones, close to the source drain, will move freely. To summarize, few layer crystalline 2H-NbSe2 enables us to investigate the physics of clean-limit, ultra-small superfluid stiffness superconductors. We find that the critical current density decreases quickly with lower layer number, making the samples sensitive to perturbations15 at finite and even zero magnetic field in the monolayer limit. It is still unclear how this sensitivity depends on specific material parameters and if this may explain metallic-like behavior in other materials. As new extremely clean 2D SCs are found, as magic angle bilayer graphene25 for example, the extreme clean limit may be achieved, 𝑙𝑚𝑓𝑝 ≫ 𝜉∥ (𝑙𝑚𝑓𝑝, electron mean free path), where the quantum mechanics of vortices dominates and new physical regimes may emerge such as a quantum vortex liquid26, a quantum Hall fluid of vortices27,28 and a fractional quantum Hall- like states29. An intriguing avenue of future research concerns the question what would emerge in this limit at non-equilibrium or when strongly coupled to other states23,24. It is also of interest if the non-equilibrium control of interfaces between different vortex states may be used to dynamically steer novel emergent physics. 11 References: 1. Bardeen, J., Cooper, L. N. & Schrieffer, J. R. Microscopic Theory of Superconductivity. Phys. Rev. 106, 162 -- 164 (1957). 2. Bardeen, J and Stephen, M., J. Theory of the Motion of Vortices. Phys. Rev. 140, A1197 (1965). 3. Abrikosov, A. A. The magnetic properties of superconducting alloys. J. Phys. Chem. Solids 2, 199 -- 208 (1957). 4. Tinkham, M. Introduction to Superconductivity. Dover Publication, Inc. (2004). 5. Fietz, W. A. & Webb, W. W. Hysteresis in superconducting alloys temperature and field dependence of dislocation pinning in niobium alloys. Phys. Rev. 178, 657 -- 667 (1969). 6. Blatter, G., Feigel'Man, M. V., Geshkenbein, V. B., Larkin, A. I. & Vinokur, V. M. Vortices in high-temperature superconductors. Rev. Mod. Phys. 66, 1125 -- 1388 (1994). 7. Dekker, C., Wöltgens, P. J. M., Koch, R. H., Hussey, B. W. & Gupta, A. Absence of a finite-temperature vortex-glass phase transition in two-dimensional YBa2Cu3O7-δ films. Physical Review Letters 69, (1992). 8. Ma, M. & Lee, P. A. Localized superconductors. Phys. Rev. B 32, 5658 -- 5667 (1985). 9. Anderson, P. W. Theory of dirty superconductors. J. Phys. Chem. Solids 11, 26 -- 30 (1959). 10. Kapitulnik, A., Kivelson, S. A. & Spivak, B. Anomalous metals -- failed superconductors. arXiv:1712.07215 (2017). 12 11. Dynes, R. C., Garno, J. P. & Rowell, J. M. Two-Dimensional Electrical Conductivity in Quench-Condensed Metal Films. Phys. Rev. Lett. 40, 479 -- 482 (1978). 12. Haviland, D. B., Liu, Y. & Goldman, A. M. Onset of superconductivity in the two- dimensional limit. Phys. Rev. Lett. 62, 2180 -- 2183 (1989). 13. Gantmakher, V. F., Golubkov, M. V, Lok, J. G. S. & Geim, A. K. A. Giant negative magnetoresistance of semi-insulating amorphous indium oxide films in strong magnetic fields. J. Exp. Theor. Phys. 82, 951 -- 958 (1996). 14. Saito, Y., Nojima, T. & Iwasa, Y. Highly crystalline 2D superconductors. Nat. Publ. Gr. 2, 1 -- 18 (2016). 15. Tamir, I. et al. Extreme Sensitivity of the Superconducting State in Thin Films. arXiv:1804.04648 (2018). 16. Wang, L. et al. One-dimensional electrical contact to a two-dimensional material. Science (80-. ). 342, 614 -- 617 (2013). 17. Telford, E. J. et al. Via Method for Lithography Free Contact and Preservation of 2D Materials. Nano Lett. 18, 1416 -- 1420 (2018). 18. Foner, S. & McNiff, E. J. Upper critical fields of layered superconducting NbSe2 at low temperature. Phys. Lett. A 45, 429 -- 430 (1973). 19. Kosterlitz, J. M. & Thouless, D. J. Ordering, metastability and phase transitions in two- dimensional systems. J. Phys. C Solid State Phys. 6, 1181 -- 1203 (1973). 20. Nader, A. & Monceau, P. Critical field of 2H-NbSe2 down to 50mK. Springerplus 3, 16 (2014). 13 21. Ambegaokar, V., Halperin, B. I., Nelson, D. R. & Siggia, E. D. Dynamics of superfluid films. Phys. Rev. B 21, 1806 -- 1826 (1980). 22. Griessen, R., Hoekstra, A. F. T., Wen, H. H., Doornbos, G. & Schnack, H. G. Negative-μ vortex dynamics in high-Tcsuperconducting films. Physica C: Superconductivity and its Applications 282 -- 287, (1997). 23. Basov, D. N., Averitt, R. D. & Hsieh, D. Towards properties on demand in quantum materials. Nat. Mater. 16, 1077 -- 1088 (2017). 24. Tokura, Y., Kawasaki, M. & Nagaosa, N. Emergent functions of quantum materials. Nature Physics 13, 1056 -- 1068 (2017). 25. Cao, Y. et al. Unconventional superconductivity in magic-angle graphene superlattices. Nature 556, 43 -- 50 (2018). 26. Rozhkov, A. & Stroud, D. Quantum melting of a two-dimensional vortex lattice at zero temperature. Phys. Rev. B 54, 12697 -- 12700 (1996). 27. Choi, M. Y. Quantum Hall effect in ideal superconducting arrays at zero temperature. Phys. Rev. B 50, 10088 -- 10091 (1994). 28. Stern, A. Quantum Hall fluid of vortices in a two-dimensional array of Josephson junctions. Phys. Rev. B 50, 10092 -- 10106 (1994). 29. Onogi, T. & Doniach, S. Simulation of quantum melting of the vortex lattice and of fractional quantum Hall-like states of the quantum vortex liquid in 2D superconductors. Solid State Commun. 98, 1 -- 5 (1996). 30. Feigel'man, M. V., Geshkenbein, V. B. & Larkin, A. I. Pinning and creep in layered 14 superconductors. Phys. C Supercond. its Appl. 167, 177 -- 187 (1990). 31. Le, L. P. et al. Magnetic penetration depth in layered compound NbSe2 measured by muon spin relaxation. Phys. C Supercond. 185 -- 189, 2715 -- 2716 (1991). 32. Mattheiss, L. F. Band Structures of Transition-Metal-Dichalcogenide Layer Compounds. Phys. Rev. B 8, 3719 -- 3740 (1973). Acknowledgements We deeply thank Dani Shahar, Daniel Rhodes and Valerii Vinokour for fruitful discussions and input. This research was primarily supported by the NSF MRSEC program through Columbia in the Center for Precision Assembly of Superstratic and Superatomic Solids (DMR-1420634), the Global Research Laboratory (GRL) Program (2016K1A1A2912707) funded by the Ministry of Science, ICT and Future Planning via the National Research Foundation of Korea (NRF), and Honda Research Institute USA Inc. We acknowledge computing resources from Columbia University's Shared Research Computing Facility project, which is supported by NIH Research Facility Improvement Grant 1G20RR030893-01, and associated funds from the New York State Empire State Development, Division of Science Technology and Innovation (NYSTAR) Contract C090171, both awarded April 15, 2010. AJM and DMK were supported by the Basic Energy Sciences Division of the U.S. Department of Energy under grant DE-SC0018218. DMK additionally acknowledges support by the Deutsche Forschungsgemeinschaft through the Emmy Noether program (KA 3360/2-1). Author information Contributions: 15 The experiment was designed by AB and EJT, devices fabricated by AB, EJT and AJW, data taken by AB, EJT and DW, analysis by AB and EJT, theory and simulation by DMK and AJM, hBN crystals grown by KW and TT. All authors contributed equally to the manuscript. Competing interests: The authors declare no competing interests. Corresponding author: Avishai Benyamini -- ab4442@columbia.edu 16 Captions: Figure 1: Equilibrium phase diagram of a 2D superconductor. a. Illustration of a fully encapsulated 2H-NbSe2 device including the measurement setup. b. Illustration of the equilibrium phase-diagram of a 2D superconductor with pinning. The normal state is shown in light red, the pinned vortex state in light blue and a vortex solid or glass state at 𝑇 → 0 in light green. c. Temperature traces of 𝑅𝐴𝐶 normalized to the normal state resistance, 𝑅𝑁, at 𝐵 = 0 for the quadrilayer, bilayer and monolayer devices with a reference to TDGL simulation showing the expected behavior from mean-field at linear response. d. 2D color map of 𝑅𝐴𝐶 for the quadrilayer device as a function of temperature and magnetic field. Dashed line shows the 0.9 ⋅ 𝑅𝑁 line. Inset: Dependence of the vortex dislocation energy scale 𝑈0 on layer number. A linear fit, 𝑈0 = 𝜀0 ⋅ 𝑁, crossing the origin gives 𝜀0 = 13.44 𝐾 per layer. From the theoretical form30, 𝑈0 = 2𝑡 𝑐𝑚Φ0 𝜇0𝜆2 (𝑐𝑚 is a model dependent constant, 𝜇0 the vacuum permeability and 𝜆 the bulk penetration depth), we get 𝑐𝑚~5.6 ⋅ 10−3 assuming 𝜆 = 250𝑛𝑚 for bulk 2H-NbSe231 and 𝑡0 = 0.62𝑛𝑚 the thickness of a single layer32, which is of the order of the vortex-vortex interaction form4, 𝑐𝑚 = 6.3 ⋅ 10−3, and an order of magnitude larger than of the dislocation mediated 2D melting30, 𝑐𝑚~2.4 ⋅ 10−4. e. 2D color map of 𝑅𝐴𝐶 for the monolayer device showing much wider transitions with respect to the quadrilayer device. Figure 2: Absence of dissipation less transport in a 2D superconductor. a. Comparison of the current induced differential resistance of monolayer and quadrilayer devices emphasizing the on- set of resistance. Black arrows donate the current at which differential resistance is observed above the noise floor, 𝐼𝑐. b-d. Full B-I colormaps of 𝑅𝐴𝐶/𝑅𝑁, shown in log-scale, for a monolayer, bilayer and quadrilayer devices. The two critical currents 𝐼𝑐 and 𝐼0 are noted at 𝐵 = 0. 17 Figure 3: TDGL simulation reproduce metallic-like behavior and main non-equilibrium experimental features. a, c. Metallic-like behavior in experiment (quadrilayer 024) and from TDGL simulation including 100 pinning sites. Resistance is shown in log-scale vs inverse temperature traces for two different currents and magnetic fields noted in the panels and on the phase diagrams, panels b and d, by corresponding light and dark gray dots. b, d. Corresponding non-equilibrium phase diagrams to panels a,c. Insets show zoom-ins on the low magnetic field and high current regime. Figure 4: Vortex size controlled by magnetic field and current. a. Activation energy dependence on 𝐼𝐷𝐶 is plotted for several magnetic fields. Black arrows point to the estimated cross- over current. Dashed lines show power-law fits to the higher current regime. b. Dependence of the exponent extracted from the power-law fits for both low (gray) and high current (black) regimes. Dash lines shows a fit to 𝛼(𝐵) = 𝛼0 ⋅ ln ( 𝐵 𝐵𝛼 ) with 𝛼0 = 0.034 and 𝐵𝛼 = 2.88𝑇 in the low current regime and 𝛼0 = 0.205 and 𝐵𝛼 = 3𝑇 for the high current regime. c. Illustration of the non- equilibrium phase-diagram summarizing the different phenomenological regimes observed and the inferred theoretical physical pictures. Figure 5: Non-equilibrium real-space control over the superconducting state. a-d. Normalized 𝑅𝑎𝑐 as a function of 1/𝑇 measured at different distance from the source-drain electrodes for four DC currents. Illustrations of the device use the color scheme used in figure 4d for the different non-equlibrium steady-states. Vortices are illustrated by black points with arrows indicating their direction of motion. Panel b for example, show that the region closer to the source drain path is in the unpinned vortex state while the furthest is in the pinned vortex state. e. Frames from simulated vortex dynamics movies, see supplementary materials, for a non- uniform current at a fixed magnetic field for four different current regimes. 18 a. Side view Ibias V Au BN BN SiO2 NbSe2 NbSe2 Top view Ibias V Au 1 B=0 G.L Equilibrium phase-diagram Normal Activated pinned vortex state Vortex solid/glass Hc1 Hc2 c. N R / C A R 0.8 0.6 0.4 0.2 0 0.5 0.6 b. 2 c T g n i t l e m T d. ] K [ T 6 5 4 3 2 1 0 1L 2L 4L 0.9 0.7 0.8 T / Tc2 Monolayer 1 RAC [Ω] 100 0.9·R N 0 0.5 1 B [T] 1.5 2 e. RAC [Ω] Quadrilayer U 0=ε 0·N 60 ] K [ 30 0 U 0 0 1 2 3 4 Layer No. 5 0.9·R N 0 1 2 3 B [T] 4 5 20 10 0 ] K [ T 4 3 2 1 0 0 Figure 1 a. 10-1 Monolayer N R / C A R 10-2 Ic Quadrilayer 10-3 0 0.2 0.8 1 0.6 0.4 IDC / I0 b. ] A μ [ C D I 16 12 8 4 0 I0 Ic Quadrilayer IDC=0 100 Ic , I0 RAC / RN N R / C A R 10-1 10-2 10-3 10-2 100 10-1 10-2 10-3 100 10-1 B / HC2 250mK 5 4 250mK 2 2.5 0 1 2 3 B [T] Monolayer IDC=0 100 N R / C A R 10-1 10-2 10-3 10-2 100 10-1 B / HC2 c. 40 30 I0 20 10 Ic Bilayer 100 IDC=0 N R / C A R 10-1 10-2 10-3 10-2 100 10-1 B / HC2 d. 120 100 80 60 40 20 0 0 0.1 0.2 0.3 B [T] 250mK 0.4 0.5 0 0 0.5 1.5 1 B [T] Figure 2 a. ] Ω [ C D R 101 10-1 b. ] A μ [ C D I 120 100 80 60 40 20 Experiment 024-4L 2.3T, 7.2μA 0.6T, 39μA c. N R / R 100 10-1 TDGL simulation including pinning 0.52·Hc2, 0.113·Jc 0.34·Hc2, 0.267·Jc 0 1 2 3 4 0 4 8 12 16 20 1/T [1/K] RDC [Ω] 30 024-4L 250mK d. 0 C J / J Tc/T no pinning R / RN 1 1 0.92 0.84 0 0.015 0 R/RN ~0.99 1 0.8 0.6 0.4 0.2 R/RN ~0.1 110 100 90 -0.15 0 R/RN ~0.99 0 -5 -4 R/RN ~0.03 -2 -3 -1 B [T] 0 0 0 0.2 0.4 0.6 B / Hc2 0.8 1 Figure 3 Activated regime b. 1.5 1 low IDC high IDC 10mT 50mT 200mT U = U 0·(Id c 0/Id c)α α 0.5 0 a. 103 ] K [ U 102 024-4L 101 100 600mT 101 IDC [uA] 102 101 102 B [mT] 103 Figure 4 Non-equilibrium phase-diagram c. J Normal steady-state Unpinned vortex state Pinned vortex state II decreased thickness Pinned vortex state I J0 JcJ1 B a. L Idc Vortex motion, R>0 Boundary Idc b. c. Idc d. Idc V1 V2 V3 W SC R=0 IDC=18.8μA 024-4L, 600mT L [μm] 1.2 2.4 3.6 100 N R / C A R 10-1 10-2 0.2 0.4 0.6 0.8 1/T [1/K] 1.0 1.2 e. J/Jc=0.4, B/Hc2=0.07 42.3μA 65.8μA 122μA 100 10-1 10-2 0.2 1.0 1.2 0.4 0.6 0.8 1/T [1/K] J/Jc=2.5 1.0 1.2 0.4 0.6 0.8 1/T [1/K] J/Jc=1.66 100 10-1 10-2 0.2 1.0 1.2 0.4 0.6 0.8 1/T [1/K] J/Jc=0.83 100 10-1 10-2 0.2 Figure 5 Supplemental information for Absence of dissipationless transport in clean 2D superconductors A. Benyamini1†*, E.J. Telford2*, D.M. Kennes3, D. Wang2, A. Williams2, K. Watanabe4, T. Taniguchi4, J. Hone1, C.R. Dean2, A.J. Millis2, A.N. Pasupathy2 S1 - Device parameters The table below summarizes the parameters for the three main devices shown in the paper. Each device had multiple contacts that showed similar results. For information on other measured devices see S3. Device 002 003 024 Layer # 1 2 4 𝑹[𝛀] 𝑻𝒄 [𝑲] 𝑯𝒄𝟐 [𝑻] 𝝃 [𝒏𝒎] 𝒍 [𝒏𝒎] 71 36 28 3.5 5 6.2 2.3 4.2 5.3 11.8 8.5 7.4 49.1 48.5 31.1 S2 -- Heating discussion Joule heating of micron size 2D superconductors can happen due to finite resistance at finite temperature of the SC or due to finite contact resistance at the interface between the embedded gold electrodes and NbSe2. The measured 4-probe sheet resistance of few layered NbSe2 is 𝑅∎~17 − 110Ω depending on layer number and the contact resistances are of the order of 100′𝑠 of ohms. Figure S1 -- Heat balance equation and heating at jumps. a. Solution of the heat balance equation for 𝛽 = 6 and 𝑈 = 24𝐾 for varying 𝑝 ≡ 𝑅0𝐼2/𝐴. For low 𝑝 𝑇𝑒𝑙~𝑇𝑝ℎ for all temperatures, as 𝑝 is increased a 'hump' is observed around a finite 𝑇𝑝ℎ which turns into an instability with two stable solutions. At low 𝑇𝑝ℎ the two stable solution are 𝑇𝑒𝑙 = 𝑇𝑝ℎ and 𝑇𝑒𝑙 = 𝑇𝑠𝑎𝑡𝑢𝑟𝑎𝑡𝑖𝑜𝑛 > 𝑇𝑝ℎ, showing that a heat balance equation can create saturation in the sample for high enough power. One feature we do not observe in experiment is the jump as a function of 𝑝 of 𝑇𝑒𝑙 to 𝑇𝑝ℎ. b. Line traces of 𝑅𝐷𝐶 versus 𝐼𝐷𝐶 at low magnetic field at 𝑇 = 250𝑚𝐾. Jumps are observed at a finite current. c. Summary of 2 , at the jump point for different magnetic fields. The blue and red dots correspond to the two the power, 𝑃 = 𝑅𝐷𝐶 ⋅ 𝐼𝐷𝐶 sides of the observed hysteresis, see S3.  To calculate the heating of the SC equations is needed1, 𝑃 = 𝐴 ⋅ (𝑇𝑒𝑙 𝛽 − 𝑇𝑝ℎ a heat balance 𝛽 ) , where 𝑃 = 𝑅(𝑇) ⋅ 𝐼2 is the power coming in, 𝑅(𝑇) = 𝑅0 ⋅ exp (− activation energy, I is the sourced current, 𝐴 is a conversion factor, 𝑇𝑒𝑙 will be the temperature of the SC, 𝑇𝑝ℎ will be the temperature of the main source for thermal equilibration and 𝛽 is the exponent. This formalism will give saturation at low temperature for a finite power. The fits to our data sets work for 𝑈 ≲ 10𝐾 which are measured only close to 𝐻𝑐2. Another feature of the heat-balance equation which isn't observed in the measured data is a jump in 𝑇𝑒𝑙 as the power gets to a critical value, see figure S1a at 𝑇𝑝ℎ → 0. ) is the resistance of the SC, 𝑈 is the 𝑈 𝑘𝐵𝑇 To check if the jumps observed for the 4-layer device are due to trivial heating we plot the power at the jump point for several magnetic fields, figure S1b-c. In panel b we show the resistance in log10 at high currents. In panel b we show the inferred power at the jumps. The power is evidently not constant at these points suggesting that a simple heating picture is not enough. In the case of heating from the contact resistance, assuming again that thermalization happens through the contacts, we would anticipate that the sample would heat uniformly, a fact that we do not observe in the non-local measurement shown in the manuscript in figure 6. Another experimental evidence is that we do not see a change in the normal state resistance which is temperature dependent at higher temperatures above 𝑇𝑐. Summarizing, although heating may be the main source for observed effect, the non-equilibrium phase diagram exhibits a wealth of physics which goes beyond a heat-balance equation. We have further results that are out of the scope of this paper and will be published independently that show a physical effect that cannot at this point be connected to heating. As our TDGL simulations, see main text and S3-4, exhibit most of the measured features we work under the assumption that vortex physics is the correct picture, albeit heating may still play a role at higher current, but not the dominant one. S3 -- Activated behavior at equilibrium Figure S2 -- Activated behavior sensitive to magnetic field. a. Temperature dependence of the resistance measured at equilibrium with a small AC current for different magnetic fields. A linear slope is seen for all traces in 𝑙𝑜𝑔𝑅 verus 1/𝑇 indicating activated behavior. b. Summary of the fitted activation energies versus magnetic field in log-scale. A cross-over is observed around 50𝑚𝑇 to a logarithmic dependence as expected from the long- range vortex interactions in 2D. The red dashed line is a guide for the eye of the activation extracted at 𝐵 = 0, second dashed lines are the errorbar amplitude. S4 -- Summary of critical currents Figure S3a summarizes the critical current densities, 𝐽𝑐/0 = 𝐼𝑐/0 𝑊𝑡 (𝑊 the flake width), dependence on layer number for all measured devices in log-scale at 𝐵 = 0. Overall an exponential dependence is observed for the lower critical current density and a weaker dependence for the upper critical current density. Both critical currents converge at four layers. Theoretically 𝐽0 can be due to the cooper pair breaking current density, 𝐽𝑝𝑏 = (𝑛𝑒 is the superfluid electron density, Δ the SC gap, 𝑚 the mass of the carriers and 𝑣𝐹 the Fermi velocity), or due to the Ginzburg-Landau thin-film critical current density, 𝐽𝐺𝐿 = . At 𝐵 = 0 𝐽0 is roughly ~1010𝐴/𝑚2 two order of magnitude lower than an estimate of 𝐽𝑝𝑏 but of 6√6𝜋𝜅𝜆 the order of 𝐽𝐺𝐿 which we associate to 𝐽0. Figure S3b shows 𝐽0 at 𝐵 = 0 as a function of 𝐻𝑐2 with a linear fit to 𝐽𝐺𝐿 giving 𝜆~280 − 310𝑛𝑚 (for 𝜉∥ = 8 − 10𝑛𝑚) on the order of 𝜆𝑏𝑢𝑙𝑘 = 250𝑛𝑚. 𝑛𝑒𝑒Δ 𝑚𝑣𝐹 𝐻𝑐2 At sufficiently large magnetic fields, a substantial number of vortices exist in the SC sample, which can be pinned either collectively or by disorder. We can attempt to understand dissipation at these fields in terms of the motion of these vortices. In this picture, the lower critical current density, 𝐽𝑐, is indicative of the minimal Lorenz force needed to free pinned vortices. At intermediate magnetic fields we observe that 𝐽𝑐 < 𝐽0 for all sample thicknesses studied here, as expected for a type-II SC with weak vortex pinning2,3. Two possible limits exist for the depinning force, depending on whether it is single vortices or a collectively pinned vortex bundle that is being depinned2,3. We can convert 𝐽𝑐 to these two limiting forces. In the limit of single vortex depinning, the critical force acting on a single vortex is 𝐹𝑆𝑉 = 𝐽𝑐𝜙0𝑡, while in the collective limit, the force acting on all vortices collectively is 𝐹𝑁 = 𝑁𝑉 ⋅ 𝐹𝑆𝑉 (𝑁𝑉 = is the number of vortices in the sample, 𝐴 = 𝐿𝑊 is sample area and 𝐿 is the sample length). The two conversions from Jc to force are shown in figure S3c as dashed and full lines respectively. The blue, red and black traces represent the data for the monolayer, bilayer and quadlayer device respectively. The correct depinning force will depend on 𝐵𝐴 Φ0 the size of the vortex bundle which should increase at larger magnetic fields2 and reside between the two limits, see illustrations in the figure. Measurement on other samples of corresponding layer number show similar force magnitudes, though the critical magnetic field varies between samples which shifts the position of the curves. We find an exponential increase of the vortex depinning force with increasing layer number. We postulate that the increase is due to enhancement of SC with layer number due to the tunnel coupling between layers or due to correlated pinning between layers as observed by the increase in 𝑇𝑐2, the SC gap or the superfluid stiffness with layer number. Figure S3 -- Behavior of the critical currents in the few layer limit. a. Summary of the critical current densities for various devices plotted versus layer. b. Dependence of 𝐽0 at 𝐵 = 0 on 𝐻𝑐2. A linear fit to 𝐽0 = shown. c. Conversion of 𝐼𝑐 to the force acting on the vortices in the two limits where the force acts on a single vortex, dash lines, and on all vortices in the sample, full line. The conversion formulas are noted and illustrations for the vortex bundle state which will get depinned at the lowest current. 6√6𝜋𝜅𝜆 𝐻𝑐2 is S5 -- TDGL simulations -- No pinning We use deterministic TDGL equations4 -- 6 to describe the dynamics of the 2D superconductor in the presence of both a magnetic field as well as a sourced current. The key quantities to monitor are the complex superconducting order parameter Δ𝑒𝑖𝜙, the charge density 𝜌 as well as the current density 𝑗⃗. The electromagnetic fields are represented by the vector potential 𝐴⃗(𝑡) and scalar potential Θ(𝑡). We choose units by ℏ = 𝑐 = 𝑒 = 1 which means that the superconducting flux quantum Φ0 = = 𝜋. ℎ𝑐 2𝑒 The equations to solve are 1 𝐷 (𝜕𝑡 + 2𝑖Ψ)Δ = 1 𝜉2𝛽 Δ[𝑟 − 𝛽Δ2] + [∇⃗⃗⃗ − 2𝑖𝐴⃗(𝑡)] 2 Δ 𝜌 = Ψ − Θ 2 4𝜋𝜆𝑇𝐹 𝑗 = 𝜎(−∇Ψ − 𝜕𝑡𝐴(𝑡)) + 𝜎 𝜏𝑠 𝑅𝑒 [Δ∗ ( ∇ 𝑖 − 2𝐴) Δ] Here 𝐷 is the normal state diffusion constant, Ψ is the electrochemical potential per electron charge, 𝜉 = √6𝐷𝜏𝑠 is related to the superconducting coherence length 𝜉0 = 𝜉/√ 𝑟 𝛽 , where 𝜏𝑠 is the spin-flip scattering time, 𝜆𝑇𝐹 is the Thomas-Fermi static charge screening length and 𝛽 is a system dependent constant that sets the magnitude of the order parameter. For definiteness we measure lengths in units of 𝜉 and time in units (which we write simply as 𝐷−1 since 𝜉 is our unit of length) and choose parameters 𝛽 = 1, 𝜎 = 0.1, of 𝜉2 𝐷 𝜏𝑠𝐷 = 1 6 and 2 𝜆𝑇𝐹 𝜉2 = 0.1. We chose those units for definiteness, but verified that none of the general conclusions is lacking. To close this set of equations we supplement them with the continuity equation 𝜕𝑡𝜌 + ∇⃗⃗⃗ ⋅ 𝑗⃗ = 0 and the Poisson equation for the scalar potential ∇2Θ = −4𝜋𝜌 Using a finite elements approach we solve the coupled partial differential equations with periodic boundary conditions in the y-direction while being open in the x-direction (choosing first derivatives to vanish at that boundary as a boundary condition). We discretized time in steps of 𝐷Δ𝑡 = 0.0001, but verified numerically that the results obtained are converged upon decreasing this numerical parameter further. For finite sourced current we choose the boundary conditions of the current density such that the one end of the open boundary acts as a particle source while the other acts as a drain 𝑗𝑥(𝑥 = 0, 𝑦, 𝑡) = 𝑗𝑥(𝑥 = 𝐿, 𝑦, 𝑡) = 𝑗0 with 𝐿 the size of the system. The initial conditions for all other variables but the order parameter Δ are chosen to be zero at 𝑡 = 0, while for Δ(𝑥, 𝑦, 𝑡 = 0) we choose initial values drawn from a random uniform distribution in the interval [0,0.001]. For given external magnetic field 𝐵 and zero external electric field we determine 𝐴 from ∇ × 𝐴 = 𝐵 assuming Coulomb gauge. We concentrate on a geometry which has open boundaries along the x direction and periodic boundary conditions along the y direction. The system we consider is thus a torus. However, when vortices move in a slap geometry with open boundaries vortices are destroyed at the one end and created at the other giving similar physics. S6 -- Simulating B-T phase diagram at equilibrium with no pinning We find that TDGL simulation of the B-T phase diagram, figure S4b (see also S5), with no pinning and no thermal fluctuations qualitatively reproduces the transition from the normal state to the activated region seen in the experiment, figure 1d and S4a. Due to the absence of pinning and thermal fluctuations in the TDGL simulation, the activated behavior is not captured well in this model. However, the dependence of the critical field on temperature is captured well. One of the findings of the TDGL simulation, see supplemental movies M1-M3, is that for samples of the sizes typically achieved in exfoliated monolayers, edge effects are of importance, and the details of the sample geometry play a significant role in the shape of the critical field line as a function of temperature. Figure S4 -- Comparison of TDGL simulation with no pinning to experiment. a. Experimental phase diagram for the quadrilayer device as shown in the manuscript, figure 1. b. TDGL B-T phase diagram. S7 - Simulating TDGL including disorder To include disorder we generalize the above equations, S4, by replacing 𝑟 → 𝑟(𝑥, 𝑦). We rewrite 𝑟(𝑥, 𝑦) = 𝑟0𝑓(𝑥, 𝑦) with 𝑟0 setting the superfluid stiffness without disorder and 𝑓(𝑥, 𝑦) describing the disorder effects. As 𝑓(𝑥, 𝑦) we choose 𝑁 𝑓(𝑥, 𝑦) = 1 − ∑ 𝛿𝑖 ⋅ exp (− 𝑖=1 (𝑥 − 𝑥𝑖)2 𝜁2 − (𝑦 − 𝑦𝑖)2 𝜁2 ) , where 𝑁 describes the total number of defects, (𝑥𝑖, 𝑦𝑖) denotes the position of the 𝑖-th defect and 𝛿𝑖 is the 𝑖-th defect's strength. We draw 𝑥𝑖 and 𝑦𝑖 from a uniform distribution (0, 𝐿] as well as 𝛿𝑖 from (0, 𝛿𝑚𝑎𝑥]. S8 -- Hysteresis at low B Hysteresis in measured resistance is observed with current sweeps at low fields. Figure S5 shows on the right the full measured diagram with a dark line showing the 0.01 ⋅ 𝑅𝑁 and 0.99 ⋅ 𝑅𝑁 resistance contours. The finite magnetic regime above ~350𝑚𝑇 show no hysteresis. The two left contours show zoom-ins on lower and lower magnetic field regimes at higher absolute currents. This lower regime exhibits a jump in the resistance which is hysteretic. This is observed by the differences in the positive and negative direct currents scans and by the black line shown in the upper middle panel representing the position where the jump occurs in the lower middle panel. Figure S5 -- Hysteresis of resistance jumps. Resistance, 𝑅𝐷𝐶 = 𝑉𝐷𝐶/𝐼𝐷𝐶, colormap in log-scale. The right panel shows the full measurement done by sweeping positive to negative currents. The data around zero current is removed due to artifacts of division by zero. S9 -- Temperature dependence at non-equilibrium The non-equilibrium, finite DC current, behavior is shown in figure S6a-c for three magnetic fields representative of the different observed physical regimes. The blue to green traces show 𝑅𝐷𝐶(1/𝑇) for the same current range on all plots. At 10𝑚𝑇 the activated behavior only weakly depends on the current amplitude while at 600𝑚𝑇 it depends strongly and stops behaving activated at intermediate currents. At higher currents at 200𝑚𝑇 and 600𝑚𝑇, yellow to red traces, shows saturation of 𝑅𝐷𝐶(1/𝑇) as 1/𝑇 → ∞, while for 10𝑚𝑇 no such behavior is observed. Jumps are observed both at 10𝑚𝑇 and 200𝑚𝑇 which reduce in amplitude and disappear at increased magnetic field. Figure S6 -- Temperature dependence of resistance at non-equilibrium. a-c. 𝑅𝐷𝐶 shown in log-scale vs inverse 𝑇 for varying 𝐼𝐷𝐶 for three representative magnetic fields 10𝑚𝑇, 200𝑚𝑇 and 600𝑚𝑇. Blue to green traces represent the same 𝐼𝐷𝐶 in a-c. Yellow to red represent larger currents as noted in each legend.  Figure S7 summarizes phenomenologically the observed phases at our lowest measuring temperature, 𝑇~250𝑚𝐾. The dashed lines are inferred from resistance contours from figure S5, the colored circles are from the different observed crossovers shown in figure S6 between the two activated regimes and between the activated to saturated regimes. The grey area is shown for currents we cannot associate with activation or saturation behaviors. Figure S7 -- Non-equilibrium phase diagram. Dashed lines and colored circles are inferred from measurements. The underlying colors represent the different phenomenological regimes discussed in more detail in the supplementary and the manuscript. To clarify the temperature dependence, we draw three phase diagrams showing the inferred physical regimes as a function of temperature and current for three different magnetic fields. The phase diagrams are shown in figure S8. The cross over from the normal state is shown by the black contour in case of a continuous change in resistance, while for a jump in the resistance a red line is shown (for hysteresis see S5). The point which we get to the noise floor is shown by the black dashed line. A further red dash line is shown to shown when a discontinuity is observed below the continuous drop form the normal state. For the lowest magnetic fields, left phase diagram, we cannot extrapolate what is the nature of the physical state below the jump and it is shown in turquoise. Figure S8 -- Non-equilibrium I-T phase diagrams. a-c. Phase diagrams summarizing S6 and S7. References: Ovadia, M., Sacépé, B. & Shahar, D. Electron-Phonon Decoupling in Disordered Insulators. doi:10.1103/PhysRevLett.102.176802 Blatter, G., Feigel'Man, M. V., Geshkenbein, V. B., Larkin, A. I. & Vinokur, V. M. Vortices in high-temperature superconductors. Rev. Mod. Phys. 66, 1125 -- 1388 (1994). Tinkham, M. Introduction to Superconductivity. Dover Publication, Inc. (2004). Gor'kov, L. P. & Kopnin, N. B. Vortex motion and resistivity of type-ll superconductors in a magnetic field. Sov. Phys. Uspekhi 18, 496 -- 513 (1975). Ivlev, B. I. & Kopnin, N. B. Theory of current states in narrow superconducting channels. Sov. Phys. Uspekhi 27, 206 -- 227 (1984). Kennes, D. M. & Millis, A. J. Electromagnetic response during quench dynamics to the superconducting state: Time-dependent Ginzburg-Landau analysis. Phys. Rev. B 96, 064507 (2017). 1. 2. 3. 4. 5. 6. 

1501.07526

1

1501

Effects of short-range electron-electron interactions in doped graphene

We study theoretically the effects of short-range electron-electron interactions on the electronic structure of graphene, in the presence of single substitutional impurities. Our computational approach is based on the $\pi$ orbital tight-binding approximation for graphene, with the electron-electron interactions treated self-consistently at the level of the mean-field Hubbard model. We compare explicitly non-interacting and interacting cases with varying interaction strength and impurity potential strength. We focus in particular on the interaction-induced modifications in the local density of states around the impurity, which is a quantity that can be directly probed by scanning tunneling spectroscopy of doped graphene. We find that the resonant character of the impurity states near the Fermi level is enhanced by the interactions. Furthermore, the size of the energy gap, which opens at high-symmetry points of the Brillouin zone of the supercell upon doping, is significantly affected by the interactions. The details of this effect depend subtly on the supercell geometry. We use a perturbative model to explain these features and find quantitative agreement with numerical results.

cond-mat.mes-hall

cond-mat

Effects of short-range electron-electron interactions in doped graphene Faluke Aikebaier, Anna Pertsova, and Carlo M. Canali Department of Physics and Electrical Engineering, Linnaeus University, Kalmar, Sweden We study theoretically the effects of short-range electron-electron interactions on the electronic structure of graphene, in the presence of single substitutional impurities. Our computational ap- proach is based on the π orbital tight-binding approximation for graphene, with the electron-electron interactions treated self-consistently at the level of the mean-field Hubbard model. We compare explicitly non-interacting and interacting cases with varying interaction strength and impurity po- tential strength. We focus in particular on the interaction-induced modifications in the local density of states around the impurity, which is a quantity that can be directly probed by scanning tunneling spectroscopy of doped graphene. We find that the resonant character of the impurity states near the Fermi level is enhanced by the interactions. Furthermore, the size of the energy gap, which opens up at high-symmetry points of the Brillouin zone of the supercell upon doping, is significantly affected by the interactions. The details of this effect depend subtly on the supercell geometry. We use a perturbative model to explain these features and find quantitative agreement with numerical results. PACS numbers: 73.22.Pr,71.55.-i,31.15.aq,71.10.Fd I. INTRODUCTION Graphene – a two-dimensional allotrope of carbon, has attracted considerable attention in recent years, largely due to its remarkable electronic properties stemming from the massless-Dirac-fermion nature of its low-energy quasiparticle states.1 Most of the electronic properties of graphene that have been studied experimentally can be well described by non-interacting single-particle theory. However, electron-electron interactions in graphene are expected to be strong. In undoped clean graphene, the density of states at the Fermi level vanishes and therefore the Coulomb potential is not screened.2,3 Recent experi- ments have shown that unscreened Coulomb interactions lead to reshaping of the ideal conical energy dispersion expected in graphene.4 More precisely, the Fermi velocity near the Dirac point acquires a logarithmic correction as a result of interactions. From a theoretical viewpoint, it can be shown that this logarithmic enhancement arises from the non-local exchange interaction, already at the level of the first- order Hartree-Fock perturbation theory.5 Hence, it is the long-range nature of the electron-electron interactions in graphene that is responsible for the logarithmic correc- tion, the most striking interaction effect observed so far in this material in the absence of external magnetic fields. As a result, theoretical work has mostly focused on inves- tigations of long-range interactions in graphene, using a variety of techniques ranging from mean field6 to renor- malization group approaches.7,8 It should be noted, however, that there are several im- portant conditions that need to be satisfied in order to observe significant long-range interaction effects exper- imentally. It is necessary to be able to probe a wide range of carrier concentrations and to tune the Fermi level sufficiently close the Dirac point, where the renor- malization of the Fermi velocity is expected to be dra- matic due to the vanishing density of states. Moreover, spurious screening effects, e.g. dielectric screening from the substrate, should be avoided. This makes undoped high-quality suspended graphene an ideal platform for studying the effects of long-range electron-electron inter- actions.4 On the contrary, in the case of graphene on a substrate or in the presence of disorder and impurities these effects are less relevant. In particular, doping introduces a finite density of states at the Dirac point of graphene and the long-range part of the Coulomb potential is screened. In this case, short-range interactions become crucial. If these interac- tions are fairly strong, they can lead to interesting effects on the electronic structure, especially on the impurity states in the vicinity of the Fermi level. In fact, estimates of the on-site Hubbard U parameter in carbon-based molecules9,10 suggest that the short-range Coulomb in- teractions among π-electron in graphene can be indeed quite large, i.e. of the order of 10 eV. A similar value is obtained by accurate ab initio calculations.11 In this paper, we study the effects of short-range electron-electron interactions on the electronic structure of doped graphene. We should note that the impor- tance of impurity effects in graphene has been addressed in many theoretical studies.12–18 A number of interest- ing features have been revealed, including the opening of the gap upon doping13,14 and the appearance of im- purity (acceptor or donor) states in the vicinity of the Fermi level.12,17 There is also a great interest in ad- dressing these properties experimentally19–21 since dop- ing graphene with impurities is one way to further explore and tune its electronic, magnetic and transport prop- erties. However, the interplay of short-range electron- electron interactions and impurity potentials in graphene has not yet been fully explored. We use a single-band (π orbital) tight-binding (TB) model to describe the electronic structure of graphene. A supercell method is employed to study the effects of finite doping. A substitutional impurity is introduced 5 1 0 2 n a J 9 2 ] l l a h - s e m . t a m - d n o c [ 1 v 6 2 5 7 0 . 1 0 5 1 : v i X r a in the TB Hamiltonian as a local modification of the on-site potential at the impurity site. Here we focus on attractive impurity potentials, mimicking nitrogen impu- rity atoms which are typical dopants in graphene.19 The short-range interactions are described by means of the Hubbard model in the mean-field approximation, which is the simplest way of treating the many-body interacting problem. Interaction terms are introduced at each site in the TB Hamiltonian. We use a numerical self-consistent scheme to account for a redistribution of electronic charge around the impurity caused by interactions. As the out- put of our numerical calculations, we obtain the band structure and the local density of states (LDOS) around the impurity site. Furthermore, we calculate scanning tunneling microscopy (STM) images by integrating the LDOS over a small energy window above the Fermi level. By using this approach, we show that short-range in- teractions introduce several remarkable features in the electronic structure of doped graphene. Importantly, they enhance the resonant character of states localized in real space around the impurity, which are induced in the vicinity of the Dirac point. The complex interplay between short-range interactions and impurity potential is also responsible for non-trivial gaps at high-symmetry crossing points in the band structure of graphene, in par- ticular at the Dirac point. The paper is organized as follows. In Sec. II we in- troduce our TB model and describe how the impurity potential and short-range interactions are incorporated in the Hamiltonian. We also provide some details of the self-consistent supercell calculations. Our findings are described in Sec. III. In particular, in Sec. III A we fo- cus on the effects of interactions on the band structure of graphene and on some issues related to the super- cell geometry. In Sec. III B we discuss the changes in the resonant character of the LDOS around the impurity for varying impurity potential strength and interaction strength. The comparison between the simulated STM topographies for non-interacting and interacting cases is provided. Finally, we draw some conclusions. II. METHODOLOGY The second-quantized Hamiltonian for interacting elec- trons on a honeycomb lattice in the presence of impurity can be written as H = † εic iσciσ+ † † 0σc0σ+U iσcjσ+Uimc tijc ni↑ni↓. (cid:88) (cid:104)i,j(cid:105)σ (cid:88) i (cid:88) iσ (1) † Here c iσ and ciσ are the creation and annihilation op- erators for electron on site i and with spin σ; εi and tij are on-site energies and hopping parameters, respec- tively. Only hopping between nearest neighbors on the honeycomb lattice is included. We assume that the TB parameters are uniform, except for the on-site energy at the impurity site, and we use the values obtained by fit- (cid:17) 2 ting the TB band structures to density functional theory calculations, namely εi = 0 and tij = −2.97 eV.22 The third term in Eq. (1) represents the local impurity potential, with Uim being the impurity potential strength (Uim < 0 for attractive impurity). In our calculations we use Uim = −10 eV and Uim = −20 eV in order to obtain visible trends for the impurity states in the vicinity of the Fermi level. The last term describes the on-site interaction between two electrons with opposite spins on site i (including the impurity site), with U (U ≥ 0) being the Hubbard U pa- rameter, which expresses the strength of the intra-atomic Coulomb repulsion. Here niσ is the number operator, † defined as niσ = c iσciσ. We consider U = 0, or non- interacting case, and U = 9.3 eV, which is the value ob- tained for graphene using the constrained Random Phase Approximation method.11 In order to extract the trends in the electronic structure with increasing the interaction strength we also use a larger value of U = 20 eV. In the mean-field approximation, the two-body inter- action term in Eq. (1) becomes (cid:88) (cid:88) (cid:16)(cid:104)ni↓(cid:105)c U ni↑ni↓ = U † i↑ci↑ + (cid:104)ni↑(cid:105)c † i↓ci↓ , (2) i i where (cid:104)niσ(cid:105) is the average electron occupation number, or density, for spin-up (σ =↑) and spin-down (σ =↓) electrons. Here we consider a non spin-polarized case so that (cid:104)ni↑(cid:105) = (cid:104)ni↓(cid:105). In pristine graphene with the Fermi level exactly at the Dirac point, the average electron occupation number is a constant equal to 1/2. Adding a mean-field field on-site potential does not break the translational invariance of the crystal and the average occupation number remains constant. In fact, in orthogonal basis such a potential merely introduces a rigid shift of the energy bands (note that in non-orthogonal basis the interplay between the overlap integrals and the on-site interactions can lead to renormalization of the Fermi velocity23). However, the presence of both mean-field on-site in- teractions and impurity potential can lead to non-trivial effects in the electronic structure. In this case, the po- tential at each site depends on the average occupation number (cid:104)niσ(cid:105), which is not necessarily the same on all sites. As a result, when a carbon atom is replaced by an impurity, there will be a redistribution of electronic charge in the system. In order to capture this effect, we need to perform self-consistent calculations for the Hamiltonian in Eq. (1) and (2). At each step of the self-consistent cycle, the average occupation number for site i is calculated as occ(cid:88) (cid:12)(cid:12)bk iσ (cid:12)(cid:12)2 k (cid:104)niσ(cid:105) = 1 N , (3) where N is the number of k-points in the Brillouin zone and bk iσ are the coefficients in the expansions of the wave- functions of the Hamiltonian in terms of the localized atomic orbitals iσ(cid:105). These are obtained by diagonaliza- tion of the Hamiltonian at each k-point. The sum runs over all occupied states up to the Fermi level. Note that all calculations are done at half-filling. 3 As initial values we use the occupation numbers calcu- lated for a non-interacting problem, i.e. for a supercell of graphene with impurity (Uim (cid:54)= 0) and with U = 0. The criterion of self-consistency is (4) (cid:12)(cid:12)(cid:12)(cid:104)niσ(cid:105)s − (cid:104)niσ(cid:105)s−1(cid:12)(cid:12)(cid:12) < η, (cid:88) iσ where s is the index of the self-consistent cycle and η is a small parameter (we choose η = 10−7). We use a linear mixing scheme, in which the input density (cid:104)niσ(cid:105)s+1 at step s + 1 is calculated as a linear combination of outputs (cid:104)niσ(cid:105)s out and (cid:104)niσ(cid:105)s−1 out in from two previous steps out + λ(cid:104)niσ(cid:105)s in = (1 − λ)(cid:104)niσ(cid:105)s−1 (cid:104)niσ(cid:105)s+1 out , (5) where λ is the mixing coefficient; we use λ = 0.25, which allows us to achieve self-consistency in less than 100 steps. In order to model the effect of finite impurity con- centration, we construct a p × p supercell by replicat- ing a graphene unit cell p time along each of the two- dimensional lattice vectors [see Fig. 1(a)]. The impurity atom substitutes a carbon atom in the supercell. In this work, we use two different supercells with p = 6 and p = 7. Atomic concentration of the dopants depends on the size of the supercell so the concentration is slightly different for the two choices, namely 1.0% for a 7× 7 and 1.4% for a 6 × 6 supercell. It is known that for p = 3q, where q is an integer, the Dirac points of graphene, K and K(cid:48), are mapped onto the Γ point of the Brillouin zone of the supercell.13,24,25 as illustrated in Fig. 1(b). This does not happen if p is not divisible by 3. Therefore, the 6 × 6 supercell is special. As we explain in Sec. III A, the effects of impurity potential and interactions in this case are rather non-trivial. This is the main reason for considering two different supercell sizes. III. RESULTS A. Bandstructure It is known that the finite amount of doping opens up an energy gap at the Dirac point of graphene.13,15,17 Here we address the question of how the details of the bandstructure near the gap are affected by interactions. We start with a special supercell geometry p×p, with p divisible by 3 (p = 6 in our calculations). Figure 2(a)-(b) shows the bandstructure of the 6×6 supercell with impu- rity potential Uim = −10 eV and Uim = −20 eV, respec- tively, for three values of the interaction strength, U = 0, U = 9.3 eV and U = 20 eV. Note that in the bandstruc- ture calculations, different impurity potential and inter- action strength introduce a shift of the energy bands with FIG. 1. (Color online) (a) 6 × 6 supercell of graphene with a substitutional impurity. A magenta sphere represents the impurity atom. Dashed lines mark the unit cell of pristine graphene and arrows show the primitive lattice vectors. A and B denote carbon atoms in the two equivalent sublattices. (b) Brillouin zone folding in graphene. Shaded area (1) repre- sents the first Brillouin zone of a p × p supercell of graphene. Numbered curves correspond to the first Brillouin zone of the unit cell for different p: p = 3q − 1 (2), p = 3q (3) and p = 3q + 1. For p = 3q the Dirac points of graphene (K, K(cid:48)) are mapped onto Γ point of the folded Brillouin zone. In other cases, K and K(cid:48) are mapped onto K and K(cid:48) of the folded Brillouin zone. respect to a reference case, i.e. non-interacting pristine graphene (Uim = 0 and U = 0). In order to examine the features around the gap for different choices of parame- ters, we align the position of the doubly degenerate state (see the discussion below) in all curves in Fig. 2(a)-(b) to the value found for U = 0 for a given impurity potential strength. As we mentioned before, in the case of p = 6 both K and K(cid:48) are mapped onto Γ point,13,24,25 producing four degenerate states at Γ in the absence of impurities and interactions. When one carbon atom in the supercell is substituted by an impurity atom, a gap opens up between two states at Γ, however the other two states remain degenerate. More precisely, for U = 0 three of the four states at Γ are degenerate while one of the states moves away from the Dirac point. This situation is referred to 4 A. Within our model, this effect is solely due to interac- tions (the contribution of sublattice B to the gap is zero in the absence of interactions13). We find the following values for the two gaps at Γ from analytical calculations (see Appendix A for details). In the non-interacting case and Uim = −10 eV, the large pseudogap is -0.56 eV. For U = 9.3 eV, the large pseudogap decreases to 0.39 eV. At the same time, a small pseudogap of 0.05 eV opens up. For U = 20 eV, the large and small pseudogaps become 0.28 eV and 0.06 eV, respectively. These values are all in good agreement with the pseudogaps found in Fig. 2(a). Note that for the larger impurity potential, the trends with increasing U are the same, however for a given U both gaps are larger than in Uim = −10 eV case. Ana- lytical calculations using the perturbative model in this case also agree with numerical results. Similar features are found for a regular 7 × 7 supercell [see Fig. 2(c) and (d)]. Note that the conduction band minima at Γ have been aligned with the reference U = 0 case. The main difference from the 6× 6 supercell is that in this case there is a real band gap at K (K(cid:48)). Our perturbative analysis shows that there is no contribution from sublattice B to the gap at the Dirac point, if we assumed that the impurity is substituted in sublattice A. As in the case of the 6 × 6 supercell, the size of the gap decreases with increasing the interaction strength. Analytically, for Uim = −10 eV we find a gap of 0.20 eV for U = 0, 0.14 eV for U = 9.3 eV and 0.10 eV for U = 20 eV. These values are in good agreement with numerical calculations. Somewhat smaller values of the gaps compared to a 6×6 supercell for the same Uim and U are expected since the atomic concentration of impurities is smaller. Bandstructure calculations presented in this section lead to a conclusion that short-range interactions ef- fectively reduce the strength of the impurity potential, which results in a decrease of the large gap (pseudogap) at the Dirac point. In oder to see how the character, e.g. the energy and the spatial extent, of the electronic states around the Dirac point is affected by interactions, we need to look at the LDOS around the impurity. B. Local density of states Calculations of LDOS at the impurity site reveal sev- eral important features. A substitutional impurity in- troduces electronic states at energies comparable to the impurity potential (Uim ∼ 10 eV), i.e. far away from the Fermi level. However, there are also states appearing in the vicinity (within ∼ 1 eV) of the Fermi level.12,13,26,27 These states are the most relevant for the low-energy electronic properties of graphene and will be examined in detail. Figure 3 shows the double- or multi-peak impurity res- onances close to the Fermi level for the 6 × 6 and 7 × 7 supercells, respectively, for different impurity potential and interaction strengths. The multi-peak structure of FIG. 2. (Color online) Bandstructure of doped graphene along the high-symmetry lines of the Brillouin zone for 6 × 6 (a,b) and 7 × 7 (c,d) supercell, for varying impurity poten- tial strength Uim and interaction strength U . Left panels are for Uim = −10 eV, right panels for Uim = −20 eV. In each panel three cases are shown: U = 0 (black), U = 9.3 eV (red) and U = 20 eV (blue). Horizontal line in (a) and (b) is the position of the doubly degenerate state at Γ, adjusted to the value found for U = 0 (see text for details). Horizontal line in (c) and (d) marks the conduction band maximum, which has been aligned with the value found for U = 0. as the pseudogap13 since there is still a pair of linearly dispersed states crossing at the neutrality point. Hence, effectively there is no band gap for this special supercell size. One can clearly see from Fig. 2(a) that the size of the pseudogap decreases with increasing the interaction strength. On-site interactions cause a redistribution of charge around the impurity. We find that in the case of the attractive impurity potential Uim = −10 eV, the total average occupation at the impurity site decreases from (cid:104)n0(cid:105) = 0.89 in the non-interacting case to (cid:104)n0(cid:105) = 0.82 for U = 9.3 eV. The on-site Coulomb repulsion pre- vents extra charge from accumulating on the impurity and therefore the strength of the impurity potential is effectively reduced. For the impurity potential which is twice stronger, the pseudogap for the same values of U is noticeably larger [see Fig. 2(b)]. In addition to a large pseudogap, for U (cid:54)= 0 there is also a smaller pseudogap which opens up at Γ [see the insets in Fig. 2(a)-(b)]. There are now only two generate states at Γ, while the other two states shift, respectively, above and below the crossing point. Interestingly, the effect of interactions on the small pseudogap is opposite to that on the large one, e.g. its value increases with increasing the interaction strength. A perturbative model described in Appendix A suggests that the smaller gap results from the contribution of the states localized on sublattice B, if we assume that the impurity is substituted in sublattice 5 gion for U = 9.3 eV. Below we elaborate more on these findings. Increasing U reduces the overall strength of the im- purity potential, which is confirmed by decrease of the energy gap at the Dirac point due to the presence of impurities (Sec. III A). However, short-range electron- electron interactions controlled by U do not only change the potential directly at the impurity site but also affect the on-site potential and the charge density around the impurity (primarily nearest and next-nearest neighbors of the impurity atom). Hence, both the amplitude and the spatial extent of the impurity potential is modified by interactions. Let us assume that an attractive impu- rity can be described by a delta-function potential well. When interactions are included, the shape of the impurity potential is smoothed out (it acquires, say, a Gaussian shape). Therefore, although the strength of the poten- tial is reduced by a certain amount with increasing U , the potential can become more long-ranged (in a certain parameter space). This, in turn, will increase the overlap of the potentials from neighboring cells and enhance the inter-supercell interaction. This seems to be the situation for Uim = −20 eV and U = 9.3 eV, for both choices of the supercell. In this case, the impurity potential decreases slightly due to interac- tions (U < Uim), leading to a small decrease of the gap at the Dirac point compared to U = 0 case [Fig. 2(b) and (d)]. At the same time, we find a significant difference be- tween the average occupation numbers of the nearest and next-nearest neighbors for U = 0 and U = 9.3 eV (this can be also partly seen in the STM images in Fig. 5(a)- (b) and (d)-(e), for states in a small energy window above the Fermi level, where the neighbors of the impurity site appear brighter in the U = 9.3 eV case compared to U = 0). As a result, the amplitude of impurity res- onances in LDOS increases but their position shifts to lower energies. In contrast to this, for Uim = −20 eV and U = 20 eV, the average occupation numbers of the nearest and next- nearest neighbors of the impurity site do not change ap- preciably compared to U = 0 case. The strength of the impurity potential for this value of U is significantly reduced, leading to a large decrease of the energy gap [Fig. 2(b) and (d)]. As a result, the amplitude of impu- rity resonances increases further, however their position remain close to the U = 0 value. These features strongly suggest that the position of the impurity resonances is sensitive to the spatial extent of the impurity potential, namely the resonances move closer to zero energy when the potential becomes more long-ranged. To further clarify the changes in the intensity and the spatial character of the low-energy impurity peaks, brought about by interactions, we present the simulated STM topographies in Fig. 4 and Fig. 5 for Uim = −10 eV and Uim = −20 eV, respectively. For this we plot LDOS for each atom in the supercell,29 integrated over the energy window ∆E above the Fermi level (we choose ∆E = 0.25 eV). This gives an estimate of the tunnel- FIG. 3. (Color online) LDOS of doped graphene at the im- purity site for 6 × 6 (a,b) and 7 × 7 (c,d) supercell, for vary- ing impurity potential strength Uim and interaction strength U . Left panels are for Uim = −10 eV, right panels for Uim = −20 eV. In each panel three cases are shown: U = 0 (black), U = 9.3 eV (red) and U = 20 eV (blue). Vertical lines mark the position of the Fermi level (see text for details). the impurity resonances most likely originate from the long-range interaction, or interference, between the im- purity potentials, caused by the periodicity of the super- cell geometry.13 With increasing the impurity potential strength, the resonant peaks move closer to low energies. This is very similar to the case of strong potential impurities on the surface of a topological insulator.28 As shown in Fig. 3(a) and (b) with U = 0, the double-peak resonance in the 6 × 6 supercell approaches the Fermi level as Uim in- creases. At the same time its amplitude decreases. This finding is in agreement with semi-analytical calculations in Refs. 12, 26, and 27. The effect of the impurity poten- tial is more striking in the case of a regular 7×7 supercell [Fig. 3(c) and (d)]. In this case, in addition to shifting the peaks to lower energies, a stronger impurity poten- tial Uim = −20 eV also makes the peaks narrower, thus enhancing their resonant character [see in particular the first peak in Fig. 3(c) and (d) with U = 0]. In the interacting case, the amplitude of the resonances increases with U . This can be seen in all panels in Fig. 3 with U = 9.3 eV and U = 20 eV, with the exception of the 6 × 6 supercell with Uim = −10 eV and U = 20 eV, In where the amplitude of the peak decreases slightly. the case of Uim = −10 eV, for both supercells the impu- rity resonances move further away from the Fermi level with increasing U . This is perfectly consistent with our observations for the non-interacting case with decreasing impurity potential. However, in the case of a very large impurity potential Uim = −20 eV, the trend in the posi- tion of the resonances is less obvious. The peaks either do not move appreciably as in the case of U = 20 eV or even seem to move slightly towards the low-energy re- 6 FIG. 4. (Color online) Simulated STM topographies (LDOS for all atoms in the supercell) for 6× 6 (left panels) and 7× 7 (right panels) supercell, for a fixed impurity potential strength Uim = −10 eV and varying interaction strength: U = 0 (a,d), U = 9.3 eV (b,e) and U = 20 eV (c,f). Arrows mark the position of the impurity atom. Note the logarithmic color scale. FIG. 5. potential strength Uim = −20 eV. (Color online) Same as Fig. 4 but with impurity neighbors is even stronger than in the U = 0 case. This correlates with the corresponding features in the LDOS discussed above. ing current as electrons tunnel out of the occupied states of the STM tip into the unoccupied states of graphene. The increase of the electronic density of states in this energy window gives rise to a bright triangular feature around the impurity. Note that for the 6 × 6 supercell the impurity is located in sublattice A, while for the 7×7 supercell it is located in sublattice B. Therefore the bright triangular features in the two supercells appear rotated by 180◦ with respect to each other. The difference between non-interacting and interact- ing cases is clearly visible in the STM images. In all cases considered, the impurity site becomes progressively brighter compared to its neighbors with increasing U . This means that the electronic states in the small energy window above the Fermi level become more localized on the impurity site as a result of interactions. This is most evident in the case of the 7 × 7 supercell [Fig. 4(d)-(f)]. For Uim = −10 eV, the overall intensity of the images decreases with U . For a stronger Uim = −20 eV impu- rity potential, the trend is similar with the exception of the intermediate interaction strength U = 9.3 eV. In this case, the contribution of the nearest and next-nearest IV. CONCLUSIONS We have presented a theoretical study of the effects of electron-electron interactions on the electronic states of graphene in the presence of substitutional impurities. Using a self-consistent TB model with on-site interac- tions treated at the mean-field level, we have shown that the size of the gap, which opens up at the Dirac point in graphene upon doping, and the character of the low- energy electronic states are modified by interactions. The mechanism for these effects is provided by the interplay between the impurity potential and the on-site repul- sion, which leads to significant re-arrangement of the electronic charge around the impurity compared to the non-interacting case. In particular, we found that the size of the gap de- creases with increasing the interaction strength. Intu- itively, this can be understood as follows. In the case of an attractive impurity potential, which mimics nitrogen dopants in graphene, adding the on-site Coulomb repul- sion effectively reduces the strength of the potential, i.e. the depth of the potential well decreases. This is due to the fact the on-site repulsion prevents extra electronic charge from accumulating on the impurity site. For a special supercell size p×p, where p is divisible by 3, both K and K(cid:48) are mapped onto Γ point of the folded Brillouin zone. Therefore, in the case of undoped non- interacting graphene, there are four degenerate states at the neutrality point. It is known that when the impu- rity potential is included, a gap (pseudogap) opens up between two of these states while the other pair remains degenerate.13 We have shown that the size of this pseudo- gap is reduced by interactions. Interestingly, in addition to this, a small gap opens up between the second pair of states at the Γ point, which are otherwise degenerate in the absence of interactions. We explain these features both qualitatively and quantitatively, using a perturba- tive model based on the generalization of the approach developed by Lambin et al.13 to the interacting case. Furthermore, we have studied the behavior of the impurity-induced electronic states with and without in- teractions. There are two groups of states which can be detected in the density of states when a carbon atom in the supercell is replaced by an impurity atom. First, there are states which emerge far away from the Fermi energy, with their energies of the same order of mag- nitude as the impurity potential ( ≈ 10 eV in our cal- culations). Second, there are states appearing close to the Fermi energy and are therefore of particular inter- est. Although the way the electron-electron interactions affect the LDOS at low energies in general depends on the impurity concentration, we found clear trends in the behavior of the impurity-resonances as both parameters, i.e. the interaction strength and the impurity potential strength, are modified. Regardless of the interactions, the impurity levels move closer to the low-energy region (e.g. to the original Dirac point) with increasing the impurity potential strength. This finding is consistent with previous calculations for graphene.12,26,27 Similar result was found in another class of Dirac materials, namely in three-dimensional topolog- ical insulators in the presence of strong potential impuri- ties on the surface.28,30 However, our self-consistent cal- culations for graphene in the presence of both impurities and interactions reveal novel features. For a fixed im- purity potential, sort-range interactions tend to enhance the amplitude of the impurity-resonances in the vicin- ity of the Fermi level. The position of the resonances is also affected by the spatial extent of the effective impu- rity potential, which is modified by interactions. As the interaction strength increases, the states become more localized on the impurity atom. The differences in the spatial distribution of the low-energy impurity states in the non-interacting and interacting cases are clearly de- tectable in the simulated STM topographies. 7 Appendix A: Band gap in doped graphene supercell in the presence of interactions We use a simple perturbative model, based on the model proposed in Ref. 13 for non-interacting graphene, to explain how the impurity potential and electron- electron interactions affect the band structure of doped graphene. We first review this model for the non-interacting case with impurity.13 The Hamiltonian for such a system can be written in Dirac's notation as H = H0 + H1, (A1) where H0 is the Hamiltonian of non-interacting pristine graphene (cid:88) (cid:88) H0 = u(cid:105) Eu (cid:104)u + u(cid:105) tuv (cid:104)v, (A2) u u,v and H1 is the perturbation introduced by the periodic arrangement of impurities H1 = u(cid:105) Uim (cid:104)u . (A3) (cid:88) u∈1 Here u(cid:105) is the atomic orbital associated with site u ((cid:104)uu(cid:48)(cid:105) = δuu(cid:48)), Eu and tuv are the on-site energies and the nearest-neighbors hopping integrals, respectively; Uim is the impurity potential and the sum over u ∈ 1 refers to all impurity atoms belonging to sublattice 1, which substitute one carbon atom in each p× p supercell . In pristine graphene [Eq. (A2)], there are four states with zero energy, two for the two sublattices and two for the non-equivalent points K and K(cid:48) in the Brillouin zone (we omit the spin indices for simplicity). The cor- responding Bloch functions can be written as eiK·u u(cid:105), (cid:12)(cid:12)(cid:12)K A(B)(cid:69) (cid:88) 1(cid:112)NA(B) (A4) = u∈A(B) where K is a vector in the reciprocal space corresponding to either K or K(cid:48), u is the position of site u in real space, and NA(B) is the number of atoms in sublattice A(B). The task is to calculate the first order corrections to the energy states at the Dirac point due to the impurity potential, by using degenerate state perturbation theory. Let us assume that the impurity is substituted in sub- lattice A. Then the states (cid:12)(cid:12)K B(cid:11) and (cid:12)(cid:12)K(cid:48)B(cid:11) have zero by the states (cid:12)(cid:12)K A(cid:11) and (cid:12)(cid:12)K(cid:48)A(cid:11). We use Eq. (A3) and amplitudes on atoms in sublattice A and these states are eigenstates of zero energy. Therefore, we only need to consider the subspace of degenerate eigenstates formed Eq. (A4) to calculate the following matrix elements V11 =(cid:10)K A(cid:12)(cid:12) H1 V22 =(cid:10)K(cid:48)A(cid:12)(cid:12) H1 (cid:12)(cid:12)K A(cid:11) = (cid:12)(cid:12)K(cid:48)A(cid:11) = V11 = Uim NA = Uim p2 , Uim p2 , (A5) (A6) V12 =(cid:10)K A(cid:12)(cid:12) H1 V21 =(cid:10)K(cid:48)A(cid:12)(cid:12) H1 = 1 NA (cid:88) (cid:12)(cid:12)K(cid:48)A(cid:11) = (cid:12)(cid:12)K A(cid:11) = V12 = u∈1 Uim p2 δK(cid:48)−K,G, Uim p2 δK(cid:48)−K,G, (A8) where NA = p2 since for p×p supercell, we have N = 2p2 atoms and NA = NB = p2 atoms in each sublattice. V12(V21) in Eq. (A7) is not zero only when the vector K(cid:48) − K is equal to the reciprocal lattice vector G. This condition is satisfied only if p is divisible by 3, i.e. if p = 3q, where p, q are integers. This is essentially the result obtained in Ref. 13 and it is confirmed by our calculations presented in Fig. 2 for U = 0 case. The first corder corrections E(1) to the energy states are then given by the eigenvalues of matrix V , with matrix elements Vij (i, j = 1, 2) defined above. This gives E(1) = 0 and E(1) = 2Uim/p2 if p = 3q, and E(1) = Uim/p2 if p (cid:54)= 3q. Hence, in the case when p is divisible by 3, all four states are mapped onto Γ point. Three of these states, two of which are localized on sub- lattice B and one on sublattice A, have zero energy. The remaining state is shifted down in energy by 2Uim/p2 (Uim < 0 for attractive impurity), producing a pseudo- gap of magnitude 2Uim/p2. This is exactly the situation shown in Fig. 2(a) and (b) for the 6 × 6 supercell. In the case when p is not divisible by 3, the degeneracy between K and K(cid:48) is lifted and there are now two states at each of these points, separated by a gap of Uim/p2. This is the situation for the 7 × 7 supercell in Fig. 2(c) and (d). Combining the results for p = 3q and p (cid:54)= 3q, to the lowest order in the perturbation theory the gap (pseudo- gap) induced by the impurity potential can be written as Egap = Uim p2 + Uim p2 δK(cid:48)−K,G. (A9) For Uim = −10 eV, Egap = −0.56 eV if p = 6 and Egap = −0.2 eV if p = 7. These values coincide with the gaps found numerically [see Fig. 2(a) and (c) with U = 0]. Below we extend this model to the interacting case. The Hamiltonian of the interacting system can be written as H = H0 + H1 + H2, (A10) where H0 and H1 are given by Eq. (A2) and Eq. (A3), respectively, and H2 is the perturbation introduced by short-range interactions which is given by u(cid:105) (U·(cid:104)nuσ(cid:105))(cid:104)u. (cid:88) H2 = (A11) u In accordance with Eq. (2), (cid:104)nuσ(cid:105) is the average elec- tron occupation number on site u corresponding to spin σ. When the mean-field interaction term is included, the energy bands acquire a rigid shift. In order to compare ei(K(cid:48)−K)·u the results to the non-interacting case, we need to sub- tract this shift. Let us assume that it is proportional to an average quantity (cid:104)¯nuσ(cid:105), which is a constant. Then the energy that needs to be subtracted is U·(cid:104)¯nuσ(cid:105). (A7) 8 We now calculate the first-order corrections to the en- ergy states at the Dirac point due to interactions using the same procedure. Since the sum in Eq. (A11) can be decomposed into the sum over u ∈ A and the sum over u ∈ B, we can calculate corrections due to these two terms separately. For each of them we need to consider either the subspace formed by the states K A and K(cid:48)A, or by the states K B and K(cid:48)B. Let us assume for simplicity that all occupation numbers for atoms in sublattice A are approximately the same and equal to (cid:104)¯nuσ(cid:105), except for the impurity site. Then the corresponding matrix elements are give by U ((cid:104)nuσ(cid:105) − (cid:104)¯nuσ(cid:105)) (cid:88) u∈A,σ W A W A 1 U = 2NA 11 =(cid:10)K A(cid:12)(cid:12) H2 12 =(cid:10)K A(cid:12)(cid:12) H2 22 =(cid:10)K(cid:48)A(cid:12)(cid:12) H2 21 =(cid:10)K(cid:48)A(cid:12)(cid:12) H2 (cid:12)(cid:12)K A(cid:11) = (cid:12)(cid:12)K(cid:48)A(cid:11) p2 ((cid:104)nu(cid:48)σ(cid:105) − (cid:104)¯nuσ(cid:105)) , (cid:12)(cid:12)K(cid:48)A(cid:11) = W A (cid:12)(cid:12)K A(cid:11) = W A = U p2 ((cid:104)nu(cid:48)σ(cid:105) − (cid:104)¯nuσ(cid:105)) δK(cid:48)−K,G, (A12) (A13) (A14) 11, 12, W A W A (A15) where u(cid:48) is the impurity site; 2 in the denominator stands for spin. As before, first order corrections to the energy states are given by the eigenvalues of matrix W A, with matrix elements W A ij (i, j = 1, 2), and differ for p = 3q and p (cid:54)= 3q. These corrections give the following contri- bution to the energy gap at the Dirac point Eint(A) gap = U p2 ((cid:104)nu(cid:48)σ(cid:105) − (cid:104)¯nuσ(cid:105)) (1 + δK(cid:48)−K,G) . (A16) In a similar way, we calculate the first-order corrections in the subspace formed by states K B and K(cid:48)B. The cor- responding matrix elements are given by U ((cid:104)nuσ(cid:105) − (cid:104)¯nuσ(cid:105)) (cid:88) u∈B,σ ((cid:104)nuσ(cid:105) − (cid:104)¯nuσ(cid:105)) , W B 1 = 2NB U p2 (cid:12)(cid:12)K B(cid:11) = 11 =(cid:10)K B(cid:12)(cid:12) H2 (cid:88) 12 =(cid:10)K B(cid:12)(cid:12) H2 (cid:12)(cid:12)K(cid:48)B(cid:11) (cid:88) 22 =(cid:10)K(cid:48)B(cid:12)(cid:12) H2 (cid:12)(cid:12)K(cid:48)B(cid:11) = W B 21 =(cid:10)K(cid:48)B(cid:12)(cid:12) H2 (cid:12)(cid:12)K B(cid:11) = W B u∈nn of u(cid:48) u∈nn of u(cid:48) U p2 = 11, 12, W B W B W B ((cid:104)nuσ(cid:105) − (cid:104)¯nuσ(cid:105)) δK(cid:48)−K,G, (A17) (A18) (A19) (A20) where we assumed that all atoms in sublattice B have approximately the same occupation (cid:104)¯nuσ(cid:105), except for the nearest neighbors of the impurity atom. This is a rea- sonable assumption since these atoms are most strongly TABLE I. Values of band shifts and average occupation num- bers used in Eq. (A22) for Uim = −10 eV and U = 9.3 eV. Band shift (cid:104)¯nuσ(cid:105) (cid:104)nu(cid:48)σ(cid:105) (imp.) (cid:104)nuσ(cid:105) (nn of imp.) 4.60 eV 0.82 0.49 0.46 (cid:88) affected by the impurity. Then the contribution to the energy gap, stemming from corrections to the states lo- calized on sublattice B, is given by Eint(B) gap = U p2 u∈nn of u(cid:48) ((cid:104)nuσ(cid:105) − (cid:104)¯nuσ(cid:105)) (1 + δK(cid:48)−K,G) . (A21) Finally, combining the corrections due to the impurity potential [Eq. (A9)] and due to interactions [Eq. (A16) and Eq. (A21)], we obtain the expression for the energy gap at the Dirac point (cid:26)(cid:18) Uim (cid:40) U p2 Eint gap = + p2 + U p2 ((cid:104)nu(cid:48)σ(cid:105) − (cid:104)¯nuσ(cid:105)) (cid:88) u∈nn of u(cid:48) ((cid:104)nuσ(cid:105) − (cid:104)¯nuσ(cid:105)) (1 + δK(cid:48)−K,G) (1 + δK(cid:48)−K,G) (cid:27) (cid:41) , (cid:19) where the expression inside the first curly bracket is due to the states localized on sublattice A, while the second one is due the states localized on sublattice B. (A22) 9 p2 + U (cid:104) Uim (cid:105) p2 · ((cid:104)nu(cid:48)σ(cid:105) − (cid:104)¯nuσ(cid:105)) Let us now summarize what happens to the four zero energy states, when both the impurity potential and in- teractions are present. When p = 3q, all four states are mapped onto Γ point. As one can see from Eq. (A22), one of the states, corresponding to sublattice A, remains at zero energy, while the other one shifts by Eint(A) gap = . We can estimate this 2 quantity by taking the values of the rigid band shift and the average occupation numbers for impurity and its nearest neighbors from our numerical calculations (see Table I, numbers are similar for the two supercells). For Uim = −10 eV, U = 9.3 eV and p = 6, this energy shift is negative and is equal to −0.39 eV. This corresponds to the large pseudogap (below the Dirac point) that we identified in Fig. 2(a) for this choice of U . In a similar (cid:80) way, one of the states localized on sublattice B remains at zero energy, while the other one shifts up in energy u∈nn of u(cid:48) ((cid:104)nuσ(cid:105) − (cid:104)¯nuσ(cid:105)) = 0.054 eV. by Eint(B) This small energy shift is identical to the small pseudogap (above the Dirac point) found in Fig. 2(a). gap = 2 U p2 When p (cid:54)= 3, the degeneracy between K and K(cid:48) is lifted. There are two states at each of these points, one localized on sublattice A and the other one (cid:80) on sublattice B. The energy gap between the states p2 · ((cid:104)nu(cid:48)σ(cid:105) − (cid:104)¯nuσ(cid:105)) + is given by Eint u∈nn of u(cid:48) ((cid:104)nuσ(cid:105) − (cid:104)¯nuσ(cid:105)). For Uim = −10 eV, U = U p2 9.3 eV and p = 7, Eint gap = 0.14 eV, which is in good agreement with Fig. 2(c). gap = Uim p2 + U 1 A. H. Castro Neto, F. Guinea, N. M. R. Peres, K. S. Novoselov, and A. K. Geim, Reviews of Modern Physics 81, 109 (2009). 2 A. K. Geim and A. H. MacDonald, Physics Today 60, 35 (2007). (2006). 13 P. Lambin, H. Amara, F. Ducastelle, and L. Henrard, Phys. Rev. B 86, 045448 (2012). 14 S. S. Pershoguba, Y. V. Skrypnyk, and V. M. Loktev, Phys. Rev. B 80, 214201 (2009). 3 V. N. Kotov, B. Uchoa, V. M. Pereira, F. Guinea, and 15 S. Latil, S. Roche, D. Mayou, and J.-C. Charlier, Phys. A. H. Castro Neto, Rev. Mod. Phys. 84, 1067 (2012). 4 D. C. Elias, R. V. Gorbachev, A. S. Mayorov, S. V. Mo- rozov, A. A. Zhukov, P. Blake, L. A. Ponomarenko, I. V. Grigorieva, K. S. Novoselov, F. Guinea, and A. K. Geim, Nature Physics 7, 701 (2011). 5 M. I. Katsnelson, Graphene: carbon in two dimensions (Cambridge University Press, 2012). 6 J. Jung and A. H. MacDonald, Phys. Rev. B 84, 085446 (2011). 7 J. Gonz´alez, F. Guinea, and M. Vozmediano, Nuclear Physics B 424, 595 (1994). 8 J. Gonz´alez, F. Guinea, and M. A. H. Vozmediano, Phys. Rev. B 59, R2474 (1999). 9 D. P. C. Parr, R. G. and I. G. Ross, J. Chem. Phys. 18, 1561 (1950). 10 D. Baeriswyl, D. K. Campbell, and S. Mazumdar, Phys. Rev. Lett. 56, 1509 (1986). 11 T. O. Wehling, E. S¸a¸s ıoglu, C. Friedrich, A. I. Lichten- stein, M. I. Katsnelson, and S. Blugel, Phys. Rev. Lett. 106, 236805 (2011). 12 Y. V. Skrypnyk and V. M. Loktev, Phys. Rev. B 73, 241402 Rev. Lett. 92, 256805 (2004). 16 C. Adessi, S. Roche, and X. Blase, Phys. Rev. B 73, 125414 (2006). 17 Y. Fujimoto and S. Saito, Phys. Rev. B 84, 245446 (2011). 18 W. H. Brito, R. Kagimura, and R. H. Miwa, Phys. Rev. B 85, 035404 (2012). 19 L. Zhao, R. He, K. T. Rim, T. Schiros, K. S. Kim, H. Zhou, C. Gutirrez, S. P. Chockalingam, C. J. Arguello, L. Plov, D. Nordlund, M. S. Hybertsen, D. R. Reichman, T. F. Heinz, P. Kim, A. Pinczuk, G. W. Flynn, and A. N. Pa- supathy, Science 333, 999 (2011). 20 B. Zheng, P. Hermet, and L. Henrard, ACS Nano 4, 4165 (2010), pMID: 20552993. 21 F. Joucken, Y. Tison, J. Lagoute, J. Dumont, D. Ca- bosart, B. Zheng, V. Repain, C. Chacon, Y. Girard, A. R. Botello-M´endez, S. Rousset, R. Sporken, J.-C. Charlier, and L. Henrard, Phys. Rev. B 85, 161408 (2012). 22 S. Reich, J. Maultzsch, C. Thomsen, and P. Ordej´on, Phys. Rev. B 66, 035412 (2002). 23 F. Aikebaier, A. Pertsova, and C. Canali, (unpublished). 24 Y.-C. Zhou, H.-L. Zhang, and W.-Q. Deng, Nanotechnol- ogy 24, 225705 (2013). 28 A. M. Black-Schaffer and A. V. Balatsky, Phys. Rev. B 85, 25 R. Martinazzo, S. Casolo, and G. F. Tantardini, Phys. 121103 (2012). Rev. B 81, 245420 (2010). 26 V. M. Pereira, J. M. B. Lopes dos Santos, Castro Neto, Phys. Rev. B 77, 115109 (2008). and A. H. 27 T. O. Wehling, A. V. Balatsky, M. I. Katsnelson, A. I. Lichtenstein, K. Scharnberg, and R. Wiesendanger, Phys. Rev. B 75, 125425 (2007). 29 The discrete values of LDOS at lattice points are smeared out by adding a Gaussian broadening of γ = Natoms/Emesh, where Natoms is the number of atoms in the supercell and Emesh is the energy mesh. 30 R. Biswas and A. Balatsky, Phys. Rev. B 81, 233405 (2010). 10 

1507.03595

1

1507

Channel Blockade in a Two-Path Triple-Quantum-Dot System

Electronic transport through a two-path triple-quantum-dot system with two source leads and one drain is studied. By separating the conductance of the two double dot paths, we are able to observe double dot and triple dot physics in transport and study the interaction between the paths. We observe channel blockade as a result of inter-channel Coulomb interaction. The experimental results are understood with the help of a theoretical model which calculates the parameters of the system, the stability regions of each state and the full dynamical transport in the triple dot resonances.

cond-mat.mes-hall

cond-mat

Channel Blockade in a Two-Path Triple-Quantum-Dot System M. Kotzian,1, ∗ F. Gallego-Marcos,2, 1 G. Platero,2 and R. J. Haug1 1Institut fur Festkorperphysik, Leibniz Universitat Hannover, Appelstrasse 2, 30167 Hannover, Germany 2Instituto de Ciencia de Materiales, CSIC, Cantoblanco, 28049 Madrid, Spain (Dated: June 15, 2021) Electronic transport through a two-path triple-quantum-dot system with two source leads and one drain is studied. By separating the conductance of the two double dot paths, we are able to observe double dot and triple dot physics in transport and study the interaction between the paths. We observe channel blockade as a result of inter-channel Coulomb interaction. The experimental results are understood with the help of a theoretical model which calculates the parameters of the system, the stability regions of each state and the full dynamical transport in the triple dot resonances. PACS numbers: 73.21.La, 73.23.Hk, 73.63.Kv, 85.35.Ds, 85.35.Gv Triple quantum dots (TQDs), which have been imple- mented only recently [1 -- 4], offer the possibility of ana- lyzing new fascinating properties which are not present in double-quantum-dot systems. These new properties, to name a few, include interference phenomena between different transport channels giving rise to dark states in triangular [5 -- 8] and linear [9] dot distributions and long distant coherent states in TQDs [9 -- 13]. TQDs are, as the smallest qubit chain, a step towards more complex archi- tectures needed in quantum computation. They allow for novel applications in the field of quantum informa- tion processing, like for example as exchange-controlled spin qubits [14, 15] or as current rectifiers [1, 16]. They provide as well the implementation of quantum cellular automata processes, a combination of charging and re- configuration events in the system being a crucial pro- cess in quantum information [17, 18]. Coherent electron transfer using adiabatic passage was proposed for TQDs in series [19]. Furthermore, decoherence due to charge fluctuations is reduced in a TQD-based coded qubit as it involves a decoherence free subspace [15, 20]. Our system is a triangular-shaped TQD with one lead attached to each dot thus consisting of two double-dot paths. A triangular geometry is suitable for studying en- tanglement and effects of interference which makes it an interesting device for quantum information processing. The flexibility of this setup makes it a convenient tool for investigating the transport properties of a TQD. Trans- port can be measured separately and simultaneously for the two double dot paths and be compared or combined to study the whole TQDs physics on the basis of the dou- ble dots. Also, transitions from double dot resonances in one path to configurations of all three dots in resonance can be studied in transport. In contrast to former pub- lished works [4] where one source and two drain leads were used, we now use one drain and two source leads. In this configuration of two-path transport the dot con- nected to the drain is shared by both paths (Fig.1 (a)). The electrons from the different paths compete for the occupation of this dot. We analyze the role of interac- tions between the charge flowing through the two differ- ent paths by transport measurements. We observe, as a consequence of inter-channel Coulomb interaction, chan- nel blockade in transport. FIG. 1. (a) Schematic of the TQD setup with capacitive and tunnel couplings. (b) AFM picture of the TQD sample with the in-plane gates G1 - G4 and a QPC for charge measure- ments. The blue lines indicate the insulating barriers written by AFM. (c) Transport through path 1. Charging lines of dots A and B (solid lines) are observed. Charging of dot C (dotted line) is observed by a shift of the charging lines. (d) Transport through path 2. Charging lines of dot A and C (solid lines) can be seen and charging of dot B (dotted line) is observed by a shift of these charging lines. TQD Sample and Characterization. The measure- ments were performed on a lateral TQD made with local anodic oxidation by atomic force microscope 5 1 0 2 l u J 3 1 ] l l a h - s e m . t a m - d n o c [ 1 v 5 9 5 3 0 . 7 0 5 1 : v i X r a G4DG3G2G1S1S2SQPCDQPCGQPC500nmABC(b)SourceS1SourceS2ABCC,ABABtC,ADDGC,GCBCU,fAC1S2DrainU,fAC2CS2C,GS1BS1C,ACACt(a)path1path2UG3(meV)UG1(meV)507011050709000.02GB(e2/h)(c)ACBpath1UG1(meV)50709000.06GC(e2/h)(d)ACBpath2(AFM) on GaAs/AlGaAs-heterostructure [21 -- 23]. A two-dimensional electron gas with an electron concen- tration of ne = 3, 47 · 1015m−2 is located in 33 nm depth below the surface. The dots A,B,C are arranged in a triangular geometry [4] with each dot placed next to the other two and one lead attached to each dot (Fig.1 (a)). Dots A and B and also A and C are tunnel coupled, dots B and C are capacitively coupled only. The source leads S1 and S2 are connected to dots B and C respectively and dot A is connected to the drain lead D. We have two transport paths: path 1 with dots A and B and path 2 with dots A and C. The sample has four in-plane gates G1-G4 (Fig.1 (b)) to control the potential of the dots, interdot and dot-lead couplings. A quantum point contact (QPC) sensitive to all three dots is placed next to dots B and C to perform charge measurements. The measurements were conducted in a dilution refrigerator. To measure the differential conductance of the two transport paths simultaneously but separately, a lock-in technique was used with ac voltages with two different frequencies f1 = 83.3 Hz and f2 = 18.3 Hz with UAC = 10 µV applied to S1 and S2 respectively. The QPC was operated by applying a dc voltage to the source of the QPC, SQPC, and measuring a dc current at the drain of the QPC, DQPC. The QPC is tuned by the gate GQPC. In our transport measurement range, the dots contain several ten electrons on the whole. The charging energies are Ech,A = 2 meV, Ech,B = 6 meV, Ech,C = 3 meV for dot A, B and C respectively. Charge Measurements. To characterize the device, the charging is studied by using the QPC as a detector. The derivative of the QPC current is plotted as a function of gate voltages UG1 and UG3 (Fig.2) with denoted charge configurations NA, NB, NC(cid:105), where Ni are the occupations of dots A,B,C. The electrons in the core of the dots are not included in Ni. The green lines indicate charging events, where one more electron is added to the system, pink lines indicate electron movement away from the detector. Charging lines with three different slopes, one slope for each dot, are visible in the measurement, as the slope depends on the capacitive coupling and thus on the distance between the dot and gates G1 and G3. The lines with the lowest slope belong to dot C as it is the least coupled to UG1, the lines with intermediate slope to dot A and the lines with the largest slope to dot B as it is the least coupled to UG3. Anticrossings of two dots in resonance are visible where two charging lines meet. At such a resonance two triple points (TPs) emerge where at each one three different charge configurations are degenerate. Charge reconfiguration lines connecting the TPs mark the charge transitions between the dots. Resonances between dot A and B (green circle), A and C (blue circle) and also between the only capacitively coupled dots B and C located in the two different paths (black circle) are observed [24]. When the double dot 2 FIG. 2. Charge measurement using the QPC. The green lines with different slopes point out the charging of the respective dot with one electron. Double dot resonances are marked by circles in green, black and blue, the triple dot resonance by a red circle. anticrossings coincide, the resonance condition for all three dots is fulfilled (red circle). We will focus on such a region in the transport measurements. Transport Measurements. To understand transport in this system, the differential conductance G is measured along path 1 and 2 simultaneously but separately, sweep- ing gate voltages UG1 and UG3 (Fig.1 (c),(d)). In doing so, the QPC is not in use. TPs with finite differential conductance are observed in both paths where the two dots are in resonance. In path 1 (Fig.1 (c)) resonances of the dots A and B can be seen. Charging of dot C is observed by a shift of the charging lines of dot A and B, where dot C comes into resonance. Analogously, charging lines of dot A and C appear in path 2 (Fig.1 (d)) and charging of dot B is detected indirectly by the shift of the charging lines of dot A and C. A triple dot resonance is formed where two double dot resonances coincide. In Fig.3 (a) where we combine path 1 and path 2 as observed in Fig.1 (c),(d) we have three double dot resonances, A and B, A and C, B and C, in close vicinity to each other. One observes regions of high differential conductance in both paths but at different gate voltages UG1 and UG3. Along the B charging line (path 1) we can identify two resonance lines where A is resonant with B. The whole triple dot physics can be observed. The different occupations of the states at the TP participating in transport are marked in Fig.3. To understand the experimental results in more detail a theoretical model is developed which reproduces and explains the transport properties of the system. We fit UG3(meV)UG1(meV)−90−60−300−200−180−160−1400,0,0i0,0,1i1,0,1i1,1,1i0,1,1i0,1,0iABC+−3 0, 1, 0(cid:105) with one more electron in B becomes occupied. Below 1, 1, 1(cid:105), for smaller values of UG3, the occupation of 1, 1, 0(cid:105) with one less electron in C increases and at the left hand side, for smaller values of UG1, the state 1, 0, 1(cid:105) with one less electron in B is occupied. All these regions obtained numerically correspond perfectly to the ones in Fig.3 (a). The small regions of 1, 0, 0(cid:105) and 0, 1, 1(cid:105) connect the states with one electron and two electrons respectively. Each of these small regions, not really seen in Fig.3 (a), contain two TPs of path 1 and two TPs of path 2. When two TPs coincide we have a quadruple point. With the information from the theoretical calculation of the occupations we determine the TPs present on the resonant lines in path 1 (Fig.3 (a)), (0, 0, 0(cid:105), 0, 1, 0(cid:105), 1, 0, 0(cid:105)), (1, 1, 0(cid:105), 1, 0, 0(cid:105), 0, 1, 0(cid:105)) and (0, 0, 1(cid:105), 0, 1, 1(cid:105), 1, 0, 1(cid:105)), (1, 1, 1(cid:105), 1, 0, 1(cid:105), 0, 1, 1(cid:105)) where in the two last TP there is one more electron in C. In near vicinity of these TP there is high positive differential conductance in path 1 due to temperature broadening of the states. Thus they merge and form a vertical line of high differ- ential conductance. Similarly, for path 2 we have high positive differential conductance at the TPs (0, 0, 0(cid:105), 0, 0, 1(cid:105), 1, 0, 0(cid:105)), (1, 0, 1(cid:105), 0, 0, 1(cid:105), 1, 0, 0(cid:105)) and, with one more electron in dot B, (0, 1, 0(cid:105), 0, 1, 1(cid:105), 1, 1, 0(cid:105)), (1, 1, 1(cid:105), 0, 1, 1(cid:105), 1, 1, 0(cid:105)). In Fig.4 we plot the measured conductance of path 1 and 2 separately ((a),(b)) as well as the results from the simulation ((c),(d)). In path 1(2) we observe the splitting of the resonance between the dots A,B(A,C) due to the interaction with dot C(B). We also observe negative differential conductance in path 1 where path 2 has high conductance (grey color in Fig.4 (a),(c)). In the following we will analyze and compare the transport features of the two paths in more detail. Channel Blockade. In Fig.5 we show a cut from Fig.4. We observe that the resonance of path 2 splits into two ((cid:78),(cid:7)) due to the interaction with the third dot present in path 1. Path 1 gets in resonance at (cid:72) and partially blocks the other path decreasing its conductance. This point is a quadruple point, four states of the two paths are coexisting in the same region of the stability diagram. The transport through path 2 is stronger than through path 1 (τAC > τAB) thus when path 2 is in resonance ((cid:78),(cid:7)) it totally blocks path 1 decreasing its conductance even to negative values (more appreciable in Fig.4 (a) and 4 (c)). This channel blockade is a consequence of Coulomb interaction between the charges flowing through the two transport channels which share dot A. When the bias voltage increases, the path with higher conductance increases its occupation in dot A blocking access to dot A from the other path and thus decreasing its transport and obtaining negative differential conductance. In Fig.6 we identify the dominating and the blocked FIG. 3. (a) Combined color plot of differential conductance through path 1 and 2 with denoted charge configurations of the stability regions. (b) Calculated steady state occupation probabilities ρSS of each state i in the stability regions of the system (bottom). States 1, 0, 0(cid:105) and 0, 1, 1(cid:105) are a small region with low occupation probability. i the transport simulations to the experimental data to extract the interdot tunnel couplings τAB = 0.012 meV, τAC = 0.020 meV, the dot-lead tunnel couplings ΓD = 0.008 meV, ΓS1 = 0.003, ΓS2 = 0.006 meV and the electron temperature Tel = 300 mK. In the model we distinguish between particles coming from S1 and S2 in order to be as close as possible to the experimental conditions. Simulation. Using Master equation techniques (see Supplementary for more detail), we calculate the trans- port through the two paths and the stability regions of each state. In Fig.3 (b) we plot the numerical result of the steady state occupations ρSS . For small and large UG1 and UG3 the states 0, 0, 0(cid:105) and 1, 1, 1(cid:105) are occupied respectively. Above 0, 0, 0(cid:105), for larger values of UG3, the state 0, 0, 1(cid:105) with one more electron in C becomes occupied and at the right hand side, for larger values of UG1, the state i (a)path1UG3(meV)UG1(meV)9010011070809010000.15GB(e2/h)0,0,0�0,0,1�0,1,0�1,1,0�1,0,1�1,1,1�(a)path2UG3(meV)UG1(meV)9010011070809010000.06GC(e2/h)0,0,0�0,0,1�0,1,0�1,1,0�1,0,1�1,1,1�(b)901001108090ρSS000UG1(meV)8090ρSS0018090ρSS100UG3(meV)90100110ρSS101ρSS011ρSS110ρSS111809001NormalizedoccupationsρSS010((cid:78)) 0, 0, 1(cid:105) ((cid:72)) 0, 1, 1(cid:105) ((cid:7)) 0, 1, 1(cid:105) 1, 0, 0(cid:105) 0, 1, 1(cid:105) 1, 1, 0(cid:105) 1, 0, 1(cid:105) 1, 1, 0(cid:105) 1, 0, 1(cid:105) 4 1, 0, 1(cid:105) 1, 1, 1(cid:105) 1, 1, 1(cid:105) FIG. 6. Transport mechanism for the peaks in Fig.5. In ((cid:78), (cid:7)) path 2 blocks path 1 and in ((cid:72)) both paths share the occupation of dot A. crucial to have two sources. If we switch the transport direction of both paths with the source lead connected to dot A and the two drain leads connected to dots B and C, the transport paths influence each other in a different way in the simulations. The electron flow splits at dot A in two paths with a probability that depends on the tunneling rates of path 1 (τAB, ΓB) and path 2 (τAC, ΓC). Conclusion. In summary, we have shown channel blockade in electronic transport through a TQD with two source leads. Coulomb interaction between electrons coming from the two sources gives rise to a blockade of transport through one path, when the other path has high conductance. The results show how interaction be- tween the joint transport paths is affecting the transport properties of the multi-terminal device and are a step towards a better understanding of transport properties in complex multi-dot systems. We are grateful to M. C. Rogge for producing the TQD sample. We acknowledge discussions with R. S´anchez and financial support from Spanish MICINN MAT2014- 58241-P. ∗ kotzian@nano.uni-hannover.de [1] A. Vidan, R. Westervelt, M. Stopa, M. Hanson, and A. Gossard, Appl. Phys. Lett. 85, 3602 (2004) [2] L. Gaudreau, S. A. Studenikin, A. S. Sachrajda, P. Za- wadzki, A. Kam, J. Lapointe, M. Korkusinski, and P. Hawrylak, Phys. Rev. Lett. 97, 036807 (2006). [3] D. Schroer, A. D. Greentree, L. Gaudreau, K. Eberl, L. C. L. Hollenberg, J. P. Kotthaus, and S. Ludwig, Phys. Rev. B 76, 075306 (2007). [4] M. C. Rogge and R. J. Haug, Phys. Rev. B 77, 193306 FIG. 4. Differential conductance in experiment along path 1 (a) and path 2 (b) and in simulation along path 1 (c) and path 2 (d). The dotted line is the cut plotted in Fig.5. FIG. 5. Cut through the transport measurement and simula- tion of Fig.4 for path 1 and 2 at UG3=103 mV. transport channels for each resonance ((cid:78),(cid:72),(cid:7)) with the information from the simulation. We show the initial and final states connected by two different transport paths, where in ((cid:78),(cid:7)) one path blocks the other and in ((cid:72)) both paths share the conductance. Blockade phenomena were previously studied for one dot attached to three leads (one drain and two sources) which contain some amount of up and down spins [25]. The spins of each path compete to occupy the dot blocking the access to the dot for the other spin by Coulomb interaction. Compared to this work for a single dot our TQD includes coherences between states of the two paths. Other papers [26 -- 28] treat two transport paths just capacitively coupled where at some situations one of the paths blocks the transport through the other by Coulomb interaction. To observe the channel blockade in our experiment, it is (2008). path1path2UG3(meV)9011000.015GB(e2/h)(a)00.06GC(e2/h)(b)UG3(meV)UG1(meV)90110708090(c)UG1(meV)8090(d)00.020.040.068085Experiment8085TheoryG(e2/h)UG1(meV)path1path2NH(cid:7)UG1(meV)NH(cid:7)5 [5] B. Michaelis, C. Emary, and C. Beenakker, Europhys. Lett. 73, 677 (2006) [16] M. Stopa, Phys. Rev. Lett. 88, 146802 (2002). [17] C. Lent, P. Tougaw, W. Porod, and G. Bernstein, Nan- [6] C. Poltl, C. Emary, and T. Brandes, Phys. Rev. B 80, otechnology 4, 49 (1993). 115313 (2009). [18] M. C. Rogge and R. J. Haug, New Journal of Physics 11, [7] M. Busl, R. S´anchez, and G. Platero, Phys. Rev. B 81, 113037 (2009). 121306 (2010). [8] C. Emary, Phys. Rev. B 76, 245319 (2007). [9] R. S´anchez, F. Gallego-Marcos, and G. Platero, Phys. Rev. B 89, 161402(R) (2014). [19] A. D. Greentree, J. H. Cole, A. R. Hamilton, and L. C. L. Hollenberg, Phys. Rev. B 70, 235317 (2004). [20] H. Sasakura, S. Adachi, S. Muto, T. Usuki, and M. Takatsu, Semicond. Sci Technol. 19, S409 (2004). [10] F. Gallego-Marcos, R. S´anchez, and G. Platero, J. Appl. [21] M. Ishii and K. Matsumoto, Jpn J. Appl. Phys. 34, 1329 Phys. 117, 112808 (2015). (1995). [11] M. Busl, G. Granger, L. Gaudreau, R. S´anchez, A. Kam, M. Pioro-Ladri`ere, S. A. Studenikin, P. Zawadzki, Z. R. Wasilewski, A. S. Sachrajda, and G. Platero, Nature Nan- otechnology 8, 261 (2013). [12] F. R. Braakman, P. Barthelemy, C. Reichl, W. Wegschei- der, and L. M. K. Vandersypen, Nature Nanotechnology 8, 432 (2013). [13] R. S´anchez, G. Granger, L. Gaudreau, A. Kam, M. Pioro- Ladri`ere, S. A. Studenikin, P. Zawadzki, A. S. Sachrajda, and G. Platero, Phys. Rev. Lett. 112, 176803 (2014). [14] L. Gaudreau, G. Granger, A. Kam, G. C. Aers, S. A. Stu- denikin, P. Zawadzki, M. Pioro-Ladri`ere, Z. R. Wasilewski, and A. S. Sachrajda, Nature Physics 8, 54 (2012). [22] R. Held, T. Vancura, T. Heinzel, K. Ensslin, H. Holland, and W. Wegschneider, Appl. Phys. Lett. 73, 2 (1998). [23] U. Keyser, H. Schumacher, U. Zeitler, R. Haug, and K. Eberl, Appl. Phys. Lett. 76, 4 (2000). [24] J. M. Elzerman, R. Hanson, J. S. Greidanus, L. H. Willems van Beveren, S. De Franceschi, L. M. K. Vander- sypen, S. Tarucha, and L. P. Kouwenhoven, Phys. Rev. B 67, 161308(R) (2003). [25] A. Cottet, W. Belzig, and C. Bruder, Phys. Rev. B 70, 115315 (2004). [26] R. Hussein and S. Kohler, Phys. Rev. B 89, 205424 (2014). [27] C. Nietner, G. Schaller, C. Poltl, and T. Brandes, Phys. [15] D. DiVincenzo, D. Bacon, J. Kempe, G. Burkard, and K. Rev. B 85, 245431 (2012). Whaley, Nature 408, 339 (2000). [28] G. Schaller, G. Kiesslich, and T. Brandes, Phys. Rev. B 82, 041303 (2010). Supplemental Material: Theoretical Model † i τi,i+1c i ci+1 +(cid:80) H = H0 + Hlead + Hint. The dot system ( H0) is a three site Anderson-like Hamiltonian [1]: H0 = (cid:80) (cid:80) Hlead =(cid:80) part of the Hamiltonian Hint =(cid:80) Here we discuss in detail the theory used to simulate the experiment. The total Hamiltonian of the system reads † i
ic i ci + † i<j Vij ni nj, where c i is the electron creation operator and ni the particle number operator of dot i.
i (i = {A, B, C}) is the chemical potential of the dots, τij the coherent interdot tunnel coupling and Vij the Coulomb interaction between the electrons in different dots. The reservoirs are modeled as a Fermi electron gas dlk that has a constant temperature T and chemical potential µl (l = {S1, S2, D}). The interaction † li γl d l ci + H.c. couples the reservoirs and the dots with a hopping parameter γl. † lk εlk d lk The energy levels are tuned with the gate voltages present in the experiment (UG1, UG2, UG3, UG4). The rates between the leads and the dots for incoming (+) and outgoing (−) electrons with respect to the dot l←i = 2π/γl2[1 − f (µl −
i)] where f system are given by Fermi's golden rule Γ(+) is the Fermi distribution function. Γl ≡ 2π/γl2 is smaller than the interdot coupling τij, thus we can apply the Born-Markov approximation [2] for the interaction of the system with the leads. From the Von Neumann equation ∂t
(t) = i/[ H,
(t)], which contains the full system time evolution, we trace over the baths degrees of freedom getting the reduce density matrix ρ(t) = Trleads
(t) [3] obtaining the master equation i←l = 2π/γl2f (µl −
i) and Γ(−) (cid:88) j ∂tρi(t) = Lijρj(t). (1) ρi(t) is the occupation probability of the i-state of the system and L is the Liouvillian superoperator that contains all the information about the system H0 and the jumping terms between the leads and the dots Γi↔l. As we just want to study the steady state properties of the system we solve the kernel of Eq. (1) to obtain the steady state occupations ρSS = Ker[L]. Taking the steady state occupations and the tunneling rates to/from the contacts we are able to calculate the current 1(cid:88) i,j=0 I = ρSS1,i,j(cid:105)Γ(−) 0,i,j(cid:105)←1,i,j(cid:105) − ρSS0,i,j(cid:105)Γ(+) 1,i,j(cid:105)←0,i,j(cid:105). (2) ∗ kotzian@nano.uni-hannover.de [1] P. W. Anderson, Physical Review 124, 41 (1961). [2] H. P. Breuer and F. Petruccione, The Theory of Open Quantum Systems (Oxford University Press, Oxford, 2002). [3] T. H. Stoof and Y. V. Nazarov, Phys. Rev. B 53, 1050 (1996). 6 

1311.0064

1

1311

Screening Charged Impurities and Lifting the Orbital Degeneracy in Graphene by Populating Landau Levels

We report the observation of an isolated charged impurity in graphene and present direct evidence of the close connection between the screening properties of a 2D electron system and the influence of the impurity on its electronic environment. Using scanning tunneling microscopy and Landau level spectroscopy we demonstrate that in the presence of a magnetic field the strength of the impurity can be tuned by controlling the occupation of Landau-level states with a gate-voltage. At low occupation the impurity is screened becoming essentially invisible. Screening diminishes as states are filled until, for fully occupied Landau-levels, the unscreened impurity significantly perturbs the spectrum in its vicinity. In this regime we report the first observation of Landau-level splitting into discrete states due to lifting the orbital degeneracy.

cond-mat.mes-hall

cond-mat

Screening Charged Impurities and Lifting the Orbital Degeneracy in Graphene by Populating Landau Levels Adina Luican-Mayer,1 Maxim Kharitonov,1 Guohong Li,1 ChihPin Lu,1 Ivan Skachko,1 Alem-Mar B. Goncalves,1 K. Watanabe,2 T. Taniguchi,2 and Eva Y. Andrei1 1Department of Physics and Astronomy, Rutgers University, Piscataway, NJ 08854, USA 2Advanced Materials Laboratory, National Institute for Materials Science, 1-1 Namiki, Tsukuba, 305-0044, Japan (Dated: April 11, 2018) Abstract We report the observation of an isolated charged impurity in graphene and present direct evidence of the close connection between the screening properties of a 2D electron system and the influence of the impurity on its electronic environment. Using scanning tunneling microscopy and Landau level spectroscopy we demonstrate that in the presence of a magnetic field the strength of the impurity can be tuned by controlling the occupation of Landau-level states with a gate-voltage. At low occupation the impurity is screened becoming essentially invisible. Screening diminishes as states are filled until, for fully occupied Landau-levels, the unscreened impurity significantly perturbs the spectrum in its vicinity. In this regime we report the first observation of Landau-level splitting into discrete states due to lifting the orbital degeneracy. 3 1 0 2 t c O 1 3 ] l l a h - s e m . t a m - d n o c [ 1 v 4 6 0 0 . 1 1 3 1 : v i X r a 1 Charged-impurities are the primary source of disorder and scattering in two dimensional (2D) electron systems[1]. They produce a spatially localized signature in the density of states (DOS) which, for impurities located at the surface, is readily observed with scan- ning tunneling microscopy and spectroscopy (STM/STS)[2]. In this respect graphene[3][4] [5], [6], with its electronic states strictly at the surface, provides a unique playground for elucidating the role of impurities in 2D [6] [7][8]. They are particularly important in the presence of a magnetic field when the quantization of the 2D electronic spectrum into highly degenerate Landau levels (LL) gives rise to the quantum Hall effect (QHE). In this regime charged-impurities are expected to lift the orbital-degeneracy causing each LL in their im- mediate vicinity to split into discrete sub-levels [9]. Thus far however the sub-levels were not experimentally accessible due to the difficulty to attain sufficiently clean samples that would allow isolating a single impurity. Instead, previous experiments [10],[11] presented a picture of bent levels which could be interpreted in a semi-classical framework in terms of electronic drift trajectories moving along the equipotential lines defined by a dense distribution of charged-impurities. In this work we employed high quality gated graphene devices which provided access to the electronic spectrum in the QHE regime in the presence of an isolated charged-impurity. We demonstrate that the strength of the impurity, as measured by its effect on the spectrum, can be controlled by tuning the LL occupation with a back-gate-voltage. For almost empty LLs the impurity is screened and essentially invisible whereas at full LL occupancy screening is very weak and the impurity attains maximum strength. In the unscreened regime we experimentally resolve the underlying discrete quantum-mechanical spectrum arising from lifting the orbital-degeneracy. The low energy spectrum of pristine graphene, consisting of two electron-hole symmetric Dirac cones, gives rise to a linear DOS, which vanishes at the charge neutrality point (CNP). In the presence of a magnetic field B, the spectrum is quantized into a sequence of LLs characteristic of massless Dirac fermions: EN = ± vF lB p2N, N = 0, ±1, ±2, ... (1) where vF is the Fermi velocity, lB = p¯h/eB the magnetic length, e the electron charge, ¯h the reduced Planck constant and ± refers to electron (hole) states with LL index N > 1(N < 1). We employed LL spectroscopy [5] to study the electronic properties of graphene 2 and their modification in the presence of a charged-impurity. The LL spectra were obtained by measuring the bias voltage dependence of the differential tunneling conductance, dI/dV , which is proportional to the DOS, DOS(E, r) , at the tip position r. Here V = (E − EF )/e is the bias voltage and E the energy measured relative to the Fermi level, EF . Samples were prepared by exfoliating graphene from analyzer-grade HOPG and deposited on a doped Si back-gate capped with 300nm of chlorinated SiO2[12]. In order to achieve high quality we used two superposed graphene layers twisted away from the standard Bernal stacking by a large angle. The large twist angle ensures that the spectrum of single layer graphene is preserved [13] [14] while reducing the random potential fluctuations due to substrate imperfections.Hexagonal boron-nitride (h-BN) flakes which significantly reduce the corrugation of graphene[15] were also employed (Figure 1a), but the data reported is restricted to the SiO2 substrate. Using the STM tip as a capacitive antenna [16] we located the samples at low temperature and performed spectroscopy measurements to identify areas of interest. A typical zero-field spectrum taken far from an impurity in Figure 1b reveals the V-shaped DOS characteristic of single layer graphene (SLG). In finite-field the spectrum develops pronounced peaks (Figure 1c) at energies corresponding to the LLs [17] that are well resolved up to N = 4 in both electron and hole sectors, attesting to good sample quality. Fitting the field and level index dependence to Equation 1 confirms the massless Dirac fermion nature of the quasiparticles with vF = 1.2 × 106m/s, a value consistent with measurements on SLG. Charged-impurities were located using LL-spectroscopy to measure the separation between EF and the CNP which coincides with the N = 0 level. LL-spectroscopy is more sensitive to the position of the CNP than the broad zero-field spectrum. The search for impurities starts with a topography image (Figure 1d) followed by STS within this area. An intensity map of the LLs as a function of position (Figure 1e) shows the fluctuations of the N = 0 level in response to charged-impurities. We focus on an area with a minimal number of impurities indicated by the arrow. Zooming into this area the impurity appears as an isolated bright region in the center of Figure 1f. To visualize its effect on the spatial distribution of the electronic-wavefunction we measured constant energy DOS maps (Figure 2). The maps are roughly radially symmetric, consistent with a charged-impurity at the center. We note that for energies within a gap between LLs the electronic DOS (bright region) is tightly localized on the impurity. In contrast, for energies in the center of the LL the electronic DOS extends 3 !"#$ !%#$ !'#$ N=0 +1 +2 +4 +3 !&#$ B=10T -4 -3 -2 -1 ) . . u ) . . u a b ( r V a ( V d d / / I I d d 4 2 0 -200 0 200 Sample Bias (mV) !(#$ ) 200 -200 0 200 Sample Bias (mV) ) . u a ( . V d / I d 2 1 0 !)#$ V m ( s a B e p m a S i l N=+1 0 N=0 N= -1 -200 0 100 200 300 400 Distance (nm) FIG. 1. (a) Schematics of gated graphene device illustrating the regions where the two stacked graphene layers are deposited directly on SiO2 and on a flake of h-BN in ( G/G/SiO2 and G/G/BN ). The two graphene layers share the same electrode and Fermi level. (b) Zero field STS taken far from the impurity shown in panel d. (c) STS at B = 10T and Vg = 0V shows well resolved quantized LLs. Red circles indicate bias voltages of the maps in Figure 2. (d) Large area topographic image indicating the line where the spectra in panel e were taken and the position of the isolated impurity. (e) STS line cut along the line shown in panel d; LLs with indices N = 0, are clearly resolved. (f) STM topography zoom-into the area with an isolated impurity (VB = 250mV, It = 20pA). The spectra in panels b, c and the LL map in Figure 3 were taken at the position indicated by the dot. 4 -!"#$% &'"#$% "#$% '"#$% !"#$% ("#$% )"#$% *""#$% FIG. 2. Spatial dI/dV maps at B = 10T near the impurity taken at indicated bias voltages. Scale bar for all maps: 8.2nm = lB . across the entire field of view while avoiding the impurity. This fully supports the picture which attributes the QHE plateaus to the existence of localized impurity states in the gaps between LLs. We next studied the effect of LL occupancy (filling) by tuning the gate-voltage, Vg , to progressively fill the LLs. The LL filling factor is ν = n/n0(B) where n ≈ 7×1010Vg(V )cm−2 is the carrier density and n0(B) = glgvgs Be 2π¯h is the degeneracy/area of the LL. Here gl = gv = gs = 2 represent the layer, valley and spin degeneracy respectively. Placing the STM tip far from the impurity we find that as Vg is swept the LL peaks produce a distinctive step-like pattern seen as bright traces in the intensity map of Figure 3a [12, 18]. Each step consists of a nearly horizontal plateau separated from its neighbors by steep slopes. The separation between the centers of steep segments, ∆Vg ≈ 28V , gives the LL degeneracy ≈ 2 × 1012cm−2 for B = 10T , as expected for this double-layer sample [19]. The plateau indicates that the Fermi-energy remains pinned within a narrow energy band around the center of the LL until the plateau states are filled. A further increase in Vg populates the sparse states in the gap producing the steep slopes[20]. 5 To explore the influence of the impurity on the LLs we follow the spatial evolution of spectra along a trajectory traversing it (Figure 1f) for a series of gate-voltages. As shown in Figure 3b, for certain gate-voltages the spectra become significantly distorted close to the impurity, with the N = 0 level (and to a lesser extent higher order levels) shifting downwards toward negative energies. The downshift indicates an attractive potential produced by a positively charged-impurity. Its strength, as measured by the distortion of the N = 0 LL, reveals a surprisingly strong dependence on LL filling. In the range of gate-voltages corresponding to filling the N = 0 LL (−15V < Vg < 9V ) the distortion grows monotonically with filling. At small filling the distortion is almost absent indicating that the impurity is effectively screened and it reaches its maximum value close to full occupancy. At full occupancy the N = 0 level shifts by as much as ≈ 0.1eV indicating that the effect would survive at room temperature. We note that this spectral distortion is only present in the immediate vicinity of the impurity. Farther away no distortion is observed for all the carrier densities studied here. We attribute the variation of the impurity strength with filling to the screening properties of the electron system. For a positively (negatively) charged-impurity and almost empty (full) LLs, unoccupied states necessary for virtual electron transitions are readily available in the vicinity of the impurity, resulting in substantial screening. By contrast for almost filled (empty) LLs, unoccupied states are scarce, which renders local screening inefficient. These properties are readily understood by examining the local DOS, Ds(E, r) = R d2rD(E, r)/S, averaged over a finite-size region, S, around the impurity. Unlike the DOS averaged over the whole sample, Ds(E, r), is manifestly particle-hole asymmetric within a given LL which translates to the particle-hole asymmetry of the local screening. Remarkably, when screening is minimal (Vg = +7V ) the N = 0 LL does not shift smoothly, but rather splits into a series of well resolved discrete spectral lines in the imme- diate vicinity of the impurity. As shown in Figure 4a,b the evolution of the spectra radially outwards from the center of the impurity exhibits a progression of peaks within the N = 0 LL. Starting with a single peak at the center of the impurity, it evolves into a well resolved double peak and then a triplet at distances ≈ 13nm, 20nm from the center respectively. This behavior can be understood by considering the quantum-mechanical electron motion in the presence of a magnetic-field and a charged-impurity. In one valley and for each spin projection, the two-component wave-function ψ = (ψA, ψB)T satisfies an effective Dirac 6 Hamiltonian: Hψ = Eψ, H = H0 + U(r), H0 = ¯hvF σ(p − eA) (2) Here, H0 corresponds to the case without the impurity, σ = (σx, σy) are the Pauli ma- trices in the sublattice space, p = −i¯h∇, B = ∇ × A , and A is the vector potential. We assume a radially symmetric impurity-potential U(r), and neglect the Zeeman-effect. In the symmetric-gauge A = 1 2 [B × r] the eigenstates are characterized by the orbital quantum- number m. Solving H0 yields the unperturbed spectrum in Equation 1, and the eigenfunc- tions ψ0 N mB(r) ( Figure 4c) where m ≥ − N . Since EN are independent of m, the LLs have infinite orbital-degeneracy. The impurity lifts this orbital-degeneracy and the N mA(r), ψ0 eigenenergies split into series of sublevels, EN m. To illustrate impurity-induced orbital-splitting we numerically solved the problem for a 1√r2+a2 corresponding to a charge Z located a distance a below the graphene plane with κ the effective dielectric constant and ǫ0 the permitivity of Coulomb potential U(r) = Z κ e2 4πǫ0 free-space. The resulting simulated-spectrum in the left panel of Figure 4d, shows that the orbital-degeneracy is lifted resulting in an m dependent energy downshift. The downshift is largest for E00, and diminishes with increasing m and/or N where the unperturbed LLs are approached. For comparison with the STS data we calculated the local tunneling DOS assuming a finite linewidth γ : D(E, r) = 4 X δγ(E − EN m)ψ†N m(r)ψN m(r) (3) N mi Here i is the sublattice index and δγ(E − EN m) = γ/[π((E − EN m)2 + γ2)] represents the broadened LL. The peak intensity is determined by the probability-density ψ†N m(r)ψN m(r) If γ < ∆EN m (∆EN m spacing between adjacent levels) the and is position dependent. discreteness of the spectrum is resolved, but for γ ≥ ∆EN m (∆EN m peaks of adjacent states overlap and merge into a continuous band. Thus, even if the spectrum is discrete, but the resolution insufficient or if impurities are too close to each other, the measured D(E, r) will still display "bent" LLs, whose energies seemingly adjust to the local potential. The resulting simulated D(E, r) , shown in Figure 4d (right-panel), captures the main features of the data. In particular, upon approaching the impurity the N = 0 LL splits into well resolved discrete peaks, attributed to specific orbital states. In both experiment and simulation the states ψ0m(r) with m = 0, 1, 2 are well 7 FIG. 3. Impurity screening by populating Landau levels. (a) DOS map at B = 10T showing evolution of LLs as a function of gate-voltage taken far from the impurity at the position indicated in Figure 1f. Dashed lines indicate gate-voltages at which the spectra in b were taken. (b) LL maps across the impurity for indicated gate-voltages. The distortion of the LL sequence by the impurity is strongest for filled levels (Vg = +7V ), diminishing as filling is reduced and becoming almost invisible at Vg = −10V . 8 resolved close to the impurity but higher order states, are less affected and their contributions to D(E, r) merge into a continuous line. Similarly, the discreteness of the spectrum is not resolved for N 6= 0, consistent with the weaker impurity effect at larger distances. We note that for partial filling (Vg = −5V, 0V ) as screening becomes more efficient and orbital- splitting is no longer observed the unresolved sublevels merge into continuous lines of "bent" Landau levels (Figure 3b). Thus the capability to tune the strength of the impurity potential by the gate-voltage allows us to trace the evolution between the discrete and the previously observed quasi-continuous regimes [11]. We now turn to the effective charge of the impurity and how its screening is affected by LL occupancy. Although we do not have in-situ chemical characterization of the impurity we can assume based on chemical-analysis of similar samples that the most likely candidate consistent with our observations is an Na+ ion adsorbed on the surface of the SiO2 sub- strate [21]. It is well known that Na+ ions are ubiquitous in cleanrooms and laboratory environments and they are readily adsorbed on SiO2 [22]. Levels of Na+ contamination as high as 1012cm−2 can be reached within just a few days of exposure to human activity. For a Na+ ion adsorbed on the substrate underneath the first graphene layer a ≈ 0.6nm ≪ lB. A rough estimate of its effect on the spectrum can be obtained by equating the measured energy shift ∆E00 ≈ 0.1eV , obtained at Vg = +7V to the calculated value in first order perturbation theory: e2 4πǫ0 e2 hψ00 ∆E00 = Z κ where Erf (x)x≪1 → 0 is the error function. Using ∆E00 ≈ Z 1√r2+a2 ψ00i = Z (π/2)(1/2)(1 − Erf (a/lB)) κ 4πǫ0lB e2 4πǫ0lB (π/2)1/2 = 0.1eV we κ obtain Z/κ = 2.5 . Factoring out the contribution of the SiO2 substrate, κSiO2 = 4, from ∼= 1 the expression for the effective dielectric constant [23] κ = κgr(κSiO2 + 1)/2, gives Z/κgr where κgr is the static dielectric constant of graphene. This indicates that graphene provides no screening for gate-voltages corresponding to fully occupied LLs and that bare charge of the impurity is Z = +1. In the limit of almost empty LLs (Vg = −10V ) the estimated κgr ≈ 5 indicates that graphene strongly screens the impurity potential [24]. The absence of screening for filled states implies strong Coulomb interactions when EF lies in a gap consistent with the observation of a fractional QHE in suspended graphene [25] . This work demonstrates that screening in graphene is controlled by LL occupancy and that it is possible to tune charge-impurities and their effect on the environment by applying a gate voltage or by varying the magnetic field. In the limit of strong screening, corresponding 9 "#$! N=0 m=0 A N=+1 N=+2 N=+3 "%$! 0 300 300 Position (l ) Position (l ) B B 2 4 6 8 10 10 ) ) 200 200 i i V m ( s a B e p m a S V m ( s a B e p m a S l l 100 100 0 -100 -100 -200 -200 -300 -300 !! ψ02 ψ01 ψ00 CB A ) . u a ( . V d / I d B N=0 m=0 m=1 C N=0 m=0 m=1 m=2 -150 0 150 300 Sample Bias (mV) "&$! "'$! FIG. 4. Lifting the orbital-degeneracy. (a) dI/dV spectra for B = 10T and Vg = 7V at the positions indicated in panel b reveal the appearance of peaks corresponding to states with N = 0 and m = 0, m = 0, 1 and m = 0, 1, 2 as marked. (b). Top: LL map across the impurity for Vg = 7V . Dashed lines at distances 0nm, 13nm and 20nm from the center of the impurity indicate the position of the spectra in panel a. Bottom: calculated probability densities for states ψ0m with m = 0, 1, 2 are consistent with the spatial distribution of the discrete spectral lines in the top panel. (c) Calculated probability densities on the two graphene sublattice A (blue) and B (red). (d) Left panel: simulated spectrum near an impurity illustrating lifting the orbital-degeneracy in different LLs. Right panel: simulated DOS near an impurity. Linewidth γ = 0.05vF /lB . Red lines represent the calculated energies, EN m shown in the left panel. 10 to low LL occupancy a positive impurity is essentially invisible, while for high LL occupancy it is unscreened and lifts the orbital-degeneracy of the LLs in its vicinity. Due to the large enhancement of the effective fine-structure constant in graphene a charged-impurity with Z ≥ 1 is expected to become supercritical[3][8] but tuning the effective charge to observe supercritically is extremely difficult[26]. The ability demonstrated here to tune the strength of the impurity in-situ opens the door to exploring Coulomb criticality and to investigate a hitherto inaccessible regime of criticality in the presence of a magnetic field [9]. Funding was provided by DOE-FG02-99ER45742 (E.Y.A and G.L), Lucent (A.L-M), NSF DMR 1207108 (I.S., C.P.L) , DOE DE-FG02-99ER45790 (M. K.), Brazilian agency Capes BEX 5115-09-4 ( A.M.B.G). We wish to thank M. Aronson for the monochromator grade HOPG crystal. [1] T. Ando, A. B. Fowler, and F. Stern, Rev. Mod. Phys. 54, 437 (1982). [2] H. Mizes and J. Foster, Science 244, 559 (1989). [3] A. H. Castro Neto, F. Guinea, N. M. R. Peres, K. S. Novoselov, and A. K. Geim, Rev. Mod. Phys. 81, 109 (2009). [4] D. Abergel, V. Apalkov, J. Berashevich, K. Ziegler, and T. Chakraborty, Advances in Physics 59, 261 (2010); M. Morgenstern, physica status solidi (b) 248, 2423 (2011). [5] E. Y. Andrei, G. Li, and X. Du, Reports on Progress in Physics 75, 056501 (2012). [6] V. N. Kotov, B. Uchoa, V. M. Pereira, F. Guinea, and A. H. Castro Neto, Rev. Mod. Phys. 84, 1067 (2012). [7] D. V. Khveshchenko, Phys. Rev. B 74, 161402 (2006); J. Chen, C. Jang, S. Adam, M. Fuhrer, E. Williams, and M. Ishigami, Nature Physics 4, 377 (2008); Y. Zhang, V. Brar, C. Girit, A. Zettl, and M. Crommie, ibid. 5, 722 (2009); T. O. Wehling, S. Yuan, A. I. Lichten- stein, A. K. Geim, and M. I. Katsnelson, Phys. Rev. Lett. 105, 056802 (2010); C. Bena, Phys. Rev. B 81, 045409 (2010); S. Das Sarma, S. Adam, E. H. Hwang, and E. Rossi, Rev. Mod. Phys. 83, 407 (2011). [8] V. M. Pereira, J. Nilsson, and A. H. Castro Neto, Phys. Rev. Lett. 99, 166802 (2007); A. V. Shytov, M. I. Katsnelson, and L. S. Levitov, Phys. Rev. Lett. 99, 246802 (2007). [9] O. V. Gamayun, E. V. Gorbar, and V. P. Gusynin, Phys. Rev. B 83, 235104 (2011); 11 Y. Zhang, Y. Barlas, and K. Yang, Phys. Rev. B 85, 165423 (2012). [10] A. Yacoby, H. Hess, T. Fulton, L. Pfeiffer, and K. West, Solid state communications 111, 1 (1999); K. Hashimoto, C. Sohrmann, J. Wiebe, T. Inaoka, F. Meier, Y. Hirayama, R. A. Romer, R. Wiesendanger, and M. Morgenstern, Phys. Rev. Lett. 101, 256802 (2008). [11] D. Yoshioka, J. Phys. Soc. Jpn. 76, 024718 (2007); Y. Niimi, H. Kambara, and H. Fukuyama, Phys. Rev. Lett. 102, 026803 (2009); D. Miller, K. Kubista, G. Rutter, M. Ruan, W. de Heer, M. Kindermann, P. First, and J. Stroscio, Nature Physics 6, 811 (2010); M. Morgenstern, A. Georgi, C. Strasser, C. Ast, S. Becker, and M. Liebmann, Physica E: Low-dimensional Systems and Nanostructures 44, 1795 (2012). [12] A. Luican, G. Li, and E. Y. Andrei, Phys. Rev. B 83, 041405 (2011). [13] J. C. Meyer, A. Geim, M. Katsnelson, K. Novoselov, T. Booth, and S. Roth, Nature 446, 60 (2007); G. Li, A. Luican, J. L. Dos Santos, A. C. Neto, A. Reina, J. Kong, and E. Andrei, Nature Physics 6, 109 (2009); A. Luican, G. Li, A. Reina, J. Kong, R. R. Nair, K. S. Novoselov, A. K. Geim, and E. Y. Andrei, Phys. Rev. Lett. 106, 126802 (2011). [14] As shown by G. Li et al. Nat. Phys. 6,109, (2009), a twist between superposed graphene layers gives rise to two peaks in the density of states (Van-Hove singularities) which flank the charge neutrality point and are separated from each other by an energy which increases with twist-angle. For twist-angles exceeding 10 degrees the low energy spectrum (< 1eV ) is indistinguishable from that of single layer graphene. The absence of Van-Hove singularities and the single layer LL spectrum in the data reported here provide direct evidence of layer decoupling. Although there is no topographic signature of the associated Moire pattern, which would require a very sharp tip, the above signatures are taken as evidence for a large twist angle. [15] J. Xue, J. Sanchez-Yamagishi, D. Bulmash, P. Jacquod, A. Deshpande, K. Watanabe, T. Taniguchi, P. Jarillo-Herrero, and B. J. LeRoy, Nature materials 10, 282 (2011). [16] G. Li, A. Luican, and E. Y. Andrei, Review of Scientific Instruments 82, 073701 (2011). [17] G. Li and E. Andrei, Nature Physics 3, 623 (2007); G. Li, A. Luican, and E. Y. Andrei, Phys. Rev. Lett. 102, 176804 (2009). [18] O. Dial, R. Ashoori, L. Pfeiffer, and K. West, Nature 448, 176 (2007). [19] D. S. Lee, C. Riedl, T. Beringer, A. H. Castro Neto, K. von Klitzing, U. Starke, and J. H. Smet, Phys. Rev. Lett. 107, 216602 (2011); J. D. Sanchez-Yamagishi, T. Taychatanapat, K. Watan- 12 abe, T. Taniguchi, A. Yacoby, and P. Jarillo-Herrero, Phys. Rev. Lett. 108, 076601 (2012). [20] Although STM explores only a small area of the sample, the gate-voltage dependence of the data in Figure 3a reflects the available states in the entire sample including those that are outside the field of view of the STM. This is because the gate covers the entire sample and can populate all available states. [21] I. Constant, F. Tardif, and J. Derrien, Semiconductor science and technology 15, 61 (2000). [22] J. Fripiat, J. Chaussidon, and A. Jelli, Chimie-physique des ph´enom`enes de surface: applica- tions aux oxydes et aux silicates (Masson, 1971). [23] E. H. Hwang and S. Das Sarma, Phys. Rev. B 75, 205418 (2007). [24] This value is comparable to the zero field RPA estimate from: E. H. Hwang and S. Das Sarma, Phys. Rev. B 75, 205418 (2007) for double layer graphene κgr ≈ 2.4,κgr = 1 + glgsgvπrs/8 ≈ 3.75 suggesting that when the LLs are almost empty screening of positive charges in graphene is not very different from the zero field case. Here rs = 4πe2/hvF (κSiO2 +1) is the dimensionless Wigner-Seitz radius which measures the relative strength of the potential and kinetic energies in an interacting quantum Coulomb system with linear dispersion. We note that for single layer graphene, gl = 1, screening would be significantly weaker, κgr ≈ 2.4. [25] X. Du, I. Skachko, F. Duerr, A. Luican, and E. Y. Andrei, Nature 462, 192 (2009); K. I. Bolotin, F. Ghahari, M. D. Shulman, H. L. Stormer, and P. Kim, ibid. 462, 196 (2009). [26] Y. Wang, V. W. Brar, A. V. Shytov, Q. Wu, W. Regan, H.-Z. Tsai, A. Zettl, L. S. Levitov, and M. F. Crommie, Nature Physics 8, 653 (2012); Y. Wang, D. Wong, A. V. Shytov, V. W. Brar, S. Choi, Q. Wu, H.-Z. Tsai, W. Regan, A. Zettl, R. K. Kawakami, et al., Science 340, 734 (2013). 13 

1208.4184

1

1208

An investigation into the feasibility of myoglobin-based single-electron transistors

Myoglobin single-electron transistors were investigated using nanometer- gap platinum electrodes fabricated by electromigration at cryogenic temperatures. Apomyoglobin (myoglobin without heme group) was used as a reference. The results suggest single electron transport is mediated by resonant tunneling with the electronic and vibrational levels of the heme group in a single protein. They also represent a proof-of-principle that proteins with redox centers across nanometer-gap electrodes can be utilized to fabricate single-electron transistors. The protein orientation and conformation may significantly affect the conductance of these devices. Future improvements in device reproducibility and yield will require control of these factors.

cond-mat.mes-hall

cond-mat

2 An investigation into the feasibility of 1 0 2 myoglobin-based single-electron transistors Debin Li1, Peter M. Gannett2, and David Lederman1 1 Department of Physics, West Virginia University, Morgantown, WV, 26506-6315, USA 2 Department of Basic Pharmaceutical Sciences, West Virginia University, Morgantown, WV, 26506-9500, USA E-mail: david.lederman@mail.wvu.edu Abstract. Myoglobin single-electron transistors were investigated using nanometer- gap platinum electrodes fabricated by electromigration at cryogenic temperatures. Apomyoglobin (myoglobin without heme group) was used as a reference. The results suggest single electron transport is mediated by resonant tunneling with the electronic and vibrational levels of the heme group in a single protein. They also represent a proof-of-principle that proteins with redox centers across nanometer-gap electrodes can be utilized to fabricate single-electron transistors. The protein orientation and conformation may significantly affect the conductance of these devices. Future improvements in device reproducibility and yield will require control of these factors. PACS numbers: 73.23.Hk, 73.23.-b, 85.35.Gv, 82.37.Rs, 87.14.E-, 87.80.Nj Submitted to: Nanotechnology g u A 1 2 ] l l a h - s e m . t a m - d n o c [ 1 v 4 8 1 4 . 8 0 2 1 : v i X r a An investigation into the feasibility of myoglobin-based single-electron transistors 2 1. Introduction Electron transfer is of fundamental importance to proteins whose functionality relies on redox reactions with other proteins, cofactors, or the external environment. In proteins containing heme groups, a single Fe ion in a protoporphyrin ring changes valence state between ferrous (Fe2+) and ferric (Fe3+) oxidation states as an electron is accepted by or donated to the protein [1]. Much experimental work has been performed to develop an understanding of protein electron transfer mechanisms [2, 3] most of it based on statistical averages of protein ensembles measured by spectroscopic and electrochemical techniques [4, 5]. Electron transfer through individual molecules has been probed recently using electrochemical scanning tunneling microscopy (EC-STM) [4, 6]. In EC-STM, a gate voltage is used to align a molecules redox level to the Fermi levels of the tip and substrate with the sample in solution at room temperature. The results are explained using resonant tunneling [4, 6] and two-step electron transfer models [5]. According to resonant tunneling models, electrons transfer via resonant tunneling due to alignment of molecular orbital energy levels with the Fermi level of the contact electrodes. In the two-step process, conformational changes in the molecule occur during tunneling which change the orbital energy levels. This process is thought to be especially prevalent in redox molecules [6, 7]. In a related group of experiments, measurements on large assemblies of localized proteins using mercury contacts have been used to distinguish between tunneling and hopping electron transfer mechanisms [8]. Nanometer-gap electrodes fabricated by electromigration techniques [9] have made it possible to study electron conductance through small single redox molecules. By applying a bias voltage VB between electrodes and a gate voltage VG to the sample's substrate, molecular energy levels can be probed and characteristic single-electron transistor (SET) behavior, such as a Coulomb blockade or a Kondo resonance, can be observed [10, 11, 12, 13]. In particular, these experiments have demonstrated that molecular junctions fabricated in this way can be used to probe the electron transfer properties of the molecules. In these measurements, the presence of Coulomb blockade phenomena by itself is not proof that the molecular properties are being probed because the characteristics of the junction itself after the breaking process can affect the measurements [14], and therefore control experiments must be performed to validate the data. These measurements, if performed correctly, rely on a resonant tunneling process which in principle indicate that tunneling is a possible electron transfer mechanism. Our aim in this work is to determine whether tunneling is a viable electron transfer mechanism in redox proteins such as myoglobin by attempting to fabricate and measure single electron transistors. Here we describe the study of single-electron transistors composed of the protein myoglobin (Mb). Mb is a roughly spherical globular metalloprotein containing a heme group surrounded by a folded single-chain polypeptide consisting of 153 amino acids (apomyoglobin) [15]. Mb was chosen because 1) it is relatively small (4.4 nm ×4.4 nm × An investigation into the feasibility of myoglobin-based single-electron transistors 3 2.5 nm), which is desirable to increase electron tunneling probability, 2) the apo form is readily available and well studied and can thus be used as a control to discern effects related to the heme group, and 3) its structure and vibrational spectra are well known. In other words, Mb is a model system with which to test bioelectronic devices based on heme-containing proteins. This paper is organized as follows. After briefly discussing the general properties of myoglobin and single-electron transistors in the Background section, we describe including break junction fabrication, electrical the experimental techniques used, measurements, and protein incorporation. In the Results section we describe results from bare junctions, junctions with apomyoglobin (apoMb), and junctions with Mb. In the Discussion section we analyze our data using the known properties of myoglobin and the results from the bare and apoMb samples. Because SET conductance can result from metal grains produced by the electromigration process [16, 14], an important distinction between samples with and without protein incorporation is the observation of inelastic tunneling from discrete vibrational modes [9, 11, 17]. We also identify heme-group resonant tunneling by comparing the behaviors of apoMb samples with Mb samples. The Conclusions section summarizes our results and identifies areas where the experiments can be improved. In Appendix A we describe differential SET conductance calculations with and without tunneling-induced structural changes in the protein. 2. Experimental techniques 2.1. Break-junction fabrication The junctions were grown on Si substrates with a SiO2 layer 200 nm thick. The resistivity of the degenerately-doped silicon wafers ranged from 0.002 Ω-cm to 0.003 Ω-cm. The Si substrate served as the gate contact. The widths of junctions were usually between 100 nm and 300 nm. The junctions were fabricated by electron-beam lithography using a lift-off technique. Pt films, 20 nm thick, were deposited on a Si/SiO2 substrate by magnetron sputtering with approximately a 15◦ deposition angle with respect to the surface normal of the sample. The samples were rotated about the surface normal during deposition. The Pt deposition rate was 0.028 nm/s and the base pressure was 5 × 10−8 Torr; an Ar pressure of 1.5 mTorr was used during deposition. The junction thickness at its narrowest point was approximately 8 nm to 10 nm because of a shadow effect during the sputtering deposition process, as shown in figure 1(a). The junction with this kind bow-tie shape tends to break only in the middle, does not require high voltages to break [figures 1(b) and (c)], and can be broken by an electrical pulse with a low negative voltage, as shown in figure 1(d) [18]. The total resistance of the junctions (including parasitic resistance) at room temperature ranged between 100 Ω and 2000 Ω. The narrowest gap between electrodes formed in this way are on the order of 5 nm, as shown in figure 1. An investigation into the feasibility of myoglobin-based single-electron transistors 4 c ) A m ( t n e r r u C 6.0 4.5 3.0 1.5 0.0 0.0 0.2 0.4 0.6 0.8 1.0 Voltage (V) Figure 1. (a) Atomic force microscopy (AFM) image for one Pt junction which 20 nm thickness in large area, 8 to 10 nm in the middle. (b) AFM image after breaking the junction by electromigration at room temperature. (c) I − V measurement for breaking the junction at 0.9 V and room temperature. (d) AFM image for another junction broken by an electrical pulse (-50 mV) at T = 6 K. 2.2. Protein incorporation and viability Mb and apoMb at least 95% pure were obtained from horse skeletal muscle (Sigma) in its powder state and essentially salt-free. The protein powders were dissolved in a 10 mM phosphate buffer saline (PBS, Sigma, pH = 7.4) with 14.2 mM and 31.5 µM concentrations for Mb and apoMb, respectively. Protein solutions were passed through a 0.2 µm pore size filter to remove larger protein clusters that may have formed. In order to remove possible contaminants from the electrode surface, junctions were rinsed with optima acetone and isopropanol and subsequently cleaned with O2 plasma. After the cleaning procedure, the protein solution was dropped on the samples. The samples were kept at 4 ◦C in order to let water evaporate. A drop of the solution was placed on the unbroken junction prior to cooling down the samples. Linker molecules were avoided because junctions exposed to them (without protein) generated resonant tunneling electronic data. These data showed a weak dependence on gate voltage, which would have made it difficult to disntiguish the protein behavior from that of the linker (see supplementary information). Raman spectra showed that the Mb was active after cooling down to T = 5 K and then warming up to room temperature [19]. Therefore, it is unlikely that the protein denatured during the cooling process. Another possible problem is that the local temperature during the electromigration process might be high enough to denature the protein [20, 21, 22]. Measurements of the electrical resistance at low voltages as a function of temperature yielded an estimate of a maximum An investigation into the feasibility of myoglobin-based single-electron transistors 5 Figure 2. Sketch of three-terminal device and AFM image of a bare Pt junction broken by electromigration. The AFM image indicates that the gap is on the order of 5 nm wide. temperature of approximately 400 K, which is consistent with estimates of 495 K for junctions broken at room temperature [23]. Additional Raman scattering measurements performed on protein heated up in air in a furnace or hot plate demonstrated that myoglobin is viable after heating to 400 K (see supplementary information). The junctions with a bow-tie configuration were fabricated from Pt thin films, 8 to 10 nm thick in the middle, 200 nm wide and deposited on a Si wafer with a 200 nm thick SiO2 insulating layer. The doped Si substrate was used as a gate contact in all samples, as shown in figure 2. Electrical measurements were performed immediately after breaking the junction at low temperature (either 77 K or 6 K). The conductance (dI/dV ) as a function of VB and VG was acquired by numerically differentiating current-bias voltage (I-V ) curves. Junctions were measured without any protein, with apomyoglobin, and with myoglobin. 3. Results 3.1. Bare junctions In the control experiments with bare junctions (without protein), 107 Pt junctions were broken by electromigration at T = 77 K. The broken junctions had a resistance ranging from 1 MΩ to 1 GΩ. All but one of the broken junctions did not show differential conductance peaks. Figure 3(a) is a typical result measured at T = 77 K. The conductivity was relatively low and there was a significant amount of noise for these structures. Only one sample had conductance peaks, as shown in figure 3(b). In addition, 31 bare Pt junctions were broken by electromigration and measured at T = 5.5 K. Seventeen of these did not have conductance peaks as shown in figure 3(c). Out of the remaining fourteen junctions that had conductance peaks, eleven broken An investigation into the feasibility of myoglobin-based single-electron transistors 6 5.4 4.8 4.2 3.6 3.0 (b) (c) 1.5 1.0 0.5 0.0 -200 0 200 -200 0 200 V B (mV) 2 ) S (a) 5 - 0 1 ( V d / I d 1 0 -200 0 200 60 30 0 ) V m ( B V -30 -60 -4 -2 0 V G (V) 2 4 6 Figure 3. Typical results for bare Pt junctions. (a) Conductance as a function of bias voltage measured at T = 77 K. No conductance peaks are found near zero bias. (b) Conductance of the only sample out of 107 junctions broken at T = 77 K that had noticeable conductance features. (c) Typical differential conductance at T = 5.5 K for bare junctions. No conductance peaks are evident. (d) SET behavior measured in a bare Pt junction at T = 5 K [gray scale from −3.5 × 10−7 S (black) to 1.8 × 10−5 S (white)]. junctions did not have a dependence on the gate voltage. The remaining three junctions did have conductance peaks whose positions were dependent on the gate voltage. Coulomb diamonds, characteristic of SET behavior, were observed in the conductance as a function of gate and bias voltages as shown in figure 3(d). This demonstrates that occasionally Pt grains are formed, as has been previously observed in Au junctions, which can mask the molecular information [16, 14, 22]. These islands have several energy levels and the conductance peaks from metal grains can be tuned by gate voltage modulation. Many Coulomb diamonds were observed, each corresponding to the addition of an electron to the island. It is important to note that the appearance of several Coulomb diamonds is unlikely to happen in single-protein junctions, since excessive charging can denature the protein. 3.2. Apomyoglobin experiments In the apoMb experiments, out of 125 Pt protein-coated junctions broken at T = 77 K, nine had conductance peaks with a very weak response to VG. Figure 4(a) is the conductance measured at T = 77 K for one of these apoMb samples. One conductance peak was approximately constant as a function of VG. This peak, measured at VG = 0 as a function of temperature, is shown in figure 4b. Figure 4(c) shows how the peak height depended on temperature. The monotonic decrease of the peak height with increasing temperature, together with saturation at low temperatures (T < 100 K) indicates that tunneling, and not Mott hopping, is the dominant electron conduction mechanism in this device [24, 25]. Tunneling is observed because the conductance is smaller than the An investigation into the feasibility of myoglobin-based single-electron transistors 7 conductance quantum of G◦ = 2e2/h = 7.75 × 10−5 S. The lack of dependence of VG means, however, that the separation of energy levels inside the protein are much larger than kBT [26]. Forty-three apoMb samples were measured at T ≈ 6 K. Fifteen of them had conductance peaks and two had a very weak VG dependence. For three of the apoMb samples, conductance lines changed slightly with VG as shown in figure 4(d). The spectra measured at VG = 0 at several temperatures is shown in figure 4(e). The zero conductance region near VB = 0 may arise from a Coulomb blockade effect, but the lack of Coulomb triangles indicates that the energy levels in the apoMb protein are too close in energy to be able to observe a Coulomb staircase. The temperature dependence of the the conductance at VB = 0 is shown in figure 4(f). For temperatures below T = 20 K the conductance is independent of temperature, characteristic of a tunneling process, but at higher temperatures the conductance quickly increases until the spectrum completely loses its tunneling gap above T = 100 K. The line that starts at VB ≈ 132.5 mV, indicated by the red arrow, may arise from the vibrational spectra of amino acid redox centers [27]. 3.3. Myoglobin experiments A total of 215 Mb samples were measured at T ≈ 5 K, out of which 21 had conductance peaks that strongly depended on VG. In another approach, junctions were broken by a single electrical pulse with a very low negative voltage with a mechanism similar to voltage-pulse-induced electromigration [18]. These thin (∼ 8 nm) Pt junctions can be broken by a low negative voltage ≈ −60 mV which avoids potentially heating the sample to high temperatures. Out of 64 Pt Mb-coated junctions broken by electrical pulse at T = 6 K, four had an obvious dependence on VG. Of the 25 samples that had gate voltage-dependent behavior, the conductance maps can be divided into three types: 1) data where a Kondo-like resonance is observed, that is, conductance near zero bias voltage that is independent of gate voltage; 2) Coulomb blockade without a two-step process; and 3) Coulomb blockade with a two-step process. Figure 5 shows results from one of the samples with a Kondo-like resonance. When the junction was initially broken by a negative electrical pulse, no conductance peaks were found. However, there was a sudden increase in the current when the voltage was increased again, indicating a modification of the contact and possible protein incorporation in the gap of electrodes [10], although this could also be due to Joule heating resulting from a sudden narrowing of the junction in the bowtie region [28]. Conductance peaks were observed thereafter. In the differential conductance map, shown in figure 5(a), two intense peaks around VB = 0 are independent of VG for 0 < VG < 3 V. For higher VG values, the peaks are composed of half a Coulomb blockade- like triangle. This could be an indication of a Kondo-resonance assisted tunneling [10], although verification would require measurement in large magnetic fields. The two lines indicated by the green arrows, corresponding to green arrows in figure 5(a), do An investigation into the feasibility of myoglobin-based single-electron transistors 8 (a) 80 40 0 ) V m ( B V -40 (d) 200 100 0 ) V m ( B V -100 -6 -4 -2 0 VG (V) (b) 77 K 90 K 109 K 129 K 148 K -200 2 4 -10 0 (e) 0.8 0.6 0.4 0.2 ) S 5 - 0 1 ( V d / I d 0.0 20 30 10 VG (V) 5.8 K 9.8 K 50 K 100 K 150 K -100 -50 0 50 100 -300 -200 -100 0 100 200 300 VB (mV) (c) -80 2.6 2.4 2.2 2.0 1.8 1.6 2.6 2.4 2.2 2.0 ) S 5 - 0 1 ( V d / I d ) S 5 - 0 1 ( x a m V d / I ) S 6 - 0 1 ( 0 = V V d / I d VB (mV) (f) 3 2 1 0 0 80 40 120 1/T (10-3 K-1) 160 d 1.8 6 7 10 11 12 13 9 8 1/T (10-3 K-1) Figure 4. ApoMb results. (a) Differential conductance (dI/dV ) measured at T = 77 K as a function of bias and gate voltages. Gray scale from 1.56 × 10−5 S (black) to 2.64 × 10−5 S (white) . (b) dI/dV measured at VG = 0 for various temperatures. (c) dI/dV of peak in (b) as a function of 1/T . (d) dI/dV for another Mb sample [gray scale from 0 (black) to 5.0 × 10−6 S (white)] measured at T = 6 K. The red arrow points to a possible vibration-assisted conduction line. (e) dI/dV at VG = 0 for various temperatures. (f) Conduction at VB = 0 and VG = 0 for sample in (d) as a function of 1/T . not depend strongly on VG for 0 < VG < 1 V. The conductance feature starting at VB ≈ 30 mV could arise from vibrational excitations as this energy coincides with the energy of the Fe-His vibration (27.3 meV). Conductance steps are also present as indicated in figure 5(b). These steps result in peaks in d2I/dV 2, indicated by the black (9 mV), blue (35 mV), and magenta (≈ 80 mV) arrows in figure 5(b). Taking into account that modes increase in energy at low T , the arrows correspond to heme- doming [29, 30], Fe-His, and, heme-group deformation (83.7 meV) modes, respectively [31]. Conductance data for a sample which displayed Coulomb blockade behavior without a two-step process are shown in figure 6(a). Figure 8(b) is a calculation using reasonable parameters that reproduce the main characteristics of the experimental data. The appearance of two wide conductance lines starting at VB ≈ 5.2 mV, indicated by red arrows in figure 6(a), arise from inelastic tunneling with the 4.96 meV vibrational mode of the heme group [29, 30]. Another sample with SET behavior is shown in figure 6(b). In this case, two An investigation into the feasibility of myoglobin-based single-electron transistors 9 80 40 0 -40 -80 1.5 1.0 0.5 0.0 ) V m ( B V ) S 5 - 0 1 ( V d / I d (a) 0 2 4 6 8 10 12 VG (V) 3 0 (b) -100 -50 0 50 100 VB (mV) ) / V S 4 - 0 1 ( 2 V d / I 2 d -3 Figure 5. Mb results. (a) dI/dV as function of VB and VG at T = 6 K [gray scale is 0 (black) to 1.33 ×10−5 S (white)]. Green arrows point to conduction lines corresponding to Fe-His vibration-assisted tunneling. (b) dI/dV (black) and d2I/dV 2 (red) spectra at VG = 11 V. Vertical arrows correspond to inelastic tunneling peaks. deformed Coulomb blockade triangles near zero bias voltage can be found without a degeneracy point in the differential conductance graph. This means that two charge states do not have the same energy, and in this case the separation between the top triangles is > 60 mV. This separation, as shown by the calculation in figure 8(c), is an indication that the two-step electron transfer process is a viable mechanism. In addition, there are three areas that exhibit negative differential conductance (NDC), indicated by the three blue arrows. One explanation of how NDC behavior may arise is from the existence of two quantum dots in series . This suggests that these measurements were the result of transport through two or more myoglobin proteins in series with each other. The faint line at VB ≈ −25 mV, indicated by the green arrow, does not have a strong dependence on VG, which indicates electron tunneling by the medium around the heme group. We also note that the conductance feature that starts at VB = −97 mV, indicated by the red arrow, could arise from one of the excited electronic level of the An investigation into the feasibility of myoglobin-based single-electron transistors 10 50 (a) ) V m 0 ( B V -1 0 VG (V) 1 -50 40 0 -40 -80 (b) ) V m ( B V -120 -2 -1 0 1 VG (V) 2 3 Figure 6. (a) Differential conductance (dI/dV ) data from a Mb sample [gray scale from 0 (black) to 5.0 × 10−6 S (white)] at T = 6 K. The red arrows point to vibration- assisted conduction lines. (b) Differential conductance T = 6 K [gray scale from −0.27 × 10−6 S (black) to 1.6 × 10−6 S (white)] for another Mb sample. The blue arrows point to regions of negative differential conductance (NDC). The green arrow points to a faint conductance line independent of VG . The red arrow points to another conductance line possibly arising from a vibrational-assisted tunneling process or an excited electronic level in the protein(s). protein or a vibrational excitation of the protein. In particular, this energy coincides with the energy of the tryptophan amino acid in Mb which is known to have an energy of 760 cm−1 = 94.2 mV (measured at room temperature) [32]. The wide conductance line that starts at VB ≈ 93 mV, indicated by the pink arrow, is another line that could arise from the excited state or vibrational excitation. An investigation into the feasibility of myoglobin-based single-electron transistors 11 Figure 7. Heme group structure (a) and energies of Fe d-orbitals in dry metMb (b). 4. Discussion Here we discuss our results in terms of the known properties of myoglobin. Figure 7(a) shows the heme group structure of dry Mb. The fifth Fe ligand binds covalently with amino acid His 93, and therefore the Fe-His vibrational mode (27.3 meV) is an excellent heme-group indicator [29]. The sixth Fe ligand is occupied by H2O for aqueous Mb but is coordinated with His 64 when dry. In the latter case the Fe atom is in the plane of porphyrin ring in a low spin (S = 1/2) configuration (metMb) [31]. The ligand field splits the Fe energy levels into two eg and three t2g orbitals, as illustrated in figure 7(b), with an energy gap ∆ ≈ 1.0 eV for metMb [29]. These orbitals split further by 0.25 eV to 0.37 meV due to molecular orbital interactions [31], which is difficult to detect even at a cryogenic temperature of T ∼ 5 K. Excited states are also difficult to probe because of low VB (< 300 mV) used in the measurements (higher voltages destroyed our samples). A Mb SET can be modeled as a single energy level device with inelastic tunneling processes possible due to the localized vibrational modes of the heme group. However, it is known that redox reactions can cause significant conformational changes of the heme group [33] and that the associated relaxation time could range from picoseconds to seconds [34]. If the relaxation time is long, the electron resonantly tunnels while the protein remains in the intermediate state and its degeneracy point is at VB = 0. If the relaxation time is short (on the order of a few ps), the protein quickly relaxes from the intermediate state to its ground state, whose level may fall below the Fermi level of electrodes, as illustrated in figure 8(a). In this case, the electron is trapped in the protein, current flow is blocked, and the Coulomb blockade lacks a degeneracy point. This is further illustrated by the calculations in figures 8(b) and 8(c) performed by solving tunneling rate equations [35, 36] (see Appendix A). When the two-step relaxation An investigation into the feasibility of myoglobin-based single-electron transistors 12 (a) (b) (b) 60 30 0 ) V m ( B V -30 -60 -1.5 -1.0 -0.5 0.5 1.0 1.5 0.0 VG (V) (c) 50 0 ) V m -50 ( B V -100 -150 -2.0 -1.5 -1.0 -0.5 0.0 0.5 1.0 1.5 2.0 VG (V) (a) Two-step electron transfer by redox center. δE is the difference in Figure 8. conformational energies between the unoccupied and occupied states. The Fermi levels of the source and drain electrodes are indicated by µS and µD, respectively. (b) Calculated differential conductance dI/dV as a function of bias and gate voltages VB and VG for T = 6 K, δE = 0 mV, VC = 0.307 V, CD : CS : CG = 24 : 40 : 1, tunneling rate 10 GHz (between the protein and the source or drain electrodes). CD, CS and CG are the capacitances of the drain, source, and gate electrical contacts, respectively. There is no two-step process in this case (δE = 0). (c) Calculated dI/dV with a two- step process for δE = 10 mV, other parameters same as in (b). See Appendix A for description of calculation. process does not take place, normal Coulomb triangles are observed in the differential conductance dI/dV map. When the relaxation process occurs, a gap opens at the intersection of the Coulomb triangles (degeneracy point). An investigation into the feasibility of myoglobin-based single-electron transistors 13 It is important to note that amino acids around the heme group may complicate the analysis by acting as intermediaries for electron hopping processes and redox centers [27]. These aromatic amino acids have low frequency vibrational modes, such as phenylalanine (124 meV), tyrosine (79.9 meV, 103 meV, and 106 meV), and tryptophan (94.2 meV, 109 meV and 125 meV) [32]. A fundamental question in analyzing our results is determining whether the protein is indeed being measured. Regarding the results of the bare junctions with respect to the Mb junctions, we notice that the conductivity maps in figures 5 and 6 do not show multiple degeneracy point like those of the bare junctions in figure 3(d), and the energy scale (gate and bias voltages) used are completely different. This is an indication that the myoglobin data come from the protein and not from metallic islands formed in the junctions. We note that a similar observation has been made for break junctions used to measure single OPV-5 molecules, where gold grains remaining in the junction create multiple Coulomb diamonds which are superimposed on the signal from the molecule [13]. As to whether the heme group plays a role in resonant tunneling processes, we note that none of the apoMb samples' conductance showed a strong dependence on gate voltage, while only Mb samples had this dependence. Another indication that the data represent the conductance of the protein are the inelastic tunneling features of the conductance plots in figures 5 and 6. As mentioned above, these features correspond to the known vibrational energies of the heme group. The effects of the protein orientation can be surmised from inelastic vibrational spectra. Two pathways for electron transfer between the heme iron and electrodes are through the porphyrin ring or the His93 group [31]. Transfer through the porphyrin ring occurs as a result of porphyrin rings p-orbital hybridization with the d-orbital of iron [31] combined with a slow porphyrin-iron structural relaxation [37]. In this case, the resonant energy level does not change during tunneling, and results like those in figures 8(b) and 6(a) are expected. On the other hand, the dip near VB = 0 and VG < 3 V in Fig. 6(b) could be due to the very fast relaxation ( ps) of the Fe-His bond [37, 38]. The qualitative difference between figures 5(a) and 6(a) could be due to different orientations of the heme group with respect to the electrodes. Moreover, because the heme group is not at the center of the protein [15], certain orientations may cause the heme to be closer to the electrodes, thus increasing the probability of observing Kondo resonances. Another complication is that the large lateral size of the break junctions could result in several proteins with different conformations being measured in parallel, or in protein dimers consisting of two proteins measured in series if the gap is somewhat larger than a single protein. We assume that conductance is dominated by the narrowest portions of the break junction since tunneling processes usually depend exponentially on the effective width of the barriers, and therefore it is unlikely that trimers or larger protein agglomerations will play much of a role in the conductance data. The data shown in figure 6(a) agree well with what is expected from a single protein conductance, whereas the data in figure 6(b) show indications of multiple protein conductance (either in series or in parallel), in addition to a two-step tunneling process. Analyzing the data with An investigation into the feasibility of myoglobin-based single-electron transistors 14 more sophisticated models that take into account multiple protein conductance may be helpful in further understanding the conductance mechanism, but this is beyond the scope of this paper. 5. Conclusions We have presented results from a series of experiments on metallic break junctions with apomyoglobin and myoglobin. Some myoglobin junctions displayed single-electron transistor behavior consistent with resonant tunneling through the heme group. In one case, tunneling appears to be a result of a two-step process where the protein changes conformation during the tunneling process. Elastic and intelastic tunneling data are also consistent with resonant tunneling from known vibrational modes of the protein. Our results underscore the importance of tunneling, as opposed to hopping, as an electron transfer mechanism in redox proteins. Future challenges include determining the spin state of a single metalloprotein via a SET configuration for sensor applications and using other metalloproteins, such as cytochrome P450, that have binding properties to specific molecules. This will require a greater yield of successful devices and control of the protein orientation and conformation on the electrodes. Acknowledgments The authors thank S. Urazhdin and his group for their help with the design of the electronic chips and the use of cryogenic equipment. J. Gu helped with the initial phases of the experiment and J. Kabulski participated in useful discussions on protein attachemnt. This work was supported by the U. S. National Science Foundation (grants EPS-0554328 and EPS-1003907), the U. S. National Institutes of Health (grant GM- 086891), and the WVNano Initiative at WVU. Some of the work was performed using the WVU Shared Research Facilities. Appendix A. Rate Equations for Two-step Electron Transfer In the following treatment we use the same approach as in references [35] and [36]. The rate equation for the transition of an electron into an energy level in a SET is dPα dt = Xβ (Γβ→αPβ − Γα→βPα), (A.1) where P is the probability that the molecule is in a specific state and Γβ→α and Γα→β are the transition rates between α and β states. The current can be calculated by electron transfer through either of the two tunnel barriers corresponding to the source and drain electrical contacts. The two currents at equilibrium are equal. The current through the An investigation into the feasibility of myoglobin-based single-electron transistors 15 source is I = IS = eXα Xβ ΓS α→βPα, (A.2) where e is the charge of the electron, ΓS multiplied by ±1 depending on the Γα→β transition contribution to the current. α→βPα is the contribution of the source to Γα→β Figure 8(a) is a diagram representing the two-step resonant tunneling process. The proteins ferric state (Fe+3) is labeled by (0) and the ferrous state (Fe+2) is labeled (1). ΓS and ΓD are the tunneling rates between the protein and the source and drain electrodes, respectively. The energy difference between the ferric and ferrous states is δE. We define P0 and P1 as the probabilities that the level in the quantum dot (protein) are unoccupied and occupied, respectively. For single electron transport, the total tunneling time τ can be estimated from τ = e/I, so that τ ∼ps for a current of I ∼ 10−7A. We therefore set up the rate equations for two-step electron transfer by making the following assumptions: (i) When the electron tunnels into the protein, the energy level 0 relaxes to level 1 via a structural distortion in a time ∼ τ . (ii) The process is not reversible and fluctuations between levels 1 and 0 resulting from nuclear distortions are neglected. This is true at low temperatures where the thermal energy is much smaller than the vibrational energy of the heme group (kBT ≪ Evib). With these assumptions, the rate of change of P0 and P1 can be written as S and f 0(1) where f 0(1) are the Fermi distribution functions of the source and drain electrodes, respectively, for the unoccupied (occupied) energy levels. These can be written as D f 0 S,D = "1 + exp S,D = "1 + exp f 1 (µN +1 − µS,D) kBT #−1 (µN +1 − δE − µS,D) kBT #−1 (A.4) (A.5) Here µN is the chemical potential of the system with N electrons and µN +1 is the chemical potential when an additional electron is added to the system. The latter can be expressed as µN +1 = E0 − eVdot = E0 − e CGVG + CSVSD Ctotal , (A.6) where Vdot is the potential of the quantum dot, VSD is the source-drain potential difference, VG is the gate potential, Ctotal = CG + CS + CD is the total capacitance of th all the electrodes in the system with CG, CS, and CD being the capacitances of the gate, source, and drain electrodes. The energy E0 ≡ eVCCG/Ctotal is proportional to the crossing potential VC, which is the potential at which current can flow at low bias ∂P0 ∂t = −P0(cid:16)ΓSf 0 S + ΓDf 0 D(cid:17)+P1hΓS (cid:16)1 − f 1 S(cid:17) + ΓD(cid:16)1 − f 1 D(cid:17)i = − ∂P1 ∂t , (A.3) An investigation into the feasibility of myoglobin-based single-electron transistors 16 due to matching of the empty electron energy level with the source electrode chemical potential. In equilibrium, ∂P0 ∂t = ∂P1 ∂t = 0 and P0 + P1 = 1. Combining this equilibrium condition with equation A.9 and the expression for the current I = e (P1ΓS (1 − fS) − P0ΓSfS) leads to and P0 = ΓS (1 − f 1 − f 1 S) + ΓD (1 − f 1 D) S) + ΓD (1 + f 0 D ΓS (1 + f 0 S − f 1 D) (A.7) (A.8) (A.9) I = eΓSh1 − f 1 S − P0(cid:16)1 + f 0 S − f 1 S(cid:17)i . In figures 8(b) and (c) dI/dV was calculated by numerically taking the derivative of the expression for I in equation A.9 with respect to VSD. As in the experiments, the chemical potential of the drain was set to µD = 0 (i.e., the drain contact was grounded). Note that if δE = 0 there is no two-step process and the result of a single-level quantum dot is recovered. References [1] Munier B, de Visser S P and Shaik S 2004 Chemical Reviews 104 3947 -- 3980 http://pubs.acs.org/doi/pdf/10.1021/cr020443g) URL (Preprint pMID: http://pubs.acs.org/doi/abs/10.1021/cr020443g 15352783 [2] Marcus R A and Reviews - http://www.sciencedirect.com/science/article/pii/030441738590014X Bioenergetics on Sutin N 1985 811 Biochimica 265 -- et 322 Biophysica Acta ISSN 0304-4173 (BBA) URL [3] Bendall D S 1996 Protein Electron Transfer (Oxford: BIOS Publishers) 76(21) [4] Tao Phys. 1996 Lett. Rev. N J 4066 -- 4069 URL http://link.aps.org/doi/10.1103/PhysRevLett.76.4066 [5] Chi Q, O Farver of Academy (Preprint http://www.pnas.org/content/102/45/16203.abstract Sciences http://www.pnas.org/content/102/45/16203.full.pdf+html) and of J Ulstrup the United 2005 States Proceedings of America the of 102 National 16203 -- 16208 URL [6] Alessandrini A, Corni S and Facci P 2006 Phys. Chem. Chem. Phys. 8(38) 4383 -- 4397 URL http://dx.doi.org/10.1039/B607021C [7] Alessandrini A, Salerno M, Frabboni S and Facci P 2005 Applied Physics Letters 86 133902 (pages 3) URL http://link.aip.org/link/?APL/86/133902/1 [8] Ron I, Sepunaru L, and Cahen D 2010 Journal of pMID: http://pubs.acs.org/doi/abs/10.1021/ja907328r 20210314 (Preprint Itzhakov S, Belenkova T, Friedman N, Pecht I, Sheves M the American Chemical Society 132 4131 -- 4140 http://pubs.acs.org/doi/pdf/10.1021/ja907328r) URL [9] Park H, Lim A K L, Alivisatos A P, Park J and McEuen P L 1999 Applied Physics Letters 75 301 -- 303 URL http://link.aip.org/link/?APL/75/301/1 [10] Park J, Pasupathy A N, Goldsmith J I, Chang C, Yaish Y, Petta J R, Rinkoski M, Sethna J P, Abruna H D, McEuen P L and Ralph D C 2002 Nature 417 722 -- 725 URL http://dx.doi.org/10.1038/nature00791 [11] Kubatkin S, Danilov A, Hjort M, Cornil J, Bredas J L, Stuhr-Hansen N, Hedegard P and Bjornholm T 2003 Nature 425 698 -- 701 URL http://dx.doi.org/10.1038/nature02010 An investigation into the feasibility of myoglobin-based single-electron transistors 17 [12] Yu L H, Keane Z K, Ciszek J W, Cheng L, Stewart M P, Tour J M and Natelson D 2004 Physical Review Letters 93 266802 URL http://link.aps.org/doi/10.1103/PhysRevLett.93.266802 [13] van der Zant H S J, Osorio E A, Poot M and O'Neill K 2006 physica status solidi (b) 243 3408 -- 3412 ISSN 1521-3951 URL http://dx.doi.org/10.1002/pssb.200669185 [14] van der Zant H S J, Kervennic Y V, Poot M, O'Neill K, de Groot Z, Thijssen J M, Heersche H B, Stuhr-Hansen N, Bjornholm T, Vanmaekelbergh D, van Walree C A and Jenneskens L W 2006 Faraday Discuss. 131 347 -- 356 URL http://dx.doi.org/10.1039/B506240N [15] Garrett R H and Grisham C M 1999 Biochemistry (Harcourt-Brace College Publishers) pp 480 -- 481, 114 [16] Houck A A, Labaziewicz J, Chan E K, Folk J A and Chuang I L 2005 Nano Letters 5 1685 -- 1688 (Preprint http://pubs.acs.org/doi/pdf/10.1021/nl050799i) URL http://pubs.acs.org/doi/abs/10.1021/nl050799i [17] Chae D H, Berry J F, Jung S, Cotton F A, Murillo C A and Yao Z 2006 Nano Letters 6 165 -- 168 (Preprint http://pubs.acs.org/doi/pdf/10.1021/nl0519027) URL http://pubs.acs.org/doi/abs/10.1021/nl0519027 [18] Hayashi T and Fujisawa T 2008 Nanotechnology 19 145709 URL http://stacks.iop.org/0957-4484/19/i=14/a=145709 [19] Li D 2009 Exploring electron transfer in myoglobin-based transistors Ph.D. thesis West Virginia University [20] Esen G and Fuhrer M S 2005 Applied Physics Letters 87 263101 (pages 3) URL http://link.aip.org/link/?APL/87/263101/1 [21] Taychatanapat T, Bolotin K I, Kuemmeth F and Ralph D C 2007 Nano Let- http://pubs.acs.org/doi/pdf/10.1021/nl062631i) URL (Preprint 7 652 -- 656 ters http://pubs.acs.org/doi/abs/10.1021/nl062631i [22] Ward D R, Halas N J and Natelson D 2008 Applied Physics Letters 93 213108 (pages 3) URL http://link.aip.org/link/?APL/93/213108/1 2006 Parviz and A B [23] Dong J Nanotechnology 17 5124 URL http://stacks.iop.org/0957-4484/17/i=20/a=014 [24] Fowler A B, Timp G L, Wainer J J and Webb R A 1986 Phys. Rev. Lett. 57(1) 138 -- 141 URL http://link.aps.org/doi/10.1103/PhysRevLett.57.138 [25] Azbel M Y, Hartstein A and DiVincenzo D P 1984 Phys. Rev. Lett. 52(18) 1641 -- 1644 URL http://link.aps.org/doi/10.1103/PhysRevLett.52.1641 [26] Kastner M A 2000 Ann. Phys. (Leipzig) 9 885 -- 894 [27] Shih C, Museth A K, Abrahamsson M, Blanco-Rodriguez A M, Di Bilio A J, Sudhamsu J, Crane B R, Ronayne K L, Towrie M, Vlcek A, Richards J H, Winkler J R and Gray H B 2008 Science 320 1760 -- 1762 (Preprint http://www.sciencemag.org/content/320/5884/1760.full.pdf) URL http://www.sciencemag.org/content/320/5884/1760.abstract [28] Strachan D R, Smith D E, Johnston D E, Park T H, Therien M J, Bonnell D A and Johnson A T 2005 Applied Physics Letters 86 043109 (pages 3) URL http://link.aip.org/link/?APL/86/043109/1 [29] Rosca F, Kumar A T N, Ye X, P M 2000 pion 4290 http://pubs.acs.org/doi/abs/10.1021/jp993617f (Preprint Journal The http://pubs.acs.org/doi/pdf/10.1021/jp993617f) Sjodin T, Demidov A A and Cham- of 4280 -- URL Chemistry Physical 104 A [30] Gruia F, Kubo M, Ye X and Champion P M 2008 Biophysical Journal 94(6) 2252 -- 2268 [31] Feng M and Tachikawa H 2001 Journal of the American Chemical Society 123 3013 -- 3020 pMID: 11457012 (Preprint http://pubs.acs.org/doi/pdf/10.1021/ja003088p) URL http://pubs.acs.org/doi/abs/10.1021/ja003088p [32] Wei F, Zhang D, Halas N J and Hartgerink J D 2008 The Journal of Physical Chemistry B 112 9158 -- 9164 pMID: 18610961 (Preprint http://pubs.acs.org/doi/pdf/10.1021/jp8025732) URL http://pubs.acs.org/doi/abs/10.1021/jp8025732 An investigation into the feasibility of myoglobin-based single-electron transistors 18 [33] Tezcan F A, Winkler J R and Gray H B 1998 Journal of the American Chemical Society 120 13383 -- 13388 (Preprint http://pubs.acs.org/doi/pdf/10.1021/ja982536e) URL http://pubs.acs.org/doi/abs/10.1021/ja982536e [34] Cherepanov D A, Krishtalik and Mulkidjanian A Y Biophysical http://www.sciencedirect.com/science/article/pii/S0006349501760845 Journal ISSN 1049 80 0006-3495 [35] Bonet E, Deshmukh M M and Ralph D C 2002 Phys. Rev. B 65(4) 045317 URL http://link.aps.org/doi/10.1103/PhysRevB.65.045317 [36] Park J 2003 Electron Transport in Single Molecule Transistors Ph.D. thesis University of California -- Berkeley S [37] Stavrov Biopolymers http://dx.doi.org/10.1002/bip.20039 2004 S 74 37 -- 40 ISSN 1097-0282 URL [38] Spiro T G 1988 Biological Applications of Raman Spectroscopy (Wiley and Sons) pp 133 -- 215 L 1033 I -- 2001 URL 

1909.03671

1

1909

Narrow autoresonant magnetization structures in finite length ferromagnetic nanoparticles

The autoresonant approach to excitation and control of large amplitude uniformly precessing magnetization structures in finite length easy axis ferromagnetic nanoparticles is suggested and analyzed within the Landau-Lifshitz-Gilbert model. These structures are excited by using a spatially uniform, oscillating, chirped frequency magnetic field, while the localization is imposed via boundary conditions. The excitation requires the amplitude of the driving oscillations to exceed a threshold. The dissipation effect on the threshold is also discussed. The autoresonant driving effectively compensates the effect of dissipation, but lowers the maximum amplitude of the excited structures. Fully nonlinear localized autoresonant solutions are illustrated in simulations and described via an analog of a quasi-particle in an effective potential. The precession frequency of these solutions is continuously locked to that of the drive, while the spatial magnetization profile approaches the soliton limit when the length of the nanoparticle and the amplitude of the excited solution increase.

cond-mat.mes-hall

cond-mat

Narrow autoresonant magnetization structures in finite length ferromagnetic nanoparticles A. G. Shagalov1∗ and L. Friedland2† 1Institute of Metal Physics, Ekaterinburg 620990, Russian Federation and Ural Federal University, 2Racah Institute of Physics, Hebrew University of Jerusalem, Jerusalem 91904, Israel Mira 19, Ekaterinburg 620002, Russian Federation and The autoresonant approach to excitation and control of large amplitude uniformly precessing mag- netization structures in finite length easy axis ferromagnetic nanoparticles is suggested and analyzed within the Landau-Lifshitz-Gilbert model. These structures are excited by using a spatially uni- form, oscillating, chirped frequency magnetic field, while the localization is imposed via boundary conditions. The excitation requires the amplitude of the driving oscillations to exceed a thresh- old. The dissipation effect on the threshold is also discussed. The autoresonant driving effectively compensates the effect of dissipation, but lowers the maximum amplitude of the excited structures. Fully nonlinear localized autoresonant solutions are illustrated in simulations and described via an analog of a quasi-particle in an effective potential. The precession frequency of these solutions is continuously locked to that of the drive, while the spatial magnetization profile approaches the soliton limit when the length of the nanoparticle and the amplitude of the excited solution increase. I. INTRODUCTION The progress in nanotechnology of magnetic materi- als stimulates theoretical and experimental research of new magnetic nanostructures [1]. It is well known that the fundamental model of magnetization dynamics, the Landau-Lifshitz equation, has a variety of exact solu- tions [2] and exhibits spatially localized objects -- soli- tons. In the small amplitude limit, these objects were studied experimentally in thin magnetic films [3 -- 6]. In these applications, the solitons comprise stretched con- figurations and long wavelength nonlinear Schrodinger (NLS) equation provides an adequate model to inter- pret experimental observations [2, 7]. In nanomagnetics, the NLS approximation for small amplitude solitons was also used in Refs. In contrast, large amplitude solitons have small spatial widths and can be observed on nanoscales only, requiring new methods of excitation and observation. These widths can be comparable to the typical length in magnetic materials, i.e., the width of the domain walls (usually, of order of 10nm). On the other hand, nanoscale samples allow to model quasi-one- dimensional (1D) configurations [1], which are frequently used in theoretical studies of solitons [2] and guarantee their stability. [8, 9]. In this work, we propose the autoresonant approach to excitation of large amplitude, uniformly rotating, nar- row magnetic structures in finite length easy axis ferro- magnetic nanoparticles by using a weak, spatially uni- form, chirped-frequency oscillating magnetic field. The autoresonance is a universal phenomenon used in numer- ous applications in many fields of physics, e.g. in parti- cle accelerators [10, 11], atomic physics [12, 13], plasmas [14], and nonlinear waves [15]. A survey of mathematical ∗ shagalov@imp.uran.ru † lazar@mail.huji.ac.il problems associated with autoresonance can be found in Ref. [16]. In recent years, autoresonance ideas were also implemented in magnetics. Examples are autoresonant waves in magnetic materials [17, 18], the switching of magnetization of single domain point-like nanoparticles [19, 20], and most recently the autoresonant excitation of large amplitude standing magnetization waves in long ferromagnetic nanowires by using rotating driving fields with short wavelength spatial modulations [21]. This re- cent approach required two resonant stages, where in the first stage one excites a rotating, but uniform magnetiza- tion of the wire, while later, in the second resonant stage, the system develops a spatially modified standing wave profile. In contrast, in the present work we study autores- onant formation of localized magnetization structures in finite length nanoparticles using a single resonance stage and without the need of a short wavelength modulation of the driving field. The localization in this case is im- posed via boundary conditions. All these modifications require a different theory, but the simplicity of the driv- ing scheme is expected to facilitate the realization of the idea in nanoscale magnets. The paper is organized as follows. In Sec. II, we de- scribe our model based on the driven Landau-Lifshitz- Gilbert (LLG) equation. A weakly nonlinear Lagrangian formulation in the dissipationless limit of this model will be used to calculate the threshold driving amplitude for autoresonant excitation of localized magnetic structures in Sec. III. Whitham's averaged variational approach [22] will be used in the calculation. Section IV will fo- cus on the effect of the dissipation on the threshold. A fully nonlinear, slow autoresonant dynamics of narrow chirped-driven magnetization structures will be discussed in Sec. V and, finally, Sec. VI will summarize our find- ings. II. AUTORESONANT EXCITATION MODEL III. THE THRESHOLD PHENOMENON 2 We consider a quasi-one-dimensional ferromagnetic nanoparticle of length L oriented along the z-axis. The particle has an easy-axis anisotropy along z and is lo- combined with a weak uniform rotating driving field cated in a constant external magnetic field H = H0bez Hd(t) = g(cos ϕdbex + sin ϕdbey) having slowly chirped rotation frequency ωd(t) = −∂ϕd/∂t. We model this system by LLG equation (in dimensionless form): mτ = h × m+ηm × mτ , (1) were m = M/M is the normalized magnetization, η is the damping parameter, and (2) Here and throughout the rest of the text (...)τ and (...)ξ denote partial derivatives with respect to τ and ξ. We use dimensionless time τ = (γK/M )t and coordinate ξ = h = mξξ + (mz + h0)bez + ε(cos ϕdbex + sin ϕdbey). z/δ, (δ =pA/K), where γ, A, and K are the gyromag- ε = M g/K, ϕd = −R Ωddτ , Ωd(τ ) = ωdM/(Kγ). We netic ratio, the exchange constant, and the anisotropy constant, respectively. (2), h0 = M H0/K, also assume that initially the magnetization mz(ξ, 0) = 1 is uniform, while at the ends ξ = 0 and ξ = l = L/δ of the particle remains fixed (mz = 1) at all times. An approach to realization of these boundary conditions will be discussed in Sec. V. In Eq. As in many other driven nonlinear systems [25], the autoresonant excitation of rotating localized magnetiza- tion structures requires (a) slow passage of the driving frequency Ωd(τ ) through a resonant frequency Ω0, in our case, Ω0 = 1 + k2 + h0 (here k = π/l), and (b) that the driving amplitude ε exceeds some threshold value εth. In particular, in the simplest case Ωd(τ ) = Ω0 − ατ , we expect the characteristic scaling εth ∼ α3/4 [25], where α is the driving frequency chirp rate. When these two conditions are met, the driven magnetic perturbation will be captured into a continuing nonlinear resonance (phase-locking) with the drive, leading to large ampli- tude excitations of the magnetization, as the frequency chirp continues. An example of a system under consider- ation could be a Permalloy sample with A = 10−11J/m, K = 105J/m3, and M = 8 × 105A/m [26]. In this case, the characteristic magnetic length is δ ≈ 10nm and the exchange length δm = 2pA/µ0M 2 It is well known that the value πδ represents a typical width of domain walls. The solitons in easy-axis magnets [2] can be interpreted as two interacting domain walls of oppo- site topological signs (a "breather" in terms of Ref. [1]) and the soliton width can be estimated as 2πδ. A sim- plest nanoparticle, where large amplitude narrow mag- netic structures having near soliton spatial profiles can be observed is a segment of a ferromagnetic nanowire of length L > 2πδ and cross-section d < 2πδm. The small cross-section guarantees quasi-one-dimensionality of the system [1] in the direction of the segment, which is as- sumed in our model [Eqs. (1) and (2)]. 0 ≈ 7nm. We proceed by illustrating the autoresonant excitation and the threshold phenomenon in system (1) in numer- ical simulations. We used the numerical approach de- scribed in Ref. [21]. The simulations solved an equiva- lent system of two coupled NLS-type equations based on the quantum two-level analog due to R. Feynman [23]. The numerical scheme used a standard pseudospectral method [24] subject to given initial and boundary condi- tions, mz(ξ, 0) = 1 and mz(0, τ ) = mz(1, τ ) = 1. In the following illustration, we neglected dissipation and used the driving frequency Ωd(τ ) = h0 + Ω′ d(τ ), where Ω′ d =(cid:26) 1 + k2 − ∆ω sin(ατ /∆ω), τ < π∆ω/(2α), 1 + k2 − ∆ω, τ > π∆ω/(2α), (3) yielding a quasi-steady-state solution in the final stage of excitation, as Ω′ d(τ ) approached a constant. The parame- ters were α = 0.005, l = 8, and ∆ω = 0.9. Figure 1 shows −mz = − cos θ and phase mismatch Φ = Φ = ϕ − ϕd (θ and ϕ being the spherical coordinates of the magne- tization vector) versus slow time T = √ατ just above (ε = 1.05ε0 th) the threshold ε0 th = 0.01 [see Eq. (17)]. Panels a,b in the figure demon- strate the excitation of a localized large amplitude mag- netic structure, which is phase-locked to the drive. In contrast, in panels c,d for ε < ε0 th, phase-locking is de- stroyed and the excitation saturates at some small am- plitude. th) and below (ε = 0.95ε0 In analyzing the autoresonance threshold in the prob- lem analytically, we first use the dissipationless limit of Eqs. (1) in spherical coordinates: mx = sin θ cos ϕ, my = sin θ sin ϕ, mz = cos θ: + Φ2 θξξ sin θ θτ = Φξξ sin θ + 2Φξθξ cos θ − ε sin Φ, Φτ = − ξ cos θ + cos θ − Ω′ d − ε cot θ cos Φ.(5) The boundary conditions are θ(0) = θ(l) = 0. Equa- tions (4), (5) satisfy the variational principle with the Lagrangian density (4) Λ = (θ2 ξ + Φ2 ξ sin2 θ) + Φτ cos θ 1 2 +Ω′ d(τ ) cos θ − 1 4 cos(2θ) − ε sin θ cos Φ, which allows to write Eqs. (4), (5) as (cid:18) ∂Λ ∂Φτ(cid:19)τ ∂Φξ(cid:19)ξ +(cid:18) ∂Λ ∂θξ(cid:19)ξ (cid:18) ∂Λ = = ∂Λ ∂Φ , ∂Λ ∂θ . (6) (7) (8) Since the autoresonance threshold is a weakly nonlin- ear phenomenon [20], our next goal is to discuss the slow, weakly nonlinear evolution in the problem and we use Whitham's averaged Lagrangian approach [22] for achieving this goal. This approach is designed to describe wave systems with slow parameters. In our case the slow parameter is the driving frequency Ωd(τ ) and the slow- ness means that this frequency does not change signifi- cantly during one period of the driving, i.e. 2π dτ ≪ 1. Ω2 d We seek rapidly rotating solution, such that ϕt ≈ Ωd(τ ), but assume that the phase mismatch Φ(τ ) = ϕ − ϕd is slow (i.e., experiences only a small change during one pe- riod of the drive). Since in the linear approximation and constant Ωd, Eqs. (4), (5) yield a phase-locked driven solution dΩd θ = a sin(kξ), Φ = π where a = ε/(Ω′ problem, we use the following small amplitude ansatz d − 1 − k2), in the slowly varying Ωd θ = a(τ ) sin Θ, Φ = f (τ ) + b(τ ) sin Θ. (9) Here Θ = kξ and it is assumed that the weak nonlin- earity, small driving, and slow variation of the driving frequency introduce slow time dependence in a, b and f . Note that this anzatz is just a truncated Fourier ex- pansion conserving the boundary conditions and that for k = π/l, the length l of the nanoparticle is one half of the periodicity length l0 = 2π/k in (9). At this stage, for simplifying the derivation, we will set b = 0, but later 3 show that to lowest order, inclusion of nonzero b does not introduce a significant change in the weakly nonlin- ear theory. The substitution of (9) into (6), the expansion to fourth order in a, and averaging over Θ ∈ [0, π] yields the averaged Lagrangian density ¯Λ = a2 4 a4 64 2 π (1 + k2 − Ω′ d − β) + (−4 + Ω′ d + β) − aε cos f, (10) where β = fτ and we omitted terms independent of f and a, as not contributing the dynamics. Note that Eqs. (9) involve only slow time dependencies and, thus, there is no need for averaging the Lagrangian over the fast os- cillations. Next, we use ¯Λ and take variation with respect to f , leaving lowest significant order terms only to get da dτ 4ε π = − sin f. (11) Here and in the next section the problem depends on τ only, so the time derivatives are now defined as d(...)/dt. Similarly, the variation with respect to a yields the sec- ond evolution equation β = df dτ = 1 + k2 − Ω′ d − 4 − Ω′ d 8 a2 − 4ε πa cos f. (12) Note that the frequency of rotation of the magnetization vector in the linearized undriven problem is h0 + 1 + k2 and assume passage through the linear resonance, i.e. Ω′ d = Ωd−h0 = 1+k2−ατ , where ατ is viewed as a small deviation of the driving frequency from the resonance. Then Eqs. (11) and (12) guarantee the assumed slowness of variation of a and f if the nonlinearity and driving am- plitude ε and chirp rate α are sufficiently small. Next, we introduce rescaled amplitude A = [(3 − k2)/8]1/2α−1/4a. This allows to rewrite Eqs. (11) and (12) as dA dT df dT = −µ sin f, = T − A2 − µ A cos f, (13) (14) where T = α1/2τ is the slow time (used in Figs. 1 and 2), and µ = [2(3 − k2)]1/2ε πα3/4 . (15) By introducing a complex dependent variable Ψ = Aeif , Eqs. (13) and (14) can be combined into a single equation characteristic of many autoresonance problems [25]: (b) 0 T 20 40 (d) 1 0 z m − −1 0.5 1 0 −1 0.5 z m − (a) 0 z/L −0.5 0 30 20 10 T (c) 0 z/L −0.5 0 30 20 10 T Φ Φ 6 5 4 3 2 1 0 6 5 4 3 2 1 0 0 20 T 40 i dΨ dT + (T − Ψ2)Ψ = µ. (16) FIG. 1: The magnetization component −mz and phase mismatch Φ versus slow time T = α1/2t. (a) and (b) just above the threshold, ε = 1.05ε0 th. (c) and (d) just below the threshold, ε = 0.95ε0 th. If starting in zero equilibrium Ψ = 0 at T < 0 (above the linear resonance), this equation guarantees phase locking at f ≈ π after passage through the resonance (i.e., for T = 0), where A increases as ∼ √T , provided the single 4 10−1 h t ε α=0.0256 α=0.0128 α=0.0064 FIG. 2: The disappearance of the spatial modulation of Φ after entering the autoresonance regime FIG. 3: The effect of the dissipation on the threshold for capture into autoresonance. The lines are given by Eq. (27) and the markers represent numerical simulations. 10−2 10−4 10−3 η 10−2 parameter µ in the problem exceeds the value of 0.41 [20]. This yields the threshold driving amplitude ε0 th = 0.41πα3/4 [2(3 − k2)]1/2 . (17) The results in Fig. 1 illustrate that this threshold is in a good agreement with numerical simulations. Next, we include the spatial variation in Φ = f (τ ) + b(τ ) sin Θ in calculating the averaged Lagrangian density. This results in the addition of a new, b-dependent part ¯Λb(a, f, b, bτ ) = 2 3π a2bτ + ε 2 ab sin f + k2 16 a2b2 (18) to already discussed Lagrangian density (10). By tak- ing the variation with respect to b, we now get a new evolution equation da dτ 3π 8 = − ε sin f − 3πk2 32 ab. Then, on using Eq. (11), one obtains ab = 32 3πk2 ( 4 π − 3π 8 )ε sin f = 0.32 k2 ε sin f. (19) (20) This result shows that b becomes negligibly small as a increases and significantly exceeds ε in the autoresonant stage, validating the derivation of the threshold driving amplitude presented above. This effect is illustrated in Fig. 2, showing the spatial form of Φ versus slow time T for parameters ε = 0.02, α = 0.005, L = 8, and ∆ω = 1. One can see how the spatial modulation of Φ nearly dis- appears after the passage through the linear resonance at T = 0 and entering the autoresonant stage of evolution. IV. EFFECT OF DISSIPATION For including a weak dissipation in the calculation of the threshold, we return to the original system of evolu- tion equations with Gilbert damping ∂Φτ(cid:19)τ (cid:18) ∂Λ ∂Φξ(cid:19)ξ − +(cid:18) ∂Λ ∂θξ(cid:19)ξ − (cid:18) ∂Λ ∂Λ ∂Φ ∂Λ ∂θ = −ηϕτ sin2 θ, = ηθτ . (21) (22) Here the left hand sides are the Lagrange's components of the dissipationless case described by the Lagrangian density (6) and, as before, ϕ = Φ + ϕd is the azimuthal angle of the magnetization vector. Within the ansatz Φ = f (τ ) and θ = a(t) sin Θ discussed above, Eq. (21) yields θτ sin θ + ε sin θ sin Φ = η(Φτ − Ωd) sin2 θ, (23) where Ωd = dϕd/dτ = Ω0 − ατ is the driving frequency. On averaging the last equation with respect to Θ ∈ [0, π], one obtains da dτ 4ε π = − sin f − ηaΩd, (24) where, assuming a continuing phase-locking (Φ remains nearly constant), we neglected Φτ in the right hand side. The last equation replaces Eq. (11) of the dissipationless case. Furthermore, this equation shows that the right hand side of Eq. (22) involves a product of two small objects and can be neglected in the following. Thus, the additional evolution equation [originating from (22)] df dτ = ατ − 3 − k2 8 a2 − 4ε πa cos f, (25) is the same as Eq. (12) for the dissipationless case. Using the rescaling of the dependent and independent variables of the dissipationless case, Eqs. (24) and (25) can be combined into a single complex equation [compare to Eq. (16)] i dΨ dT + (T − Ψ2)Ψ + i γ 2 Ψ = µ, (26) where γ = 2ηΩdα−1/2. The capture into autoresonance in this problem was studied in Ref. [27], yielding the following threshold driving amplitude for γ < 1. εth ≈ ε0 th(1 + 1.06γ + 0.67γ 2). (27) We compare this result with the numerical simulations of the original LLG equation in Fig. 3, showing εth versus η for three values of the chirp rate α = 0.0064, 0.0128, and 0.0256. We used parameters l = 8, ∆ω = 0.9, h0 = 5 and frequency variation of the form (3). The simulation was limited to values of γ < 1 and used the initial and final simulation times T0 = −10 and T1 = 30, respec- tively, reaching the quasi-steady-state at the final time. Arriving in simulations at this quasi-steady-state without dephasing served as the criterion for finding εth. V. NONLINEAR QUASI-STEADY-STATE In our numerical simulations (e.g., Fig.1) we observe that in the autoresonant stage of evolution, both Φ and θ perform slow oscillations around some smooth time vary- ing averages. We will refer to the smooth evolution as the quasi-steady-state (it still slowly varies in time). Simi- lar slow oscillations around the smooth average were ob- served and studied in many autoresonant problems [25] and reflect the stability of the quasi-steady-state if per- turbed. The problem of stability of fully nonlinear au- toresonant evolution is important, but very complex in our case and will remain outside the scope of the present work and, thus, we focus on describing the quasi-steady- state only. We proceed from Eqs. (4) and (5) and assume a perfect continuing phase-locking, Φ = π (uniform pre- cession of the magnetization vector, ϕt = Ω′ d(τ )), and neglect θτ despite the time variation of the driving fre- quency. This leaves us with a single equation θξξ = −Vθ, (28) where we have also neglected the driving term and de- fined the effective potential V = −Ω′ d(τ ) cos θ + 1 4 cos(2θ) − [ 1 4 − Ω′ d(τ )]. (29) We have added a shift in V so that V = 0 at θ = 0 at all times, which is convenient for classifying different types of solutions (see below). Equation (28) can be viewed as defining the motion of a quasiparticle having coordinate θ in the effective potential V , ξ playing the role of time and Ω′ d(τ ) being a parameter. The solution of (28) subject to boundary conditions θ = 0 (mz = 1) at ξ = 0, l, corresponds to a single round trip of the quasiparticle in the effective potential starting from θ = 0 at ξ = 0 and returning to the same point θ = 0 at ξ = l. The form of the quasi-potential allows to qualitatively describe the motion of the quasiparticle in the problem. Note that for Ω′ d < 1 the quasipotential has a zero max- imum at θ = 0, a minimum Vmin = − (Ω′2 d /2 + 1/4) 0.6 0.4 0.2 0 −0.2 V 0 0.2 A>0 A=0 A<0 Vmin 0.4 5 Vmax θ max/π 0.6 0.8 1 θ/π FIG. 4: The effective potential. Two classes of oscillations correspond to different values of the quasi-energy: A > 0 (solid horizontal line)is the case of the zero boundary conditions used in this work, A < 0 (dished horizontal line)is the standing magnetization wave of Ref. [21]. The limit A → 0 (dotted horizontal line) corresponds to the soliton solution of infinite extent (λ → ∞). 2 θ2 d is 0 < Ω′ d, and another maximum Vmax = 2Ω′ at cos θ = Ω′ d at θ = π, as illustrated in Fig. 4 for Ω′ d = 0.25. There- fore, the quasiparticle passes θ = 0 (i.e. one satisfies the boundary conditions θ = 0 at ξ = 0 and l) provided (a) Vmax > 0 (i.e. the allowed range of Ω′ d < 1) and (b) the quasi-energy A(τ ) = 1 ξ + V is in the inter- val [0, Vmax] (see the solid red horizontal line in Fig. 4). The solutions for θ(ξ) in this case have a single maximum and a finite extent in ξ. In addition to this type of solu- tions, there also exists a motion of the quasiparticle, such that Vmin < A < 0 (see the horizontal red dashed line in Fig.4), but this motion does not reach θ = 0 and, thus, can not satisfy our boundary conditions. Autoresonant magnetic excitations of this different type were discussed in Ref. [21] and comprise spatial oscillations of θ having a finite periodicity length. Interestingly, the two types of solutions coalesce in the limit A → 0 (the dotted red line in Fig. 4) and θ(ξ) assumes a form of a single peak de- fined on the infinite ξ domain, i.e a soliton (see Ref [2]). The physical width of θ in this limit can be estimated as (30) ∆ξ ≈ 2π/κm, d )1/2 is the frequency of linear os- where κm = (1 − Ω′2 cillations around the minimum of V . For example, if 0 < Ω′ d < 0.5, one has 2π < ∆ξ < 7.25. In our case of a finite length of the nanoparticle, θ(ξ) can only approx- imately approach the form of the soliton, but, neverthe- less, Eq. (30) can still be used as an estimate of the with of the excited solution. Note that Eq. (28) can be solved in quadratures for finding θ(ξ, τ ) at a given time. However, this so- lution also requires the knowledge of the quasi-energy A(τ ) = 1 ξ + V . Finding this energy, seems to require including the drive and solving the full driven PDE. Nev- ertheless, if the system evolves in autoresonance, one can 2 θ2 calculate A(τ ) by using simplified arguments. Indeed, in the autoresonance, the azimuthal frequency of magneti- zation follows that of the drive, while the extent λ of the spatial profile of θ remains constant λ = l. The preser- vation of λ(τ ) at value l, despite the variation of Ω′ d in time allows calculation of the quasi-energy at all times using the definition dθ . (31) λ = 2Z θmax 0 p2[A(τ ) − V (θ, Ω′ d)] As an example, we have solved this problem numeri- cally (for details of this calculations see Appendix) and found A for different values of Ω′ d = 1 + k2 − D in the case of l = 8 (k = 0.393) and 0.7 6 D 6 1.05 (re- call that D represents the driving frequency deviation D = ∆ω sin(ατ /∆ω) from the linear resonance frequency in all our numerical simulations). Note that λ → ∞ at the limiting values A = 0 and A = Vmax and, there- fore, λ has a minimum at some A between 0 and Vmax. Therefore, for having the solution of equation λ = l, l must be above this minimum. We have found that such solutions in our example exists for D < 1.05. The knowledge of A allows to calculate the spatial profile θ(ξ) ) x a m θ ( s o c ) x a m θ ( s o c 1 0.5 0 −0.5 −1 −5 1 0.5 0 −0.5 −1 −5 (a) ∆ω=0.7 ∆ω=0.8 ∆ω=0.9 ∆ω=1.0 0 5 10 T 15 20 25 30 (b) 0 5 10 T 15 20 25 30 FIG. 5: The axial magnetization component mz = cos(θmax) at the maximum θmax of the autoresonant excitation versus slow time T . (a) the quasi-steady-state theory, (b) numerical LLG simulations. 6 (b) ∆ω=0.7 1 0.5 0 (a) z m ∆ω=0.4 ∆ω=0.5 ∆ω=0.6 ∆ω=0.7 0 10 T 20 −0.5 0 0.5 z/L 1 z m 0.8 0.6 0.4 0.2 0 −0.2 −0.4 FIG. 6: The autoresonant excitation in the presence of dissipation η = 0.01 (LLG simulations). (a) The axial magnetization component mz = cos(θmax) at the maximum θmax of the autoresonant excitation versus slow time T , (b) The comparison of the excited waveform (solid line) with the exact dissipationless soliton solution [2] (dashed line). by solving Eq. (28) in quadratures, but, for simplic- ity, we have limited the calculation to just finding the maximum of the autoresonant excitation θmax = θ(l/2) by solving V (θmax) = A at each time. We summarize this quasi-steady-state evolution in Fig. 5a, showing mz(l/2) = cos θmax (circles) versus slow time T for four values of ∆ω = 0.7, 0.8, 0.9, and 1.0. The results are in a good agreement with the full numerical LLG simula- tions shown in Fig. 5b for the same ∆ω, confirming the quasi-steady-state theory. Note that the quasi-steady- state calculations involve solution of algebraic equations only, which is significantly simpler that the full numerical simulations. It is interesting to compare the amplitudes of the solu- tions in Fig. 5a to those of the exact solitons for easy-axis magnets [2]: tan2(θmax/2) = Ω′ d/(Ω′ d − ∆ω) − 1. (32) This equation yields cos(θmax) = −0.20,−0.38,−0.56, −0.73 for ∆ω = 0.7, 0.8, 0.9, 1.0, respectively, in a very good agreement with the values in Fig. 5a at the final time. It is also interesting to discuss the effect of dissipa- tion. Figure 6a shows the results of the LLG simulations similar to those in Fig. 6b, but for η = 0.01 and ∆ω = 0.4, 0.5, 0.6, 0.7. The final amplitudes mz(l/2) = cos θmax of the solutions in this cases were 0.31, 0.13,−0.04 and −0.21. In addition, Fig. 6b shows the solution mz(z/L) found in numerical LLG simulations (solid blue line) at ∆ω = 0.7 and compares it to the form of the exact soliton solution (red dashed line) [2]. The good agreement in the figure shows that the boundary conditions imposed suf- ficiently far from the localized magnetization structure only slightly deform its shape preserving the proximity to the soliton solution, i.e., the autoresonant drive effi- ciently compensates for dissipation. The amplitude of the autoresonant steady-state solu- tions depends on the final driving frequency shift ∆ω. We have found numerically that there exists some crit- ical value ∆ωcr such that stable phase-locked solutions can be excited for ∆ω < ∆ωcr only. One finds that ∆ωcr depends mainly on length l and weakly on ε and α if ε > εth, see Fig. 7. In the dissipationless case, ∆ωcr is close to unity for l > 7 and rapidly decreases for l < 7. Note that l ≈ 7 is near the typical size ∆ξ of the solitons [see Eq. (30)], while the maximum value of ∆ωcr ≈ 1.05 is in agreement with D = 1.05 case in Fig. 9 in the Ap- pendix. In the dissipative case, the autoresonant driving compensates the effect of damping, providing stability of the driven phase-locked soliton. In this case, ∆ωcr has the maximum value of 0.75 at l ≈ 7, yielding the largest amplitude of the autoresonant excitation for this damp- ing parameter. We have also found that with dissipation, ∆ωcr rapidly decreases for l < 7 and that stable solitons do not exist for l < 3 with and without dissipation. We conclude this Section by discussing the realization of the assumed boundary conditions, i.e., that the spins of the nanoparticle are 'pinned' at the boundaries (mz = 1). It turns out that this condition can be realized for free spins as well. Indeed, we can exploit the fact that the proposed excitation approach involves a resonant driving mechanism. Therefore, if one destroys the resonance at the ends of the nanoparticle, the magnetization of the ends will not be affected by the resonant drive and, thus, remain in the direction of the guiding magnetic field H as set initially. For example, if the anisotropy of the end sec- tions of the particle is larger than of its middle part, the ends will remain steady when the driving field is tuned to resonate with the middle section. We illustrate this ap- proach in numerical LLG simulations in Figs. 8a and 8c showing two narrow structures of different height excited in a nanoparticle of length l = 100 with the two end sec- tions of length l/3 each having (normalized) anisotropy constant of 1.25, while the anisotropy constant of the middle l/3 section is 1.0, as before. In addition, Figs. 8b r c ω ∆ 1.2 1 0.8 0.6 0.4 0.2 0 10 20 l 30 40 FIG. 7: The dependence of the critical driving frequency shift ∆ωcr on l: η = 0 (upper line) and η = 0.01 (lower line) for ε = 2εth and α = 0.005. 7 FIG. 8: The excitation of narrow magnetization structures using increased anisotropy of the end sections of the particle to preserve mz = 1 boundary conditions. Panels (a) and (c) show the −mz for ∆ω = 0.7 and 1.0 respectively, while panels (b) and (d) present the corresponding effective potentials V and the quasi-energies A (red horizontal lines) at final time T = 100. and 8d show the effective potential for both excitations and the corresponding quasi-energies, A ≈ 0, as expected for solitary waves. In these simulations, we set mz = 1 at ξ = 0, l initially, but did not impose this condition at later times. The figure shows two solutions of different amplitude excited along the particle for ∆ω = 0.7 and 1.0. The rest of the parameters were α = 0.005, h0 = 5, η = 0.005, ε = 1.2εth. As expected, the boundary con- dition of mz = 1 at the ends of the particle is preserved (the ends are off resonance) and the excited autoreso- nant solutions have a width ∆ξ ≈ 7 [see Eq. (30)] much smaller than the length of the particle. VI. CONCLUSIONS In conclusion, we have suggested and analyzed the au- toresonant approach to excitation of LLG localized solu- tions in easy axis ferromagnetic nanoparticles. The ap- proach involved driving the system under fixed bound- ary conditions by a spatially uniform, but oscillating magnetic field having chirped frequency, which passes through the linear resonance with initially uniform mag- netization. When the driving amplitude exceeded a threshold εth, we have observed formation of large ampli- tude rotating magnetization structures (see Fig. 1). We have used a weakly nonlinear Lagrangian formulation in the dissipationless limit of the model for calculating the threshold [Eq. (17)] and discussed the effect of dissipa- tion on the threshold [Eq. (27)]. All these predictions were in a good agreement with simulations. We have also analyzed fully nonlinear autoresonant quasi-steady- state localized solutions using a simple analog of a quasi- particle in an effective potential (see developments in 0.5 0.45 κ 0.4 κ=k 0.35 0.3 0 D=0.7 0.75 0.8 0.85 0.9 0.95 1.0 1.05 0.05 A 0.1 0.15 FIG. 9: The parameter κ = π/λ versus the quasi-energy A for different driving frequency shifts D and l = 8. Condition κ = k can be satisfied for D < 1.05 only. Sec. V). In these calculations, the decreasing driving fre- quency gradually approached some fixed target value at distance ∆ω from the linear resonance, which also defined the target amplitude of the excited solution. We found numerically that there exists a critical ∆ωcr, such that the excited structures remained stable if ∆ω < ∆ωcr. This critical ∆ωcr depended mainly on the dimensionless length l and damping parameter η and decreased rapidly for l < 7 (the approximate width of the solution). In the dissipationless case, ∆ωcr ≈ 1.05, yielding almost complete inversion of magnetization at the solution max- imum. We have found that the autoresonant drive effec- tively compensates the effect of dissipation on the excited solutions, but the dissipation lowers their amplitude. Fi- nally, we have suggested and illustrated an approach to realization of the assumed fixed boundary conditions by increasing the anisotropy of the end sections of the parti- cle. A further development of the Whitham's-type vari- ational theory in order to explain the stability (seen in simulations) of the chirped-driven large amplitude local- 8 ized solutions is important. Finally, it is known that the dissipationless LLG equation in 1D is integrable and has a large variety of solutions [2]. The autoresonant exci- tation and control of some of these solutions comprises another important goal for future research. VII. ACKNOWLEDGEMENT This work was supported by the Israel Science Foun- dation Grant No. 30/14 and the Russian state program AAAA-A18-118020190095-4. VIII. APPENDIX In calculating the quasi-energy A of the slow evolution of the autoresonant quasi-steady-state, instead of solving Eq. (31) defining λ, we proceed by introducing the action I(A) = Then, since 2 π Z π 0 Re[p2(A − V )]dθ. =(cid:18) dI dA(cid:19)−1 , π λ κ = (33) (34) we calculate the action for different values of Ω′ d = 1 + k2 − D and take its derivative (dI/dA)−1 numerically to find κ(A). The results are shown in Fig. 9 in the case of l = 8 (k = 0.393) and 0.7 6 D 6 1.05. Such calculations and using Ω′ d of form (3) allow to find the quasi-energy A such that κ = k (λ = l) at different times. Note that κ = 0 at the limiting values of A = 0 and Vmax and, therefore, κ has a maximum at some A between 0 and Vmax, as can be seen in Fig. 5 for D = 1.05. Thus, for having the solution with κ = k in the driven problem (k = 0.393 in our example), k must be below this maximum as for D 6 1.0 in Fig. 5, while no such solution exists for D = 1.05. [1] H.B. Braun, Adv. Phys. 61, 1 (2012). [2] A.M. Kosevich, B.A. Ivanov, A.S. Kovalev, Phys. Rep. [9] Zai-Dong Li, Qiu-Yan Li, Lu Li, W.M. Liu, Phys.Rev. E 76, 026605 (2007). 194, 117 (1990). [3] M.M. Scott, M.P. Kostylev, B.A. Kalinikos, C.E. Patton, Phys. Rev. B 71, 174440 (2005). [10] E. M. McMillan, Phys. Rev. 68, 143 (1945). [11] V. I. Veksler, J. Phys. USSR 9, 153 (1945). [12] B. Meerson and L. Friedland, Phys. Rev. A 41, 5233 [4] M. Wu, M. A. Kraemer, M.M. Scott, C.E. Patton, B.A. (1990). Kalinikos, Phys. Rev. B 70, 054402 (2004). [13] H. Maeda, J. Nunkaew, and T. F. Gallagher, Phys. Rev. [5] M. Wu, P.Krivosik, B.A. Kalinikos, C.E. Patton, Phys. A 75, 053417 (2007). Rev. Lett. 96, 227202 (2006). [14] J. Fajans, E. Gilson, and L. Friedland, Phys. Plasmas 6, [6] M. Chen, M.A. Tsankov, J.M. Nash, C.E. Patton, Phys. 4497 (1999). Rev. Lett. 70, 1707 (1993). [15] M.A. Borich, A. G. Shagalov, L. Friedland, Phys. Rev. E [7] A.K. Zvezdin, A.F. Popkov, Sov. Phys. JETP 57, 350 91, 012913 (2015). (1983). [16] L.A. Kalyakin. Russ.Math.Surv. 63, 791 (2008). [8] P.B. He, W.M. Liu, Phys. Rev. B 72, 064410 (2005). [17] M.A. Shamsutdinov, L.A. Kalyakin, A.L. Sukhonosov, A.T. Kharisov, The Physics of Metals and Metallography. 110, 430 (2010). [22] G. B. Whitham, Linear and Nonlinear Waves (Wiley, New York, 1974). [23] R. Feynman, F. L. Vernon, and R. W. Hellwarth, J. Appl. [18] M.A. Shamsutdinov, L.A. Kalyakin, A.T. Kharisov, Phys. 28, 49 (1957). Technical Physics 55, 860 (2010). [19] G. Klughertz, P-A. Hervieux, and G. Manfredi, J. Phys. D: Appl. Phys. 47, 345004 (2014). [20] G. Klughertz, L. Friedland, P-A. Hervieux, and G. Man- fredi, Phys. Rev. B 91, 104433 (2015). [21] L. Friedland, A.G. Shagalov, Phys. Rev. B 99, 014411 (2019). [24] C. Canuto, M. Y. Hussaini, A. Quarteroni, and T. A. Zang, Spectral Methods in Fluid Dynamics (Springer- Verlag, New York, 1988). [25] L. Friedland, Scholarpedia 4, 5473 (2009). [26] W. Wang, Z. Zhang, R. A. Pepper, C. Mu, Y. Zhou, H. Fangohr, J. Phys.: Condens. Matter 30, 015801 (2018). [27] O. Naaman, J. Aumentado, L. Friedland, J.S. Wurtele, I. Siddiqi, Phys. Rev. Lett. 101, 117005 (2008). 9 

1711.06001

1

1711

Hund and anti-Hund rules in circular molecules

We study the validity of Hund's first rule for the spin multiplicity in circular molecules - made of real or artificial atoms such as quantum dots - by considering a perturbative approach in the Coulomb interaction in the extended Hubbard model with both on-site and long-range interactions. In this approximation, we show that an anti-Hund rule {\it always} defines the ground state in a molecule with $4N$ atoms at half-filling. In all other cases (i.e. number of atoms {\it not} multiple of four, or a $4N$ molecule away from half-filling) both the singlet and the triplet outcomes are possible, as determined {primarily} by the total number of electrons in the system. In some instances, the Hund rule is always obeyed and the triplet ground state is realized {\it mathematically} for any values of the on-site and long range interactions, while for other filling situations the singlet is also possible but only if the long-range interactions exceed a certain threshold, relatively to the on-site interaction.

cond-mat.mes-hall

cond-mat

a Hund and anti-Hund rules in circular molecules National Institute of Materials Physics,POB MG-7, 77125, Bucharest-Magurele, Romania M. Nit¸a and M. T¸ olea Department of Physics and Astronomy, Clemson University, Clemson, South Carolina 29634, USA D. C. Marinescu School of Science and Engineering, Reykjavik University, Menntavegur 1, IS-101 Reykjavik, Iceland A. Manolescu We study the validity of Hund's first rule for the spin multiplicity in circular molecules - made of real or artificial atoms such as quantum dots - by considering a perturbative approach in the Coulomb interaction in the extended Hubbard model with both on-site and long-range interactions. In this approximation, we show that an anti-Hund rule always defines the ground state in a molecule with 4N atoms at half-filling. In all other cases (i.e. number of atoms not multiple of four, or a 4N molecule away from half-filling) both the singlet and the triplet outcomes are possible, as determined primarily by the total number of electrons in the system. In some instances, the Hund rule is always obeyed and the triplet ground state is realized mathematically for any values of the on-site and long range interactions, while for other filling situations the singlet is also possible but only if the long-range interactions exceed a certain threshold, relatively to the on-site interaction. I. INTRODUCTION The study of experimental atomic spectra in connec- tion with the theoretical behaviour of quantum particles led Hund to formulate the rules that express the spin and angular momenta of many electron atoms. Hund's first rule (HFR) states that for a given electronic con- figuration, corresponding to incompletely occupied outer orbitals, the state with the maximum spin (i.e. multi- plicity) is the ground state [1, 2]. A textbook example of HFR confirmation is the elec- tron filling of the three degenerate p orbitals in an atomic sub-shell (px, py, pz), when the triplet, rather than the singlet, configuration is realized, as it happens in the C atom. This was explained first by Slater [4] who considered the antisymmetric nature of the elec- tronic wave function that generates a higher value for the Coulomb repulsion in the quantum state with the lower spin value. More recent theoretical analyzes were developed in Refs. [2, 5]. Hund rule's relevance has long exceeded the bound- aries of atomic physics where it was first formulated, over the years being investigated in many other systems such as quantum dots [6 -- 17], artificial molecules created by quantum dots [18 -- 21], metal clusters [23 -- 26], bipartite lattices [27, 28], ultrathin films [29] new carbon systems [30, 31] or even in optical lattices [32, 33]. A further understanding of the physical mechanism be- hind HFR is also offered by the study of physical systems where the rule is reversed, i. the two highest en- ergy electrons form a singlet rather than a triplet. Such a situation is known to exist, for example, in quantum dots, where the zero spin ground state is associated with a spin density wave [7], or in artificial molecules, when the increase of level splitting overcame the exchange en- ergy gain by parallel spin alignment [9]. In semiconduc- e. tor artificial atoms under magnetic field, the Hund rule violation is noticed in connection with changes in the ground state symmetry [34], while in quadratically con- fining quantum dots is related to the modification of the localization properties of some singlet states [16]. Other exceptions to Hund's rule, in close relation with the phenomenology studied in this paper, are known to exist in physical systems that exhibit, as a common fea- ture, degenerate, non-overlapping single-particle states in the mid-spectrum of the electronic Hamiltonian. Such states, that do not have any common sites around the ring, are called disjoint orbitals and have been identified in ring-like molecules [35 -- 38], graphene nanoflakes [27] and small Lieb lattices [28]. The classic example concerning HFR validity in molecules is given by a four-atom molecule such as square cyclobutadiene, for which the Huckel model gives an en- ergy spectrum with four states, two of them being the disjoint orbitals in the middle [36]. The model is equiv- alent with a quadruple quantum dot molecules as de- scribed in [43, 44]. Using a four-electron wave function it was shown that the singlet state has a lower energy on account of Coulomb correlations associated with sin- gle particle excitations which are absent in the triplet state [35]. The quantitative calculation of this result was developed on the base of the spin polarization phenom- ena, where within a self-consistent field approximation, the Brillouin theorem specifies which of the transitions between the many particle spin state are canceled out. When performed for C4H4 and H4 molecules in the sec- ond order of interaction this algorithm yields negative singlet-triplet energy gap in a violation of the HFR [37]. In this paper we discuss the spin properties of circular molecules with an arbitrary number of atoms whose one- particle spectrum, in general, is composed from a ladder of degenerate electronic states [39 -- 42]. This property recommends them as adequate physical systems to inves- tigate the HFR applicability, motivating the significant number of previous studies, as briefly described above. If eigenvectors of the noninteracting Hamiltonian are used to construct the many-particle states of the inter- acting system, the Brillouin theorem as in Ref. [37] is not applicable anymore, but the transition probability can be analytically investigated as done for the Hubbard model in Ref. [28] where the negative singlet-triplet gap for an octagon molecule was shown to result from the two mid-spectrum disjoint orbitals and from the electron-hole symmetry of the spectrum. As a technical detail, we consider the extended Hub- bard model for the general case of a circular molecule by including also a long-range interaction potential, as de- scribed in Section II. The single-particle spectrum which consists of a ladder of double degenerate states is dis- cussed in Section III. The particular case of the molecules with 4N atoms which present a pair of degenerate non- overlapping levels at mid-spectrum is emphasized. Our main formal results are given in Section IV, and then ap- plied to some particular cases of interest in Section V. Section VI concludes the paper. II. CIRCULAR MOLECULE AND THE INTERACTING POTENTIAL In a tight-binding approximation, we describe a cir- cular molecule composed of Ns sites (either real atoms or artificial ones, such as quantum dots) occupied by Ne electrons that interact through a long range Coulomb in- teraction by the Hamiltonian [45], H = −tXn (c† n+1cn + c† ncn+1) + 1 2 Xn,m,σ,σ′ Vnmc† nσc† mσ′ cm,σ′ cnσ, (1) where c† nσ and cnσ are the creation and annihilation op- erators for an electron state of spin σ = ±1/2 at location n = 1,··· , Ns. Every site n can host a maximum of two electrons, of opposite spins. The interaction potential between two electrons local- ized on the sites n and m with coordinates rn, rm is con- sidered within the extended Hubbard model to be given by Vnm = VL rn − rm (1 − δnm) + UH δnm, (2) with VL the long range parameter and UH the Hubbard interaction term. If, say, R1 is the distance between the nearest sites, VL/R1 and UH are measured in the energy unit t of the hopping integral set equal to 1. A ring geometry with Ns = 16 is depicted in Fig. 1. Previously, this model was used in [20, 46, 47] to inves- tigate the interaction effect in quantum dot molecules, in core-shell nanowire with corner localised electric charge 2 y n=3 2 1 x B A ϕδ (A,B) ϕβ (B) ϕγ (A,B) εδ εα,β=0 εγ k=π ϕα (A) k=π/2 k=0 FIG. 1: (up) The circular molecule with Ns = 16. (down) The single particle eigenstates are represented by horizontal lines, while the circles indicate occupied states at half-filling. At mid-spectrum where k = π/2 states ϕα and ϕβ do not share any common site (disjoint) and are occupied by two electrons in the singlet state, thus breaking the first Hund rule. For all other degenerate states, away from half-filling, the Hund rule is discussed in the text. [48, 49], or in discretized quantum rings[22]. It was also found to be a good approximation to describe the elec- tronic dynamics in planar models of circular molecules as cyclobutadiene [35] or cyclooctatetraene [50, 51] when the Hukel model is used. The extended Hubbard model is also used in chemistry in the frame of Pariser-Parr-Pople model Hamiltonian [61]. III. SINGLE PARTICLE STATES It is well known that in the absence of the interaction, the single-particle spectrum of circular molecules consist of a ladder of double degenerate states [39 -- 42]. Here we briefly outline some characteristic properties, useful in the ensuing discussion. Following the notations in Ref. [41] for a quantum ring ❍ ❍ ❍ ❍ ❍ ❍ ❍ ❍ ❍ ❍ ❍ ❍ ❍ ❍ ❍ ❍ one can express the energy of the twice degenerate states to be ǫk = −2t cos k, where k = 2πl/Ns (l = 1, 2, Ns/2 − 1) is the wavevector (with R1 the distance between the sites set equal to unity). The associated eigenstates are, k i = r 2 ϕ(1) k i = r 2 ϕ(2) Ns Xn Ns Xn sin nkni, cos nkni. (3) (4) In Fig. 1 this result is shown for Ns = 16 sites. For k = π/2 and l = Ns/4, the eigenstates identified above, located at energy ǫ(π/2) = 0, satisfy hnϕ(1) π/2ihϕ(2) π/2ni = 0, for all n ∈ [1, Ns], (5) Fig. 1, it follows that ϕ(1) tion, and ϕ(2) which means that they do not share any common site. If we think of the quantum ring as a bipartite lattice with A sites for n odd and B sites for n even as sketched in π/2i has only A sites localiza- π/2i has only B sites localization. Such non- overlapping states are also present in the flat band of a Lieb lattice [52] or as different localized edge states in 2D materials [53 -- 55]. In the frame of the molecular Huckel model used in the field of quantum chemistry, this situa- tion defines the disjoint non-bonding orbitals [36, 56, 57] which have relevance for spin properties at half-filling. For wave numbers k = 0 and k = π (l = 0, Ns/2) the double degeneracy of the spectrum is lifted as the single particle energies are ǫ0 = −2t and ǫπ = +2t, respectively. In this case ϕ(1) k i (Eq. 4) is the only good eigenstate, which upon normalization becomes, k i is zero, while ϕ(2) 0 i = r 1 ϕ(2) π i = r 1 ϕ(2) Ns Xn Ns Xn ni, (−1)nni. (6) (7) IV. THE INTERACTING GROUND STATE In this section, we investigate the applicability of the HFR for a pair of electrons that occupy the top states in the single particle spectrum identified above when several electronic occupancies are realized in a molecule with Ns sites in the presence of the Coulomb interaction. As such, we evaluate only the differences between the lowest en- ergies of the interacting system when a pair of electron spins form a singlet state, corresponding to total spin momentum S = 0 or a triplet with S = 1. With E0 and E1 denoting the lowest energies in the spin sectors S = 0 and S = 1 respectively, we define the magnetic energy or the singlet-triplet gap as: 3 In a perturbative approach, ∆E is obtained, in a first order approximation, to be equal to the exchange en- ergy associated with the Coulomb interaction between the parallel spins in the triplet configuration [28]. If the exchange energy is zero, a second order calculation in the Coulomb interaction is performed. In the eigenfunction representation Eqs. 3 and 4, the Coulomb matrix element for any four single particle quantum states, ϕα, ϕβ, ϕγ and ϕδ, is written as Vαβ,γδ, Vαβ,γδ = Xn1,n2 ϕα(n1)⋆ϕβ(n2)⋆Vn1n2 ϕγ(n1)ϕδ(n2), (9) with n1 and n2 counting the positions from 1 to Ns of the two electrons in the system and Vn1,n2 from Eq. (2). First, we consider the case of a quantum molecule with Ne number of electrons that fill the energy levels with every k ≤ k0, with the last two electrons occupying the degenerate states ϕ(1) k0 with a given k0 6= 0, π/2 or π. Actually k0 = 0 or π correspond to the lowest and highest non-degenerate levels, not of interest for our discussion, and k0 = π/2 is discussed later in this Section. The singlet-triplet gap ∆E is equal to the exchange k0 and ϕ(2) energy Vαβ,βα, where ϕα = ϕ(1) k0 and ϕβ = ϕ(2) k0 , [28]: ∆E = 2Vαβ,βα. (10) From Eqs. (3), (4) and (9), Vαβ,βα = (cid:18) 2 Ns(cid:19)2 × Ns Xn1,n2=1 sin n1k0 cos n2k0Vn1n2 cos n1k0 sin n2k0, (11) which can be written as a difference of two terms: Vαβ,βα = 1 2N 2 s Ns Xn1,n2=1 cos 2(n1 − n2)k0Vn1n2 1 2N 2 s Ns Xn1,n2=1 − cos 2(n1 + n2)k0Vn1n2. (12) The first term in Eq. 12 is just the Fourier transform 1 2 V (2k0) (see also Appendix). The second term can be shown to vanish for all the allowed values of the wave vector k0 = 2πl/Ns, except for k0 = 0, π/2 and π. For k0 = π/2, should such a value exist in spectrum (for 4N molecules), the second term cancels exactly the first one and an evaluation in the second order of the Coulomb interaction is needed. Leaving this single exception aside for the moment, the spin splitting energy (Eq. 10) is therefore given by the Fourier transform for the wave number 2k0: ∆E = V (2k0). (13) ∆E = E0 − E1. The above equation is one of the main formal results of our paper. Consequently, the sign of V (2k0) determines (8) 0 = H U r o f ) q ( V 1.8 1.6 1.4 1.2 1 0.8 0.6 0.4 0.2 0 -0.2 -0.4 NS=160 NS=12 π/3 0 0.5 1 q/π 5π/3 1.5 2 FIG. 2: Long-range part of V(q) (constants disregarded i.e. R = 1 and VL = 1) for Ns = 12 and Ns = 160. The physically allowed values for the wave number are multiples of 2π . We Ns notice V (π) → − ln 2/π. the spin configuration in the ground state, since ∆E > 0 means a triplet ground state and ∆E < 0 a singlet ground state. Since the sign of the function V (q) dictates the spin of the ground state, a legitimate question is wether neg- ative values can be obtained - i.e. singlet ground state and the anti-Hund rule, or is V (q) always positive. The minimum value of V (q) can be inferred by formally con- sidering q as a continuous variable in Eq. A4 which upon differentiation generates qmin = π [62]. An interesting analytical result is obtained from the large Ns limit. We can consider Ns = even, since only for even number of sites the wave vector k0 = π/2 is a physically allowed value. One can show that, for Ns = even (see Appendix): 1 VL V (π) → Ns(cid:2)UH − 2 ln 2 ∆ (cid:3), for Ns → ∞, ∆ = 2πR/Ns. (14) This implies that a given Fourier component like V (2k0) is always positive for any UH > 2 ln 2 VL ∆ ≃ 1.386 VL ∆ . This represents, for instance, reasonable val- ues of the Hubbard and long range parameters ratio for an artificial quantum dot arrays model used in [28]. In this case the spin energy gap from Eq. 13 is always pos- itive (∆E > 0). The examples presented in the next Section all suggest that the triplet is ground state for physically reasonable reasons. However, as Eq. (13) indi- cates, mathematically situations with a preferred singlet ground state are possible. To separate the two instances, it is insightful to plot the long-range part of V (q), i.e. for UH = 0. From Fig. 2, one notices that for q < π/3 or q > 5π/3, V (q) > 0 while for the middle interval q ∈ (π/3, 5π/3) the long range part of V (q) takes negative values, which 4 are to be compared with the on-site Hubbard part (which is always positive) in order to decide the sign of ∆E. In conclusion when the double degenerate states with the wave number k 6= π/2 are occupied with the last two electrons, the ground state is decided by the sign of the exchange energy in the triplet configuration. The triplet is always the ground state if k ∈ (0, π/6) or k ∈ (5π/6, π), for any values of the interaction parameters UH or VL. For intermediate values k ∈ [π/6, 5π/6], the singlet can become ground state if the long range interaction exceeds a k dependent threshold value (relatively to UH ). A significant exception to this rule, as mentioned pre- viously, occurs in the case of a molecule whose number of sites Ns is a multiple of four, occupied by Ne = Ns. In this case the two mid spectrum states with k0 = π/2 are occupied by two electrons. The exchange energy van- ishes for this case and the perturbative calculation in the Coulomb potential must be carried out in the second or- der to determine ∆E. As we show below, the singlet state is always the interacting ground state of the system, as ∆E < 0, and an anti-Hund rule situation is obtained. For the beginning, we consider a single particle excita- tion process from one state with wave number π/2− q′ < k0 to another state with wave number π/2 + q > k0. These states need to have the same symmetry properties to allow single particle excitation between them. With the notations ϕα = ϕ(1) , ϕβ = ϕ(2) 2 −q′ and ϕδ = ϕ(1) 2 +q we have the following formula for the spin splitting energy in the second order of perturbation [28]: π , ϕγ = ϕ(1) π π 2 π 2 ∆E = 2Vαβ,βα + 4Vδα,αγVδβ,βγ ∆δ,γ , (15) where ∆δ,γ is the excitation energy ∆δ,γ = ǫδ−ǫγ. A sim- ilar formula as Eq. 15 can be obtained with self consistent orbitals from Ref. [37] if one consider that the singlet and triplet wave functions are the same. The Coulomb matrix elements that enter in Eq. 15 are obtained straightforwardly: Vαβ,βα = 0, 1 2 Vδα,αγ = Vδβ,βγ = −Vδα,αγ, [V (q) + V (π − q)] δq,q′ , (16) (17) (18) for q, q′ ∈ (0, π/2). The first order cancellation in Eq. 16 is readily obtained when using the disjointness relation of the two states with wave number k0 = π/2 from Eq. 5. For Eqs. 17 and 18, after further arrangements, we use the summation of the Fourier transformations from Eq. A5. The negative sign in Eq. 18 is the one that will lead to negative splitting energy and HFR violation. If we consider now the single particle excitation be- tween the cosine functions ϕγ = ϕ(2) 2 −q′ and ϕδ = ϕ(2) 2 +q we find out that the Coulomb matrix elements Vδα,αγ and Vδβ,βγ only change the sign compared to those gen- erated by the sine functions. The difference is that in π π this case q, q′ can have also the value π/2 corresponding to the transition between the two extreme energy states from Eqs.6 and 7, which means that q, q′ ∈ (0, π/2]. We are holding now all possible single particle tran- sition processes between the states with wave numbers π/2 − q and π/2 + q for any possible value of q. Using above considerations in Eq. 15 and summing the terms for all pairs of single particle states ϕγ, ϕδ we obtain the following relation for the spin energies splitting: ∆E = − 2(V (π/2))2 ǫ0 −Xq (V (q) + V (π − q))2 ǫ π 2 −q (19) with ǫπ/2−q and ǫ0 the single particle energies. The first term of the above equation accounts for exci- tations from the lowest non-degenerate state (k0 = 0) to the highest one (k0 = π), which is also non-degenerate. In the case of the four atom molecule, it is the only term existent. For all the other 4N molecules with N ≥ 2, the second term must be considered as well, taking ac- count for the allowed excitations between double de- generate states symmetrically placed below and above the mid-spectrum. The summation is over the values q = 2π . Eq. 19 therefore shows a nega- Ns tive sign of the spin splitting energy (i.e. ∆E < 0) and therefore a singlet ground state and anti-Hund situation for the half-filled 4N molecule. ,··· , (Ns−4)π , 4π Ns 2Ns V. EXAMPLES In this Section, we show calculations of the singlet- triplet level spacing for some simple molecules (either made of atoms or of quantum dots), using for the Hub- bard or long-range interactions values or formulas pro- posed in literature. Two situations when the Hund or anti-Hund situations are decided by the ratio between Hubbard and the long- range interactions are shown in Table I for a triangle molecule and for an octagon molecule at various filling factors except for the half-filling. For an artificial molecule constructed with quantum dots, one may use for instance the dot confinement model described in [58] where the authors calculate the interac- tion parameters as: UH = V (Rn) = e2 , 2√2πǫd e2 4πǫRn (20) (21) , with d the dot diameter and Rn the inter-dot distance of order n. One can modify the V (Rn) and UH by varying either the dots diameter or inter-dot distances. Using Eqs. 20 and 21 the energy splitting for triangle molecule in Table I is always positive for the dot con- finement d < R1√2π ≃ 2.5R1 which is true in this case. 5 System Electron Configuration ∆E = ES − ET t>0 t<0 ∆E = 1 3 (UH − V (R1)) t>0 t<0 ∆E = (UH + V (R4)) 1 8 − 1 4 V (R2) TABLE I: Singlet-triplet splitting when the last two degener- ate occupied orbitals are away of half-filling. The calculations are done in the first order approximation. We mention also that in [59] the exact results for a tri- angle are calculated and our results are recovered in the limit UH − V (R1) ≪ t. As a matter of fact, using the condition from Eq. 14, it is easy to show that the triplet state is always the ground state away from half-filling (as situations in Table I) for the dot diameter d < R1 when considerring the model in Eqs. 20 and 21. The above values for UH and V (Rn) may also be used to compute, for instance, the singlet-triplet splitting in half-filled 4N molecules, when one has always an anti- Hund rule. For the square and the octagon, the results are given in Table II. System Electron configuration Singlet-Triplet Splitting ∆E = ES − ET ∆E = − 1 16t (UH − V (R2))2 1 (UH − V (R4))2 16√2t (UH + V (R4) − 2V (R2))2 ∆E = − − 1 64t TABLE II: Negative singlet-triplet splitting for the half-filled square and octagon molecules. We now present results obtained by the formula of ∆E from Table II to approximate the singlet-triplet energy splitting in the case of chemical molecules. As an example we consider the square model of cyclobutadiene molecule which has Ns = 4 Carbon atoms. We use standard pa- rameters of hydrocarbures from [60, 61], UH = 11.26eV, hopping energy t = 2.4eV , and distance between atoms R1 = 1.44A. The long range interaction is calculated ❍ ❍ ❍ ❍ ❍ ❍ ❍ ❍ ❍ ❍ ❍ ❍ ❍ ❍ ❍ ❍ ❍ ❍ ❍ ❍ ❍ ❍ ❍ ❍ ❍ ❍ ❍ ❍ ❍ ❍ ❍ ❍ ❍ ❍ now with Pariser-Parr-Pople model Hamiltonian [63] and Ohno formula [61] V (Rn) = 14.397 r(cid:16) 14.397 UH (cid:17)2 , + R2 n (22) with V (Rn) and UH in eV and Rn in A. From the for- mula of Table I we obtain the value ∆E = −0.72eV for a square model of cyclobutadiene, close to numerical val- ues ∆ESP = −0.71eV obtained in [37] considerring the spin polarization effect and minimal base for molecular calculation. For the planar model of cyclooctatetraene molecule with Ns = 8 carbon atoms, if we keep the same pa- rameters as above, except R1 = 1.40A, we obtain ∆E = −0.88eV that is lower than other values reported in liter- ature, but comparable as ∆EST = −0.68eV or −0.34eV in [64]. The perturbative approach described in this paper, al- though valid only for small values of the interacting po- tential, offers a good qualitative description of the spin configuration of the ground state in the case of circular molecules with a large number of atoms. Full analytical results in the presence of interaction are only available for the smallest N = 3 molecule [59], while exact diagonal- ization numerical results are already somewhat compu- tationally demanding even for the N = 8 (see [28]). Such methods are not at all feasible for large N . In a direct comparison, for the smaller circular molecule with N = 3, our perturbative result from Table 1 differs from the ex- act result [59] by less than 2 % for UH−V (R1) < 0.1t and less than 6% for UH − V (R1) < 0.3t. In [28], for N = 8, if only the on-site Hubbard interaction is considered, we find good correspondence between the perturbative re- sults and the exact diagonalization ones for UH < t. VI. CONCLUSION In this paper we have studied the first Hund rule in circular molecules, for cases when the two most ener- getic electrons occupy a pair of degenerate levels. The quantity of interest is the singlet-triplet energy gap ∆E, which was expressed in terms of the Fourier transform of the interacting potential. Both on-site (UH ) and long range (VL) interactions have been considered within an extended Hubbard model. A special case is found for the 4N molecule at half filling, for which the first order energy correction (i.e. the exchange energy) vanishes and the second order gives always the singlet as ground state, and thus an anti-Hund situation. Since the 4N molecule is a bipartite lattice, we find ourselves in the frame of the Lieb theorem [65], but with a more complex potential including arbitrary long range interaction. For all the other cases, the exchange energy does not vanish and its sign decides the ground state. Our results 6 show that, depending on the total number of electrons in the system (i.e. the wave number k0 of the highest occu- pied levels) we meet the two distinct situations. A triplet ground state is realized for any values of interaction pa- rameters if k0 ∈ (0, π/6) or k0 ∈ (5π/6, π). On the other hand, for k0 ∈ [π/6, π/2)∪ (π/2, 5π/6] the singlet ground state is mathematically possible, with the highest prob- ability for k0 close to π/2. A necessary condition for sin- glet ground state around k0 = π/2 is VL/∆ > UH/(2 ln 2) (∆ is the nearest neighbors distance measured on the cir- cle, i.e. ∆ = 2πR/Ns). The described formalism is applied for some few-atoms circular molecules, either real or artificial, in Section V. The results hold for arbitrary Hubbard or long range interactions, as well as for any number of atoms in the circular molecule. Such generality is owed to the fact that the singlet-triplet level spacing was analytically expressed in terms of the Fourier transform of the interaction po- tential. Apart from providing detailed spectral calculations for molecules of potential interest, our studies may be also relevant for understanding various origins of non-trivial spin alignment. Acknowledgements. Thw work is supported by the Core Program grant No. PN16-480101 and National Re- search Program PN-III-P4-IDPCE-2016-0221. Appendix A: The Fourier transform of the interaction potential [the V(k) function] The long range part of the potential Vnm in Eq. 2 de- pends only on the distance Rnm between the points n, m that, for the ring geometry, counts only the minimum number of sites from n to m. Then we can define the potential V (Rn) for any integer n: V (Rn) = VL Rn (1 − δRn,0) + UHδRn,0 with Rn = 2R sin πn/Ns, (A1) with the length Rn measuring the distance between two points separated by n succesive sites on a circle of radius R. The potential V (Rn) from Eq. A1 has the periodicity V (Rn) = V (Rn+Ns) and we define the Fourier transfor- mation V (k) = 1 Ns Ns Xn=1 eiknV (Rn), (A2) with the wave number k = 2π Ns l with l integer. In the calculation of ground state properties from Sec- tion IV we use the following properties of V (k): V (k) = V (k)⋆, Ns V (k) = 1 Ns cos knV (Rn) Xn=1 = UH Ns + VL Ns Ns−1 Xn=1 cos kn Rn . (A3) (A4) The Eq. A3 is immediate using V (Rn) = V (RNs−n) in Eq. A2 and Eq. A4 follows from Eq. A3 using also the explicit form of the potential from Eq. A1. Using the definition of the Fourier transform from Eq. A2 we obtain the relation V (q) + V (π − q) = 2 Ns Ns Xn=2(even) cos qnV (Rn) (A5) that is used to obtain Eqs. 17 and 18. As mentioned also in the main text, if we treat k as a continuous variable, then the derivative of Eq.A4 cancels for k = π, where the function has a minimum (as seen in Fig.2). Whether this minimum is negative or remains positive, depends on the ratio VL/UH . In order to calcu- late V (π) in the limit of large number of sites (Ns → ∞) one evaluates, up to a constant: 7 lim N→∞h π N N −1 Xn=1 cos πn sin πn/Ni. This is done by taking into account that, lim N→∞ N Xn=1 (−1)n n = − ln 2 , (A6) (A7) an equality that reproduces the first terms in Eq. A6, since for small n, cos nπ = (−1)n while sin πn/N ≃ πn/N in the limit N → ∞. Terms calculated for in- termediate values of n generate vanishing contributions. The last term in Eq. A6 (n → N ) reproduces in magni- tude terms present in Eq. A7, as sin (π − a) = sin a. Whether the terms are reproduced with the same sign or opposite one is decided by the parity of N . As a result, the limit of Eq. A6 is −2 ln 2 for N = even and 0 for N = odd. This proves Eq. 14. [1] F. Hund, Z. Phys. 33, 345, 855 (1925). [2] W. Kutzelnigg, J. D. Morgan, Z. Phys. D 36, 197 (1996). [3] W. Pauli, Z. Phys. A 31, 765 (1925). [4] J. C. Slater, Phys. Rev. 34, 1293 (1929). [5] G. Gryn'ova, M. L. Coote, and C. Corminboeuf, WIREs Comput. Mol. Sci. 5, 440 (2015). [6] S. Tarucha, D. G. Austing, T. Honda, R. J. van der Hage, and L. P. Kouwenhoven Phys. Rev. Lett. 77, 3613 (1996). [7] M. Koskinen, M. Manninen, and S. M. Reimann, Phys. Rev. Lett. 79, 1389 (1997). 1283 (2002). [19] T. Ota, M. Rontani, S. Tarucha, Y. Nakata, H. Z. Song, T. Miyazawa, T. Usuki, M. Takatsu, and N. Yokoyama, Phys. Rev. Lett. 95, 236801 (2005). [20] M. Korkusinski, I. P. Gimenez, P. Hawrylak, L. Gau- dreau, S. A. Studenikin, and A. S. Sachrajda, Phys. Rev. B 75, 115301 (2007). [21] S. Florens, A. Freyn, N. Roch, W. Wernsdorfer, F. Bale- stro, P. Roura-Bas, and A. A. Aligia, J. Phys.: Condens. Matter 23 243202 (2011). [8] O. Steffens, U. Rossler, and M. Suhrke, Europhys. Lett. [22] C. Daday, A. Manolescu, D. C. Marinescu, and Vidar 42, 529 (1998). Gudmundsson, Phys. Rev. B 84, 115311 (2011). [9] B. Partoens and F. M. Peeters, Phys. Rev. Lett. 84, 4433 [23] V. Kumar and Y. Kawazoe, Phys. Rev. B 64, 115405, (2000). (2001). [10] P. Matagne, J. P. Leburton, D. G. Austing, and S. Tarucha, Phys. Rev. B 65, 085325 (2002). [11] L. He, G. Bester, and A. Zunger Phys. Rev. Lett. 95, [24] D. Yoshida and H. Raebiger, Sci. Rep. 5, 15760 (2015). [25] P. Sen, J. Clust. Sci. 27, 795 (2016). [26] A. C. Reber, V. Chauhan, and S. N. Khanna, J. Chem. 246804, (2005). Phys. 146, 024302 (2017). [12] Y. Sajeev, M. Sindelka, and N. Moiseyev, J. Chem. Phys. [27] W. Sheng, M. Sun, and A. Zhou, Phys. Rev. B 88, 085432 128, 061101 (2008). (2013). [13] T. Sako, J. Paldus, and G. H. F. Diercksen, Phys. Rev. A 81, 022501 (2010). [28] M. T¸ olea and M. Nit¸a, Phys. Rev. B 94, 165103 (2016). [29] A. Belabbes, G. Bihlmayer, F. Bechstedt, S. Blgel, and [14] T. Sako, J. Paldus, A. Ichimura, and G. H. F. Diercksen, A. Manchon Phys. Rev. Lett. 117, 247202 (2016). J. Phys. B: At. Mol. Opt. Phys. 45 235001 (2012). [30] F. Zhang, D. Tilahun, and A. H. MacDonald Phys. Rev. [15] J. Katriel and H.E. Montgomery Eur. Phys. J. B 85 394 B 85, 165139 (2012). (2012). [31] H.-C. Jiang and S. Kivelson Phys. Rev. B 93, 165406 [16] S. Schroter, H. Friedrich, and J. Madronero, Phys. Rev. (2016). A 87, 042507 (2013). [17] A. Hofmann, V. F. Maisi, C. Gold, T. Krhenmann, C. Rssler, J. Basset, P. Mrki, C. Reichl, W. Wegscheider, K. Ensslin, and T. Ihn, Phys. Rev. Lett. 117, 206803 (2016). [32] A. F. Ho, Phys. Rev. A 73, 061601(R) (2006). [33] K. Karkkainen, M. Borgh, M. Manninen, and S. M. Reimann, New J. Phys. 9, 33 (2007). [34] C. F. Destefani, J. D. M. Vianna and G. E. Marques, Semicond. Sci. Technol. 19 L90 (2004). [18] S. M. Reimann and M. Manninen, Rev. Mod. Phys. 74 [35] W. T. Borden, J. Am. Chem. Soc. 97, 5968 (1975). 8 [36] W. T. Borden, H. Iwamura, and J. A. Berson, Acc. Chem. 125428 (2013). Res. 27, 109 (1994). [37] H. Kollmar and V. Staemmler, Theoret. Chim. Acta [53] A. Onipko, Phys. Rev. B 78, 245412 (2008). [54] B. Ostahie, M. Nit¸a, and A. Aldea, Phys. Rev. B 91, (Berl.) 48, 223 (1978). 155409 (2015). [38] S. Zilberg and Y. Haas, J. Phys. Chem. A, 102, 10851 [55] B. Ostahie and A. Aldea, Phys. Rev. B 93, 075408 (1998). (2016). [39] L. Salem, "The molecular orbital theory of conjugated systems", publisher: W. A. Benjamin 1966. [40] Y. Jones, R. Arnold, Physical and Mechanistic Organic Chemistry, Cambridge University Press 1979. [56] P. W. Atkins and R. S. Friedman, Molecular Quan- tum Mechanics, Third edition, Oxfort University Press (2001). [57] W. T. Borden and E. R. Davidson, J. Am. Chem. Soc., [41] Sigridur Sif Gylfadottira, Marian Nit¸a, Vidar Gudmunds- 99, 4587 (1976). son, Andrei Manolescu, Physica E 27, 278 (2005). [58] H. Tamura, K. Shiraishi, T. Kimura, and H. Takayanagi, [42] J. M. Mercero A. I. Boldyrev, G. Merino and J. M. Phys. Rev. B 65, 085324 (2002). Ugalde, Chem. Soc. Rev. 44, 6519 (2015). [59] M. Niklas, A. Trottmann, A. Donarini, and M. Grifoni, [43] R. Thalineau, S. Hermelin, A. D. Wieck, C. Bauerle, L. Saminadayar, and T. Meunier, Appl. Phys. Lett. 101, 103102 (2012). Rhys. Rev. B 95, 115133 (2017). [60] Y. Anusooya and Z. G. Soos, Current Science, 75, 1233 (1998). [44] I. Ozfidan, A. H. Trojnar, M. Korkusinski, and P. Hawry- [61] S. Sahoo, V. M. L. Durga Prasad Goli, S. Ramasesha and lak, Solid State Comm. 172, 15 (2013). [45] G. D. Mahan, Many-Particle Physics, Kluwer Aca- demic/Plenum Publishers, New York (2000). [46] B. R. Bu lka, T. Kostyrko, and J. Luczak, Phys. Rev. B 83, 035301 (2011). [47] K. Wrzeniewski and I. Weymann, Phys. Rev. B 92, 045407 (2015). [48] I. Ozfidan, M. Vladisavljevic, M. Korkusinski, and P. Hawrylak, Phys. Rev. B 92, 245304 (2015). [49] A. Sitek, M. T¸ olea, M. Nit¸a, L. Serra, V. Gudmundsson and A. Manolescu, Scientific Reports 7, 40197 (2017). [50] C. Trindle and T. Wolfskill, J. Org. Chem. 56, 5426 (1991). D. Sen, J. Phys. Condens. Matter 24 115601 (2012). [62] For the present discussion, which temporarily regards q as a continuous variable, it is irrelevant that the value q = π is precisely the exception discussed in text for which Eq. 13 does not give the energy splitting. [63] R. Pariser and R. G. Parr, J. Chem. Phys. 21 767 (1953); J. A. Pople, Trans. Faraday Soc. 49 1375 (1953); K. Ohno, Theor. Chim. Acta 2 219 (1964). [64] D. A. Hrovat, W. T. Borden J. Am. Chem. Soc.114, 5879 (1992); P. G. Wenthold, D. A. Hrovat, W. T. Borden and W. C. Lineberger, Science, 272, 1456 (1996); W. C. Lineberger and W. T. Borden, Phys. Chem. Chem. Phys. 13, 11792 (2011). [51] T. Nishinaga, T. Ohmae and M. Iyoda, Symmetry 2, 76 [65] E. Lieb, Phys.Rev.Lett. 62, 1201 (1989). (2010). [52] M. Nit¸a, B. Ostahie, and A. Aldea, Phys. Rev. B 87, 

1608.02789

1

1608

Efficient simulations with electronic open boundaries

We present a reformulation of the Hairy Probe method for introducing electronic open boundaries that is appropriate for steady state calculations involving non-orthogonal atomic basis sets. As a check on the correctness of the method we investigate a perfect atomic wire of Cu atoms, and a perfect non-orthogonal chain of H atoms. For both atom chains we find that the conductance has a value of exactly one quantum unit, and that this is rather insensitive to the strength of coupling of the probes to the system, provided values of the coupling are of the same order as the mean inter-level spacing of the system without probes. For the Cu atom chain we find in addition that away from the regions with probes attached, the potential in the wire is uniform, while within them it follows a predicted exponential variation with position. We then apply the method to an initial investigation of the suitability of graphene as a contact material for molecular electronics. We perform calculations on a carbon nanoribbon to determine the correct coupling strength of the probes to the graphene, and obtain a conductance of about two quantum units corresponding to two bands crossing the Fermi surface. We then compute the current through a benzene molecule attached to two graphene contacts and find only a very weak current because of the disruption of the $\pi$-conjugation by the covalent bond between the benzene and the graphene. In all cases we find that very strong or weak probe couplings suppress the current.

cond-mat.mes-hall

cond-mat

Efficient simulations with electronic open boundaries Andrew P. Horsfield1, Max Boleininger2, Roberto D'Agosta3,4, Vyas Iyer1, Aaron Thong1, Tchavdar N. Todorov5, Catherine White1. 1Department of Materials and Thomas Young Centre, Imperial College London, South Kensington Campus, London SW7 2AZ, U.K. 2Department of Physics and Thomas Young Centre, Imperial College London, South Kensington Campus, London SW7 2AZ, U.K. 3Nano-Bio Spectroscopy Group and ETSF, Universidad del País Vasco, CFM CSIC-UPV/EHU , 20018 San Sebastián, Spain 4IKERBASQUE, Basque Foundation for Science, E-48013 Bilbao, Spain and 5Atomistic Simulation Centre, School of Mathematics and Physics, Queen's University Belfast, Belfast BT7 1NN, U.K.∗ We present a reformulation of the Hairy Probe method for introducing electronic open boundaries that is appropriate for steady state calculations involving non-orthogonal atomic basis sets. As a check on the correctness of the method we investigate a perfect atomic wire of Cu atoms, and a perfect non-orthogonal chain of H atoms. For both atom chains we find that the conductance has a value of exactly one quantum unit, and that this is rather insensitive to the strength of coupling of the probes to the system, provided values of the coupling are of the same order as the mean inter-level spacing of the system without probes. For the Cu atom chain we find in addition that away from the regions with probes attached, the potential in the wire is uniform, while within them it follows a predicted exponential variation with position. We then apply the method to an initial investigation of the suitability of graphene as a contact material for molecular electronics. We perform calculations on a carbon nanoribbon to determine the correct coupling strength of the probes to the graphene, and obtain a conductance of about two quantum units corresponding to two bands crossing the Fermi surface. We then compute the current through a benzene molecule attached to two graphene contacts and find only a very weak current because of the disruption of the π-conjugation by the covalent bond between the benzene and the graphene. In all cases we find that very strong or weak probe couplings suppress the current. I. INTRODUCTION Atomic scale computer simulations of nanoscale sys- tems of necessity have to approximate the environment that the system finds itself in as it is of unlimited size. One way to incorporate a model environment is through the boundary conditions of the system being treated ex- plicitly. Here we focus on the boundary conditions for the electrons. Traditional choices include: free bound- aries, where the system is treated as an isolated cluster in vacuum; periodic boundaries, where the system plus its near environment are repeated periodically to make an effectively infinite system; and open boundaries, where the system is finite but electrons can enter and leave as though connected to an external reservoir. The correct choice of boundary conditions is determined by the prob- lem being addressed. There exist very efficient algorithms for free and peri- odic boundary atomistic simulations [1], and these will not be considered further here. Open boundaries are im- portant for a number of problems [2], and mature open boundary codes also exist [3–6]. However, relative to free and periodic boundaries, they tend to be more compu- tationally expensive to implement, and simulations can require more human effort to set up. These technical con- ∗ a.horsfield@imperial.ac.uk siderations tend to limit the range of problems addressed, often to molecular conduction, whereas if they could be overcome new areas would become accessible, such as electrochemistry. Our purpose here is to map out a pos- sible way forward by extending a light weight scattering theory scheme known as Hairy Probes [7] to systems more general than those to which it was originally applied, and to show that simulations can be made computationally efficient and easy to set up. Hairy Probes originally was designed to address time dependent problems; here we only consider the case of steady state current and static atoms. Using an Empirical Tight Binding (ETB) model we investigate a Cu atomic wire, and then using a Density Functional Tight Binding (DFTB) model [8], we apply the method to the study of a chain of H atoms. These two simple, but well understood systems, allow us to in- vestigate the correctness of the method. For both the Cu and H wires we get ideal ballistic conductance pro- vided the strength of the coupling to the probes is neither too large nor too small: extreme couplings suppress the current. We then look at current flow through a ben- zene molecule between two graphene contacts as a way to investigate the properties of graphene as a contact for molecular electronics [9, 10]. We find that the presence of a covalent bond between the benzene and the graphene suppresses the current as it disrupts the π-conjugation. As preparation for this calculation, current flow through a carbon nanoribbon is studied to find the correct cou- pling strength for the probes to the graphene. We obtain a conductance of slightly less than two quantum units corresponding to two bands crossing the Fermi level. II. FORMALISM The Hairy Probes formalism was originally introduced for orthogonal tight binding models, and covered both static and time dependent simulations [7]. Here we gen- eralize the static limit to the non-orthogonal case [11], summarizing the key steps in the theory. The expres- sions are derived using the Lippmann-Schwinger formal- ism [12, 13], which is equivalent to using non-equilibrium Green's functions (NEGF) for non-interacting or mean field Hamiltonians [14]. We note that this method has a number of similari- ties with the sink-source potential method [15–17]; how- ever, additional simplifications allow for arbitrary bias, any number of terminals, and full self-consistency. Sim- ilar simplifications have also been achieved previously by applying the wide band limit directly to the leads [18]. However, we note that Hairy Probes can deliver accurate results for low dimensional systems, and charge self-consistency can be introduced straightforwardly, as demonstrated below. The starting point is to imagine that our system is connected to one or more particle reservoirs by a set of atomically thin leads (which we call probes) that each attach to just one atomic orbital in the system of interest. The reservoir of electrons from which a probe emerges is characterized by a chemical potential and a temperature for the electrons. Each probe, then, is a bit like a wire attached at one end to a terminal of a battery, and at the other end attached by a kind of alligator clip to an atomic orbital. Each probe thus corresponds to both a source of incoming electrons of given chemical potential and temperature, and a channel for outgoing electrons. In practice, the probes are attached to contact regions in much the same way that leads are attached to con- tacts in many NEGF calculations: see Fig. 1. However, because the probes are not system specific, we can define them in a manner that is computationally convenient. Thus there is no need to compute surface Green's func- tions, the embedding self-energy can be made energy in- dependent while avoiding imposing the wide band limit directly to the leads, and the mean field self-consistent potential profile is taken care of automatically. These at- tributes are what enable the Hairy Probe formalism to be easy to use (you just need to specify where the probes are to be attached, and how strongly, but do not have to build Green's functions for the leads), and computation- ally very efficient (an effective Hamiltonian is produced that can be diagonalized, and all subsequent integrals can then be performed analytically). We note that it is shown in [7] that in the limit of long electrodes and small coupling the Hairy Probes steady state reduces to the conventional 2-terminal Landauer picture. 2 Figure 1. The arrangement for a calculation involving hairy probes. The argument we present here that leads to the Hairy Probes equations is based on the Lippmann-Schwinger formulation of scattering theory. That is, we treat each probe as transporting independent electrons from a reser- voir, with the electron wavefunctions being viewed as scattering states that travel down the probes and scat- ter off the system of interest, being partially transmitted (producing a current) and partially reflected. As we employ ETB and DFTB models, the basis set used to expand the single particle wave functions is com- posed of atomic orbitals. Let our atomic basis set be denoted by α(cid:105) where α is a combined index that spans both atomic sites and orbitals. We can now define the Hamiltonian and overlap matrices by Hαα(cid:48) = and Sαα(cid:48) = (cid:104)α α(cid:48)(cid:105), respectively. Note that the Hamilto- nian includes all the terms associated with self-consistent charge redistribution [8, 19]. We partition these orbitals between the system (β(cid:105)) and the probes (pγp(cid:105)), where p is the index of the probe and γp indexes an orbital in probe p. (cid:12)(cid:12)(cid:12) α(cid:48)(cid:69) (cid:12)(cid:12)(cid:12) H (cid:68) α γp pγp(cid:105) where φ(pnp) (cid:12)(cid:12)φ(pnp)(cid:11) = (cid:80) note by(cid:12)(cid:12)ψ(pnp)(cid:11) =(cid:80) Consider a state with index np in probe p with en- ergy Epnp that is stationary before the probe is con- Let us denote this state by nected to the system. φ(pnp) is an expan- γp sion coefficient. A scattered wave forms from this state after the probe is attached to the system, which we de- p(cid:48)γp(cid:48) p(cid:48)γp(cid:48)(cid:105), where ψ(pnp) is an expansion coefficient for orbitals in the system, and ψ(pnp) is an expansion coefficient for orbitals p(cid:48)γp(cid:48) in probe p(cid:48). The scattered wave is related to the initial state by the Lippmann-Schwinger equation, giving β(cid:105) +(cid:80) p(cid:48)γp(cid:48) ψ(pnp) β ψ(pnp) γp β β (cid:88) β(cid:48)γp ψ(pnp) β = GR ββ(cid:48)(Epnp )Wβ(cid:48),pγp (Epnp )φ(pnp) γp (1) where GR is the retarded Green's function matrix for the whole system, including all probes, and Wβ(cid:48),pγp (E) = Hβ,pγp − ESβ,pγp is an effective coupling matrix el- If we ement between the system and probe p. (cid:80) now define the retarded self energy Σ(p)R ββ(cid:48) (E) = is the retarded Green's function matrix for isolated probe p,β(cid:48)(E), where G(p)R γp,γ(cid:48) Wβ,pγp (E)G(p)R γp,γ(cid:48) (E)Wpγ(cid:48) γpγ(cid:48) p p p (cid:33) Σ(p)R ββ(cid:48) (E) p, we get the following central results δββ(cid:48)(cid:48) = ρββ(cid:48) = (cid:88) (cid:32) (E + iη) Sββ(cid:48) − Hββ(cid:48) −(cid:88) (cid:88) (cid:88) β(cid:48)β(cid:48)(cid:48)(E) β(cid:48)(cid:48)β(cid:48)(cid:48)(cid:48) β(cid:48)(cid:48)β(cid:48)(cid:48)(cid:48)(E) − Σ(p)R Σ(p)A β(cid:48)(cid:48)β(cid:48)(cid:48)(cid:48)(E) f (p)(E) (cid:111) ββ(cid:48)(cid:48)(E) GR GA p p β(cid:48) ×GR 1 2πi ×(cid:110) β(cid:48)(cid:48)(cid:48)β(cid:48)(E) dE (3) (2) β(cid:48) β(cid:48)(cid:48)(cid:48)β(cid:48) and Σ(p)A where η is a positive infinitesimal, GA β(cid:48)(cid:48)β(cid:48)(cid:48)(cid:48) are the advanced Green's function and self energy respec- tively, ρββ(cid:48) is the single particle electronic density ma- trix, and f (p)(E) is the occupancy of the levels inside the isolated probe p, and hence the energy distribution with which electrons are injected into the system by probe p. We now introduce the Hairy Probe anzatz for the re- tarded self energy. We note that we want the simplest possible form that still possesses the properties required by a self energy. Making it (almost) energy independent allows us to reduce the problem of finding the scattering states to a simple diagonalization, and by making it lo- cal to one orbital we minimize the parameters we have to set. We then end up with the following form 2 iΓpδββp E ≥ Ep,c E < Ep,c (cid:40)− 1 ββ(cid:48) (E) = δββ(cid:48) Σ(p)R (4) 0 where βp is the index of the orbital in the system to which the probe p is attached, and Ep,c is the bottom of the band for the electronic states in probe p, taken to be well below any energy levels in the system. As the self energies are imaginary, they have the effect of allowing electrons to be added to, and removed from, the system [15]. The quantities Γp set the broadening of the states in the system, and define the rates at which electrons can enter or leave. Provided Ep,c lies below all levels in the system, then we can substitute Eq. 4 into Eqs. 2 and 3 to give (cid:34) (cid:88) (cid:32) (cid:88) β(cid:48) ×Gβ(cid:48)β(cid:48)(cid:48)(E) ∞ 1 2π Γp Ep,c p δββ(cid:48)(cid:48) = ρββ(cid:48) = (E + iη) Sββ(cid:48) − Hββ(cid:48) − δββ(cid:48) i 2 (cid:35)(cid:33) (cid:88) Γpδββp p (5) f (p)(E)GR ββp (E)GA βpβ(cid:48)(E) dE (6) Note that Eq. 6 offers an alternative, albeit unphysi- cal, interpretation of Ep,c: it is the energy of the lowest occupied state in probe p, with states with E < Ep,c be- ing unoccupied. A discussion of the implications of the choice of Epc is presented in the Appendix C. We write the retarded Green's function as (cid:88) r β ζ (r)∗ χ(r) β(cid:48) E + iη −
(r) GR ββ(cid:48)(E) = 3 where ζ (r) the corresponding complex eigenvalue. These satisfy β(cid:48) are left and right eigenstates, and
(r) β and χ(r) (cid:34) (cid:88) Hββ(cid:48) − δββ(cid:48)(cid:88) i 2 Γpδββp p (cid:35) β(cid:48) =
(r)(cid:88) (cid:88) χ(r) δrs = ββ(cid:48) Sββ(cid:48)χ(r) β(cid:48)(8) β(cid:48) ζ (r)∗ β Sββ(cid:48)χ(s) β(cid:48)(9) β = ζ (r)∗ β In principle setting χ(r) should satisfy Eq. 9 as the Hamiltonian matrix is symmetric, but we have found that better results are found by solving Eq. 9 explicitly, especially in the presence of degeneracies. To solve these equations, the numerical procedure we have adopted is as follows. We first transform Eq. 8 from a generalised eigenvalue problem to an ordinary one in the usual way. First we carry out a Cholesky decom- position of the overlap matrix and use the resulting tri- angular matrices to express the Hamiltonian (including the self-energies) in an orthogonal representation. We then diagonalise the Hamiltonian matrix using a general complex eigensolver as the problem is complex and sym- metric, rather than Hermitian, and obtain the complex eigenvalues and right eigenvectors. The left eigenvectors are then obtained by inverting the square matrix of right eigenvectors, and then all eigenvectors are transformed back to the original representation using the triangular matrices from the Cholesky decomposition. Substituting Eq. 7 into Eq. 6 gives β χ(s)∗ β(cid:48) frsχ(r) ρββ(cid:48) = (cid:88) (10) where frs = 1 2π (cid:88) Γpζ (r)∗ βp ζ (s) βp p rs ∞ Ep,c (cid:0)E −
(r)(cid:1)(cid:0)E −
(s)∗(cid:1) dE f (p)(E) (11) and can be thought of as a generalized occupancy. We note that in the limit of very weakly coupled leads (Γp → 0) all having the same coupling strength, the occupancy simplifies to (cid:80) (cid:80) p f (p)((cid:60)
(r))ζ (r) βp p ζ (r) βp 2 frs → δrs 2 (12) This limiting form of the occupancy matrix is real and di- agonal, so the system carries no current, and is a weighted sum of the contributions from each probe. We include it in the spirit of moving open boundaries to problems outside the usual range, as it might be relevant to the case of an electrode in an electrochemical cell. Finally we note that if the populations f (p)(
) are independent of the probes, then we get back the usual equilibrium expression for the density matrix. orbitals β and β(cid:48) from We compute the current through the bond between (7) Iββ(cid:48) = − 4e  (Hββ(cid:48)Imρββ(cid:48) − Sββ(cid:48)ImEββ(cid:48)) (13) 4 = (cid:80) ´ ∞ βp Ep,c and ζ (s) βp grs and β χ(s)∗ rs grsχ(r) β(cid:48) Ef (p)(E) (E−
(r))(E−
(s)∗) (cid:80) where Eββ(cid:48) = p Γpζ (r)∗ a 1 2π factor of 2 for spin degeneracy has been included. A derivation of this expression is given in the Appendix B. We use finite temperature occupations for the elec- trons in the probes. To enable analytic and efficient evaluation of the integrals involving the occupancies, we use the following piecewise linear approximation to the Fermi-Dirac distribution function: dE, 1 0 f (p)(E) = 1 2 − 1 4 (cid:16) E−µp kB Tp (cid:17) −2kBTp < E − µp < +2kBTp E − µp ≤ −2kBTp E − µp ≥ +2kBTp (14) where µp and Tp are the chemical potential and tempera- ture for the electrons in probe p. The integrals are given in the Appendix A. Finally we note that the transmission between two probes p1 and p2 is given by (cid:12)(cid:12)(cid:12)GR (cid:12)(cid:12)(cid:12)2 T12(E) = βp2 βp1 (E) Γp1 Γp2. (15) III. RESULTS A. Atomic wires The Hairy Probe algorithm has been implemented in the tight binding program Plato [20]. To test the method we first investigated an atomic wire made from 300 Cu atoms; probes were attached to the first 100 and last 100 atoms. We used the orthogonal TB parameterization of Sutton et al. [21] that assigns just one s orbital to each atom. That is, there are 200 probes in all, one per orbital on each of the 200 lead atoms. The probes all have the same coupling strength Γp, and the same temperature kBTp = 0.001 Ry. Open boundary calculations are car- ried out in two stages. First, every probe is assigned the same chemical potential, and its value is adjusted until the system as a whole is charge neutral; this we term the reference chemical potential. Each atom individually is allowed to acquire a net charge, described by a monopole with a gaussian charge distribution [22], and charge self- consistency is imposed. Second, a bias is applied with the chemical potential on the left probes being raised by half the bias relative to the reference chemical potential, and the chemical potential on the right probes being lowered by half the bias. This allows the wire to acquire a net charge, though this is typically less than 1 electron for the whole system for biases up to 3.9 V. The first step is necessary because the probes do not correspond to a known physical system, so an anzatz is needed to give them sensible characteristics. We computed the current as a function of applied volt- age for a range of coupling strengths of the probes; the results are shown in Fig. 2 a). We see that the current Figure 2. (Color online) a) The current through a wire com- posed of 300 Cu atoms as a function of the applied bias for a range of probe coupling strengths (indicated by the sym- bol G). b) The current through a chain of 300 H atoms as a function of the bias voltage for a range of probe coupling strengths (indicated by the symbol G). Note that in both pan- els the curves for coupling strengths of 0.1 Ry and 0.03 Ry lie on top of each other. varies close to linearly with bias for all probe coupling strengths, and that the slope (conductance) is nearly in- dependent of that coupling for values in the range 0.01 Ry to 0.10 Ry, and in this range the slope is equal to the quantum unit of conductance (G0 = 7.748× 10−5 S). The current is reduced for both larger and smaller cou- plings. With small couplings the current is restricted by the rate at which charge can be injected and removed by the probes. At very large couplings the hopping matrix elements between atoms in the wire become a weak per- turbation on the interaction between the atoms and the probes; in this limit incoming electrons are reflected back into the probes before they can contribute to the current in the wire. The lower and upper bounds for reasonable couplings are roughly the mean spacing between levels (to ensure we have a continuous density of states) and the bandwidth (to ensure the probes do not overwhelm the system). a)b)5 The variation of potential with position can be under- stood in the following way. The potential in the probe free wire is uniform as it is metallic and the electrons can move to screen out any charge accumulation; current in a perfect conductor requires no field, locally [23]. That leaves the regions with probes. Consider electrons arriv- ing at the left region with probes from the middle region, with energies within the conduction window. In this re- gion, the lifetime of electrons before being absorbed into a probe is τ = /Γ and λ = vgroupτ is the mean free path, with vgroup being the group velocity of the elec- trons at the Fermi energy. For a cosine band with band filling ξ we have vgroup = 2av sin(πξ)/, where a is the interatomic spacing, and v is the hopping integral be- tween neighbouring sites. The fraction of electrons that make it to position x (measured from the junction be- tween the perfect wire and the region with probes) dies out as exp(−x/λ). To keep the metal neutral, the band- bottom has to adopt the same shape, to compensate. We thus have the following form for the potential at position x (cid:16) (cid:16)− x (cid:17) − 1 (cid:17) λ φ(x) ∼ 1 2 eV exp (16) Figure 3. (Color online) a) A contour plot of the density of states projected onto each atom. The position corresponds to the index of each atom in the wire. The green regions correspond to low density of states, while the blue and red regions correspond to a high density of states. b) The average potential as a function of position. In Fig. 3 a) is shown the density of states (DOS) pro- jected onto each atom (the atom index is on the x axis) as a function of the electron energy (y axis) for a wire with a bias of 1 V applied, and a coupling of 0.01 Ry for each probe. We see that in the middle of the wire (atom position 150) we have a DOS that is sharply peaked at the band edges. This is consistent with the cosine band structure associated with an infinite chain of atoms with one s orbital per atom. At the ends of the wire there is considerable weight towards the middle of the energy range, consistent with the square root type DOS associ- ated with the end atom of a semi-infinite chain of s or- bitals. We note that the states in the lead regions (atoms 1 to 100, and 201 to 300) are significantly broadened by the probes. Finally, the potential in the wire region is essentially independent of position (Fig. 4 b)). This is to be contrasted with the interface regions where the probes end and begin; here there is a clear variation of potential with position suggesting that this is where the potential drop occurs. where V is the applied voltage. The functional form of φ(x) clearly has a shape corre- sponding to that seen in Fig. 3 b). From Eq. 16 we get φ(x)/φ(∞) = 1 − exp(−x/λ). If we let x1/2 be the point where φ(x1/2)/φ(∞) = 1 2 then we get λ = x1/2/ ln 2. From Fig. 3 b) we see that x1/2 ≈ 20a and hence λ ≈ 29a. As the hopping integral is v = 0.212 Ry, the band fill- ing is 0.243 [21], and Γp = 0.01 Ry, we would expect λ/a = 2v sin(πξ)/Γp ≈ 29; this is in full agreement with the measured value. We have repeated the above calculations using a non- orthogonal DFTB model for hydrogen [24]: an atomic wire made from 300 H atoms with probes attached to the first 100 and last 100 atoms. The resulting current against bias plot is shown in Fig. 2 b). We see that it has the same structure as for the orthogonal Cu wire (see Fig. 2 a)), and that the maximum conductance is again one quantum unit. This suggests that the method for including overlap into the formalism is correct. We note that, for the case of orthogonal tight bind- ing, agreement with the two terminal Landauer solu- tion was demonstrated previously for a non-uniform wire, provided a sufficiently large number of probes was employed[7]. B. Graphene contacts Having studied simple one dimensional atomic wires, and found good agreement with the expected conduc- tance, we now consider electron transport through a more complex system: a benzene ring attached to two graphene contacts by means of covalent bonds. We have a)b)6 Figure 4. (Color online) a) The arrangement for the Hairy Probe calculation of current through a nanoribbon. The probes are attached to the atoms within the blue boxes. b) The current through the nanoribbon as a function of the bias voltage for a range of probe coupling strengths (indicated by the symbol G). Figure 5. (Color online) a) The arrangement of the atoms for the calculation of a current through a benzene molecule attached covalently to two graphene flakes. The probes are attached to the edges of the two graphene flakes. b) The cur- rent through the benzene molecule as a function of the bias applied between the two graphene contacts. c) The trans- mission through the contacts and benzene molecule. The red arrow indicates the location of the chemical potential of the probes at zero bias. selected this system because graphene's electrical prop- erties [9] suggest it might make a good contact material for molecular electronics [10]. As we shall see below, care will have to be taken with how connection to the contacts is made. We note that this system has some similarities to the well studied benzene-dithiol between two gold con- tacts [25]. To estimate the correct coupling strength of the probes to the graphene contacts we first perform calculations of current through a carbon nanoribbon. To compute the current through a small carbon nanoribbon, whose edges have been terminated with hydrogen atoms (see Fig. 4), we again use a non-orthogonal DFTB model [24]. The probes all have the same coupling strength Γp, and same temperature kBTp = 0.001 Ry. The variation of cur- rent with bias is shown in Fig. 4 for a range of cou- pling strengths. For coupling strengths of 0.1 Ry and below we find that the current increases roughly linearly with coupling strength for a given bias, and is sensitive to details of the electronic structure of the nanoribbon. The current is fairly insensitive to coupling strength for 0.3 Ry ≤ Γp ≤ 1 Ry. At large coupling strengths the current is again heavily suppressed. From this we con- clude that for carbon flakes of this size, setting Γp = 0.4 Ry is appropriate. At this coupling, the conductance is 1.44 × 10−4 S which is 1.9 times the quantum of conduc- tance; this can be understood as resulting from two bands crossing the Fermi energy forming two conductance chan- nels. a)b)b)c)µOur final simulation is now of the current through a benzene ring coupled covalently to a pair of graphene con- tacts. The contacts are modelled as small flakes, whose edges are terminated with hydrogen (see Fig. 5 a)). The probes are then attached to the atoms around the edges of each flake, with the probes on one flake all having the same electron chemical potential. The difference be- tween the potentials of the two flakes then creates the bias across the benzene molecule. We use the probe cou- pling strength of Γp = 0.4 Ry found from our nanoribbon calculations. Comparing the current versus voltage plot from Fig. 5 b) with that from Fig. 4 b), the first thing to notice is that the current has dropped by a factor of over 1000. This can be understood by looking at the transmission function for the the benzene molecule (Fig 5 c)). Here we see that the reference chemical potential sits well within a tunelling gap several eV wide, thus there are very few free carriers. As the bias increases a small number of holes appear in the valence band; the benzene molecule acquires a small positive charge of order 0.03e, which grows between 1.5V and 4V to about 0.04e. The presence of the band gap is a consequence of the covalent bond between the benzene ring and the graphene: at the point of contact, the carbon atom in the graphene adopts sp3 hybridization, disrupting the π-conjugation. Thus, to form a good contact, a method is required that maintains the conjugation. Finally, we note that the transport is dominated by holes rather than electrons because the ref- erence chemical potential lies about 0.46 eV closer to the valence band than to the conduction band. C. Graded probes Above we have applied the simplest implementation of the Hairy Probes battery, where all probes have the same coupling strength to their respective atoms. This implementation has the conceptual advantage of corre- sponding most closely to the physical interpretation of the Hairy Probes as external particle baths, in which the system is immersed. In Ref. [7] it was shown that when the length of the hairy leads increases, and Γ decreases (while always remaining larger than the lead energy-level spacing), the Hairy Probes steady state tends to the con- ventional 2-terminal Landauer steady state. However in practice one would like to keep the leads as short as possible for computational reasons. The rough rule of thumb for the optimal Γ then is that it should be as small as possible, while remaining larger than the level spacing in the leads. The resultant steady states then approximate the conventional 2-terminal limit, but not exactly. This is not right or wrong as the Hairy Probe battery is intended to be a stand-alone transport setup, with its own interpretation (as above). But the need to consider finite-size effects, and the precise choice of Γ, could then be seen as irksome. To overcome this complication, a simple alternative is to make Γ position-dependent, so that its value rises 7 Figure 6. The transmission as a function of energy for a per- fect wire of length four atoms to which are connected two leads of length 10 atoms, each lead atom having one probe with a fixed coupling (G) attached. gradually from zero, as we move along each lead, away from the central region. We refer to this scenario as Graded Probes. Below we compare these two implemen- tations numerically, and then comment. The comparison uses the simplest case of a perfect lin- ear atomic chain, with 10-atom long leads with probes and a 4-atom central region without probes. For sim- plicity we use a single-orbital orthogonal model, with a nearest-neighbour hopping integral set to −1, defining the energy unit. The corresponding energy band then lies in the energy interval (−2, 2), and the 2-terminal Landauer solution has unit transmission throughout that interval. First we consider the earlier implementation of the Hairy Probes, with a position-independent coupling Γ in each 10-atom long lead. The surface plot in Fig. 6 shows the transmission as a function of energy and Γ. Consider the limit of small Γ first. In that limit, the 10 + 4 + 10 atom system thinks of itself as a 24-atom linear molecule weakly coupled to an envirnment, which just broadens its 24 molecular states into 24 narrow res- onances. This is the origin of the 24 sharp transmission peaks at the small-Γ end of the plot. To understand the opposite limit - large Γ - consider first each 10-atom lead coupled to its probes, but not yet to the central piece (corresponding to the isolated leads in the usual Green's- function partitioned approach). If Γ is big enough, it dominates all other energy scales in the lead, ultimately making the lead itself a wide-band system, with a den- sity of states (DOS) going down as 1/Γ. If we now couple the components together, then the 4-atom central region just sees low-DOS adjoining leads, with a correspond- ingly small embedding self-energy. The upshot is that now the 4-atom central region behaves as a resonant sys- tem with weakly broadened states, producing the 4 res- onances at the large-Γ end of the plot. 8 broadening of the system states by the probes to produce a continuous density of states. There is still more work to be done to understand completely the properties of the probes. In addition to studying the dependence of cur- rent on applied bias, there are a number of calculations that could be performed, such as the transmission as a function of electron energy for different couplings, or the self-consistent charge distribution. The method is sufficiently simple that it can be imple- mented by finding the eigenvalues and eigenvectors of an effective energy independent Hamiltonian, and then per- forming all the subsequent integrals over energy analyti- cally to produce the single particle density matrix. This results in an efficient algorithm that makes self-consistent open boundary simulations easy to carry out, as it elimi- nates the need to construct lead self-energies and to per- form numerical integrals over energy. The most time con- suming part of the calculations is the construction of the density matrices (Eq. 10). For sparse Hamiltonian and overlap matrices, the scaling for building the density ma- trix is O(N 3), which is no worse than the diagonalization step. The absence of numerical integrals also helps keep the prefactor low. The method was applied to the problem of current flow through a benzene ring attached by covalent bonds to two graphene contacts. It was found that the formation of the contact covalent bonds disrupts the π-conjugation, and thus heavily suppresses the current. We thus conclude that the contacts must either involve physisorption, or a different way to form covalent bonds must be found. We have also introduced a possible way to acceler- ate the convergence of the current with respect to lead length by using graded coupling strengths for the probes. The results shown here look very promising, though more work is needed to fully undeerstand them. ACKNOWLEDGMENTS Figure 7. The transmission as a function of energy for a per- fect wire of length four atoms to which are connected two leads of length 10 atoms each having probes attached. The probes now have graded coupling strengths: see main text for the details. In between these two extremes, there is an optimal region of Γ-values, as expected, producing a roughly uni- form transmission close to 1, but for the given short leads the corrugation always remains visible. The reason is that even at its optimal value, the finite Γ results in an effective interface (between the regions with and without probes), which - like any interface - generates additional scattering. The longer the leads - and the smaller the optimal Γ - the weaker the disruption. The Graded Probes provide an alternative way to sup- press this boundary scattering, without having to make the leads long. The plot in Fig. 7 shows the Graded Probes transmission, with Γ rising linearly from zero to 1.4 along each 10-atom long lead. It is clear that - at no extra computational expense - we are now much closer to the ideal 2-terminal limit, even for the given modest lead length. The Graded Probes thus provide an alterna- tive, if one wishes to avoid very long leads, or having to consider the precise choice of Γ in the uniform-Γ setup. IV. CONCLUSIONS The primary purpose of this paper is to show how to extend the Hairy Probe open boundary method for the steady state to non-orthogonal atomic orbital basis sets. By considering the well understood case of the one di- mensional atomic wire (using both orthogonal and non- orthogonal basis sets) we find that we obtain the ex- pected conductance provided the coupling of the wire to the probes has a suitable value. Couplings that are either too large or too small suppress the current: small couplings reduce the rate of charge injection, while large couplings result in high levels of reflection of electrons back into the probes. The optimal value results in the We gratefully acknowledge funding from the Lever- hulme Trust (RPG-2014-125). M.B. was supported by funding from EOARD (FA8655-12-1-2105) and through a studentship in the Centre for Doctoral Training on The- ory and Simulation of Materials at Imperial College Lon- don, funded by EPSRC under Grant No. EP/G036888/1. A.T. was funded by A-star. We also gratefully acknowl- edge support from the Thomas Young Centre under grant TYC-10. R.D'A. acknowledges support by NAN- OTherm (CSD2010-00044) of the Spanish Ministerio de Economia y Competitividad, and the Grupo Consolidado UPV/EHU del Gobierno Vasco (Grant No. IT578-13). Finally we thank Matas Petreikis for providing improved data for Fig. 4b. −2−101200.20.40.60.81EnergyTransmission 2−terminal LandauerGraded ProbesAppendix A: Analytic integrals over energy For electrons at finite temperature, in principle we should use the Fermi-Dirac distribution, f (p)(E) = (cid:2)1 + e(E−µp)/kB T(cid:3)−1. However, it is not then possible to evaluate the integrals analytically. We thus use the piecewise linear approximation from Eq. 14 with Tp = T for all p: 1 4 0 kB T 2 − 1 (cid:16) E−µp E − µp ≤ −2kBT 1 (cid:17) −2kBT < E − µp < +2kBT (cid:18) µp + 2kBT −
(cid:19) (cid:18) µp − 2kBT −
ln (µp + 2kBT −
) E − µp ≥ +2kBT (cid:19) (A1) 4kBT ln (µp − 2kBT −
) − −1 − ln (Epc −
) 4kBT (A2) f (p)(E) = If we define Jp(
) = then, using the piecewise linear approximation, the inte- grals become ∞ Ep,c (cid:16) f (p)(E) (cid:0)E −
(r)(cid:1)(cid:0)E −
(s)∗(cid:1) dE
(r)(cid:17) − Jp (cid:16)
(s)∗(cid:17)(cid:111) (cid:0)E −
(r)(cid:1)(cid:0)E −
(s)∗(cid:1) dE (cid:16) (cid:16)
(r)(cid:17) −
(s)∗Jp (cid:110)Jp (cid:110)
(r)Jp Ef (p)(E) 1 1
(r) −
(s)∗ ∞ Ep,c
(r) −
(s)∗ I (0) rs = = I (1) rs = = (A3)
(s)∗(cid:17)(cid:111) (A4) Appendix B: Formula for electric currents To evaluate the electric current that flows across a plane, we divide our system into two parts (A and B), each defined by the list of atoms within it. If we use an atomic orbital type basis set, this is equivalent to defin- ing the regions by the set of orbitals associated with the atoms. We label orbitals in A by α and those in B by β. The index for all orbitals (spanning A and B) shall be ν. Let the number of electrons in A be NA, which can be computed from the expression NA = 2Tr , where ρ is the single particle electron density matrix and PA is a partition function for region A. We require PA to be symmetric and to satisfy 1 = PA + PB, where PB is the corresponding partition function for region B. Note that these partition functions are not projectors in general as A (cid:54)= PA). The current they need not be idempotent ( P 2 IA is the time rate of change of the number of electrons (cid:110) in A, namely (cid:105)(cid:111) (cid:110) (cid:111) ρ PA (cid:104) PA, H ρ IA = −e dNA dt = − 2e i Tr (B1) where Kp(
) = I (0) rs = I (1) rs = 9 density matrix is defined by ρ = (cid:80) define PA = (cid:80) where we have used the quantum Liouville equation, the fact that operators permute under a trace, and H is the Hamiltonian. Note that the matrix of coefficients of the νν(cid:48) ν(cid:105) ρνν(cid:48) (cid:104)ν(cid:48), and the inverse overlap matrix we call T (= S−1). We now νν(cid:48) ν(cid:105) PA,νν(cid:48) (cid:104)ν(cid:48) and let the matrix of (cid:19) coefficients PA have the form (cid:18) TAA 1 2 TAB (B2) PA = 1 2 TBA 0 where Sαα(cid:48) = (SAA)αα(cid:48) etc. The current is then found to be IA = − 4e  (HβαImρβα − SβαImEβα) (B3) (cid:88) αβ sum over bond currents, IA = (cid:80) where the matrix E satisfies Hρ = SE, and we have made use of the fact that H and S are symmetric, while ρ and E are Hermitian. We can interpret Eq. B3 as a αβ Iαβ, where Iαβ = − 4e (HβαImρβα − SβαImEβα). Appendix C: Choice of Epc In the Hairy Probe formalism we assume the self- energies are independent of energy; this allows us to use the simple spectral representation of the Green's func- tion. Having results depend on the value of the lower cutoff in the integrals is not consistent with this assump- tion. Here we investigate the internal consistency of the theory. Let us rewrite Eq. A2 as a sum of a term that is independent of Epc (Kp(
)) and a term that depends on Epc Substituting Eq. C1 into Eqs. A3 and A3 then gives (C1) ln (µp − 2kBT −
) (C2) ln (µp + 2kBT −
) − 1 − Kp 4kBT 4kBT Jp(
) = Kp(
) − ln (Epc −
)
(r) −
(s)∗ 1 (cid:18) µp + 2kBT −
(cid:19) (cid:18) µp − 2kBT −
(cid:19) (cid:0)
(s)∗(cid:1) (cid:0)
(r)(cid:1) − Kp (cid:18) (cid:18) sign(Epc) −
(r) Epc sign(Epc) −
(s)∗ Epc (cid:18) − lnEpc (cid:18) sign(Epc) −
(r) Epc sign(Epc) −
(s)∗ Epc (cid:0)
(r)(cid:1) −
(s)∗Kp
(r) −
(s)∗ ln +
(r)Kp (cid:0)
(s)∗(cid:1)
(r) −
(s)∗ ln
(r) −
(s)∗ ln
(r) −
(s)∗ ln
(r) −
(s)∗ (cid:19) (cid:19) (cid:19) (cid:19)
(s)∗
(r) − − + 1 (C3) 2πEpc 1 2πEpc r p Γp (cid:88) (cid:88) lnEpc(cid:88) lnEpc(cid:88) S−1 ββp p p p (C4) = ∆Eββ(cid:48)(Epc) → − 1 2π = − 1 2π ΓpS−1 βpβ(cid:48) (cid:88) Γp s ζ (r)∗ βp χ(r) β r S−1 ββp ΓpS−1 βpβ(cid:48) (C7) χ(s)∗ β(cid:48) ζ (s) βp (cid:88) s (C8) In the limit that Epc (cid:29)
, and for Epc < 0, we have (cid:18) ln sign(Epc) −
Epc = ln (cid:19) (cid:12)(cid:12)(cid:12)(cid:12)sign(Epc) −
(cid:18) Epc (cid:12)(cid:12)(cid:12)(cid:12) (cid:19) +i arg sign(Epc) −
Epc → −iπsign((cid:61)
) Substituting Eq. C4 into Eq. C3, and noting that sign((cid:61)
(r)) = −1, we get (cid:0)
(s)∗(cid:1) − i2π (cid:0)
(r)(cid:1) − Kp (cid:0)
(r)(cid:1) −
(s)∗Kp
(r) −
(s)∗
(r) −
(s)∗ rs → Kp I (0) rs →
(r)Kp I (1) − lnEpc (cid:0)
(s)∗(cid:1) − iπ(cid:0)
(r) +
(s)∗(cid:1) (C5) Thus for large enough Epc, I (0) rs becomes independent rs varies with Epc as − lnEpc, which is of Epc, while I (1) independent of r and s. We now recall the expressions for the density matrices (cid:88) (cid:88) p Γp Γp (cid:88) (cid:88) rs p rs ρββ(cid:48) = Eββ(cid:48) = 1 2π 1 2π rs ζ (r)∗ I (0) βp χ(r) β ζ (s) βp χ(s)∗ β(cid:48) rs ζ (r)∗ I (1) βp χ(r) β ζ (s) βp χ(s)∗ β(cid:48) (C6) 10 Let us define ∆ρββ(cid:48)(Epc) and ∆Eββ(cid:48)(Epc) to be those parts of the density matrices that depend on Epc. Com- bining Eq. C5 with Eq. C6 we get ∆ρββ(cid:48)(Epc) → 1 (cid:88) (cid:88) ζ (r)∗ χ(s)∗ β(cid:48) χ(r) β ζ (s) βp βp The contribution from ∆ρββ(cid:48) becomes arbitrarily small for large enough Epc, while the contribution from ∆Eββ(cid:48) is logarithmically divergent. We note that the Epc depen- dent parts of both matrices are real and symmetric, thus they make no contribution to the electric current, but make a contribution to the atomic forces (not discussed further in this paper). These contributions decrease as the distance from the probes increases, and can be sup- pressed entirely if we set to zero those overlap matrix el- ements that link orbitals not attached to probes to those that are attached to probes. In the main text we offer an alternative interpretation of Epc that allows us to avoid these difficulties, but at the expense of being unphysical: it can be interpreted as the lowest energy for which states in the probes are populated. (2004). (2015). 035116 (2010). [1] R. Martin, Electronic Structure: Basic Theory and Prac- tical Methods (Cambridge University Press, 2004). [2] R. M. Metzger, Chemical Reviews 115, 5056 (2015), pMID: 25950274, http://dx.doi.org/10.1021/cr500459d. [3] M. Brandbyge, J.-L. Mozos, P. Ordejón, J. Taylor, and K. Stokbro, Phys. Rev. B 65, 165401 (2002). [4] A. R. Rocha and S. Sanvito, Phys. Rev. B 70, 094406 [5] T. Ozaki, K. Nishio, and H. Kino, Phys. Rev. B 81, [6] QuantumWise, "Atomistix toolkit, quantumwise a/s," [7] E. J. McEniry, D. Bowler, D. Dundas, A. P. Horsfield, C. G. Sánchez, and T. N. Todorov, Journal of Physics: Condensed Matter 19, 196201 (2007). [8] T. Frauenheim, G. Seifert, M. Elsterner, Z. Hajnal, G. Jungnickel, D. Porezag, S. Suhai, and R. Scholz, phys- ica status solidi (b) 217, 41 (2000). [9] M. J. Allen, V. C. Tung, and R. B. Kaner, Chemical Reviews 110, 132 (2010), pMID: 19610631, http://dx.doi.org/10.1021/cr900070d. [10] G. Wang, Y. Kim, M. Choe, T.-W. Kim, and T. Lee, Advanced Materials 23, 755 (2011). [11] E. G. Emberly and G. Kirczenow, Phys. Rev. B 58, 10911 (1998). Chemical Physics 101, 6849 (1994). [13] T. N. Todorov, G. A. D. Briggs, and A. P. Sutton, Jour- nal of Physics: Condensed Matter 5, 2389 (1993). [14] J. Wang and H. Guo, Phys. Rev. B 79, 045119 (2009). [15] F. Goyer, M. Ernzerhof, and M. Zhuang, The Journal of Chemical Physics 126, 144104 (2007), http://dx.doi.org/10.1063/1.2715932. [16] P. Rocheleau and M. Ernzerhof, The of Chemical Physics 137, nal http://dx.doi.org/10.1063/1.4764291. 174112 Jour- (2012), [17] B. T. Pickup, P. W. Fowler, M. Borg, and I. Sciriha, The Journal of Chemical Physics 143, 194105 (2015), http://dx.doi.org/10.1063/1.4935716. [18] C. J. O. Verzijl, J. S. Seldenthuis, and J. M. Thijssen, The Journal of Chemical Physics 138, 094102 (2013), http://dx.doi.org/10.1063/1.4793259. [19] A. Horsfield, physica status solidi (b) 249, 231 (2012). [20] S. Kenny and A. Horsfield, Computer Physics Commu- nications 180, 2616 (2009), 40 {YEARS} {OF} CPC: A celebratory issue focused on quality software for high performance, grid and novel computing architectures. [21] A. P. Sutton, T. N. Todorov, M. J. Cawkwell, and J. Hoekstra, Philosophical Magazine A 81, 1833 (2001), http://dx.doi.org/10.1080/01418610108216639. [22] P. Soin, A. Horsfield, and D. Nguyen-Manh, Computer [12] V. Mujica, M. Kemp, and M. A. Ratner, The Journal of Physics Communications 182, 1350 (2011). [23] R. Landauer, Journal of Physics: Condensed Matter 1, 8099 (1989). [24] D. Porezag, T. Frauenheim, T. Köhler, G. Seifert, and R. Kaschner, Phys. Rev. B 51, 12947 (1995). [25] P. Delaney and J. C. Greer, Phys. Rev. Lett. 93, 036805 (2004). 11 

1905.10905

1

1905

Pressure-controlled interlayer magnetism in atomically thin CrI3

Stacking order can significantly influence the physical properties of two-dimensional (2D) van der Waals materials. The recent isolation of atomically thin magnetic materials opens the door for control and design of magnetism via stacking order. Here we apply hydrostatic pressure up to 2 GPa to modify the stacking order in a prototype van der Waals magnetic insulator CrI3. We observe an irreversible interlayer antiferromagnetic (AF) to ferromagnetic (FM) transition in atomically thin CrI3 by magnetic circular dichroism and electron tunneling measurements. The effect is accompanied by a monoclinic to a rhombohedral stacking order change characterized by polarized Raman spectroscopy. Before the structural change, the interlayer AF coupling energy can be tuned up by nearly 100% by pressure. Our experiment reveals interlayer FM coupling, which is the established ground state in bulk CrI3, but never observed in native exfoliated thin films. The observed correlation between the magnetic ground state and the stacking order is in good agreement with first principles calculations and suggests a route towards nanoscale magnetic textures by moir\'e engineering.

cond-mat.mes-hall

cond-mat

Pressure-controlled interlayer magnetism in atomically thin CrI3 Tingxin Li1, Shengwei Jiang2, Nikhil Sivadas1, Zefang Wang1, Yang Xu1, Daniel Weber3, Joshua E. Goldberger3, Kenji Watanabe4, Takashi Taniguchi4, Craig J. Fennie1, Kin Fai Mak1,2,5*, and Jie Shan1,2,5* 1 School of Applied and Engineering Physics, Cornell University, Ithaca, NY 14853, USA 2 Laboratory of Atomic and Solid State Physics, Cornell University, Ithaca, NY 14853, USA 3 Department of Chemistry and Biochemistry, The Ohio State University, Columbus, OH 43210, USA 4 National Institute for Materials Science, 1-1 Namiki, Tsukuba, 305-0044, Japan 5 Kavli Institute at Cornell for Nanoscale Science, Ithaca, NY 14853, USA *Correspondence to: km627@cornell.edu, jie.shan@cornell.edu These authors contributed equally: Tingxin Li, Shengwei Jiang. Stacking order can significantly influence the physical properties of two-dimensional (2D) van der Waals materials1. The recent isolation of atomically thin magnetic materials2-22 opens the door for control and design of magnetism via stacking order. Here we apply hydrostatic pressure up to 2 GPa to modify the stacking order in a prototype van der Waals magnetic insulator CrI3. We observe an irreversible interlayer antiferromagnetic (AF) to ferromagnetic (FM) transition in atomically thin CrI3 by magnetic circular dichroism and electron tunneling measurements. The effect is accompanied by a monoclinic to a rhombohedral stacking order change characterized by polarized Raman spectroscopy. Before the structural change, the interlayer AF coupling energy can be tuned up by nearly 100% by pressure. Our experiment reveals interlayer FM coupling, which is the established ground state in bulk CrI3, but never observed in native exfoliated thin films. The observed correlation between the magnetic ground state and the stacking order is in good agreement with first principles calculations23-27 and suggests a route towards nanoscale magnetic textures by moiré engineering28. Intrinsic magnetism in 2D van der Waals materials has received growing attention2-22. Of particular interest is the thickness-dependent magnetic ground state in atomically thin CrI3. In these exfoliated thin films, the magnetic moments are aligned (in the out-of-plane direction) in each layer, but anti-aligned in adjacent layers3,12-22. They are FM (or AF) depending on whether there is (or isn't) an uncompensated layer. The relatively weak interlayer coupling compared to the intralayer coupling allows effective ways to control the interlayer magnetism, which have led to interesting spintronics applications including voltage switching 12-14, spin filtering16-20 and spin transistors21. The origin of interlayer AF coupling is, however, not well understood since interlayer FM order is the ground state in the bulk crystals. Recent ab initio calculations23-27 and experiments22,29,30 have suggested that stacking order could provide an explanation but a direct correlation between stacking order and interlayer magnetism is lacking. In bulk CrI3, the Cr atoms in each layer form a honeycomb structure, and each Cr atom is surrounded by six I atoms in an octahedral coordination (Fig. 1a). The bulk crystals undergo a structural phase transition from a monoclinic phase (space group C2/m) at room temperature to a 1 rhombohedral phase (space group R3) at around 210-220 K (Ref.31,32) with a small volume reduction. The major difference between the two phases is the stacking order while the in-plane structure remains nearly unchanged. In the monoclinic phase, each layer is displaced by a translation vector of (1/3, 0), while in the rhombohedral phase, by (1/3, -1/3) in the ABC order (Fig. 1a). The structural change can be characterized by polarized Raman spectroscopy (see Methods for details). Figure 1b is the spectra recorded at 300 K and 90 K in the back-scattering crossed-polarization configuration. The peak near 107 cm-1 is a two-fold degenerate Eg mode for the rhombohedral phase, whose energy and intensity are independent of the polarization angle (right panel, Fig. 1c). In contrast, for the monoclinic phase with lower symmetry, the mode splits into an Ag and Bg mode with distinct selection rules29,32, which give rise to one unresolved peak with a four-fold polarization dependence for the peak energy (left panel, Fig. 1c). In this study, we apply hydrostatic pressure on exfoliated CrI3 thin films of 2 -- 9 layers to investigate the pressure effect on the material's magnetic properties. Because of the strong structural anisotropy, hydrostatic pressure on layered materials modifies practically only the interlayer structure such as the layer separation33,34 and stacking order. For all measurements we used samples in the form of tunnel junctions with atomically thin CrI3 as a barrier and few-layer graphene as electrodes (Fig. 2a). The devices are encapsulated with hexagonal boron nitride (hBN) to protect CrI3 from the environmental effect. A piston high-pressure cell compatible with electrical transport measurements was used to apply pressure up to 2 GPa. The pressure effect on the magnetic properties of CrI3 was monitored by the tunnel magnetoresistance (TMR). Magnetic circular dichroism (MCD) microscopy and polarized Raman spectroscopy were also employed to characterize the magnetic state and the crystal structure of CrI3, respectively, before and after applying pressure. The MCD data were taken at about 3.5 K, and the tunnel conductance data, at about 1.7 K unless otherwise specified. Since pressure can be varied only at room temperature, the samples typically went through many thermal cycles (under zero magnetic field) during the measurements. (See Methods for details on the bulk crystal growth and device fabrication, as well as the pressure cell and the MCD measurements.) Figure 2b is an optical micrograph of an exfoliated CrI3 film before being integrated into a tunnel junction. The film consists mostly of 2-layer and 5-layer regions. We applied pressure of 1.8 GPa on the sample at room temperature, cooled it down to 1.7 K with pressure, and released pressure after the sample was warmed back up to room temperature. Figure 2c shows the MCD image of the sample at 633 nm under zero field before (left) and after (right) applying pressure. Accordingly, figure 2d is the magnetic-field dependence of MCD at selected locations. The MCD probes the sample magnetization, but its magnitude in samples of different thicknesses cannot be compared directly due to the variations in their spectral response and local field factor. Before applying pressure, the MCD signal under zero field is nearly vanishing across the entire 2-layer region, and is finite (but generally weak) in the 5-layer region. This is consistent with the interlayer-AF order: 2-layer CrI3 is an antiferromagnet, and 5-layer, a ferromagnet due to an uncompensated layer. This is further seen in Fig. 2d (left panel). In addition to the FM loop centered at zero field, both 5-layer regions exhibit two spin-flip transitions: one near 0.75 T and the other near 1.7 T. These transitions correspond to spin flip at the surface layers and the interior layers of the sample, respectively. Above the second transition field, the magnetic moments in all layers are aligned. Since interior layers are absent in bilayers, there is only one 2 spin-flip transition near 0.75 T. At these transitions (first-order phase transitions), hysteresis is clearly visible. These observations are in good agreement with the reported results3,12-17. The MCD image also shows the presence of inhomogeneities. In particular, the nonvanishing signal in even-layer-number regions and the significantly higher signal in certain 5-layer regions are likely from stacking faults which can introduce interlayer FM coupling as we discuss below. Results become dramatically different after applying pressure (right panel, Fig. 2c, d). A significantly higher MCD signal at zero field emerges almost everywhere, suggesting potentially an interlayer AF-to-FM phase transition in atomically thin CrI3. This is supported by the magnetic-field dependence of MCD in both 5-layer regions and the second 2-layer region (denoted by a 'red' dot in Fig. 2b): a clear hysteresis loop centered at zero field is observed. The FM phase persists to about 60 K in both 5-layer and 2-layer regions (Fig. 2e), which agrees well with the Curie temperature of the bulk crystals. The MCD result also shows that the pressure- induced AF-to-FM phase transition is not complete in many locations across the sample. For instance, a mixed FM and AF response with a visible spin-flip transition is observed over a probe area of about 1 µm2 in the first 2-layer region ('black' dot in Fig. 2b). Next we correlate the pressure-induced interlayer AF-to-FM transition with the crystal structure by performing polarized Raman measurements similar to that on the bulk crystals in Fig. 1b, c. Because the Raman efficiency decreases rapidly with sample thickness, we focus only on the 5- layer region ('green' in Fig. 2b). Figure 3 shows the polarization angle dependence of the Raman mode near 107 cm-1 in the crossed-polarization configuration (See Supplementary Sect. 3 for more Raman data). Before applying pressure, the Raman response shows a clear four-fold pattern with a maximum peak shift of about 1.5 cm-1 at 300 K (Fig. 3a). This is consistent with the monoclinic structure observed in the bulk crystals at high temperature. However, in contrary to the bulk case, the four-fold pattern remains on cooling to 10 K (Fig. 3c for 90 K), and the rhombohedral phase is not observed. This observation is consistent with a recent Raman study of CrCl3 of 17 nm in thickness29 and a second harmonic generation study of few-layer CrI3 (Ref. 30). Atomically thin crystals exfoliated at room temperature are likely kinetically trapped in the monoclinic phase at low temperature by the encapsulation layers (graphene or hBN). This is possible since the energy difference between the two stacking polytypes is small (Ref. 23-27). After applying pressure (Fig. 3b, d), the Raman response changes to that for the rhombohedral structure: no polarization dependence can be observed for the peak energy within an uncertainty of 0.2 - 0.3 cm-1. The response at 90 K and 300 K does not change, again supporting that the stacking order is locked in encapsulated atomically thin films. Our result therefore shows that there is a pressure-induced monoclinic-to-rhombohedral stacking order change in exfoliated thin CrI3 films, and interlayer AF (FM) coupling is preferred in the monoclinic (rhombohedral) phase. This correlation between the interlayer magnetic ground state and the stacking order agrees with the recent ab initio calculations23-27. We have investigated more than ten samples, several of which consist of regions of different thicknesses. The effect of pressure and cooling on the magnetic state of these samples is summarized in Supplementary Table S1. We find that in general minimum pressure of 1.7 GPa with cooling is required to achieve a relatively thorough phase transition in bilayer CrI3 and the requirements become less stringent for thicker films. Such a result can be understood from a simple energy consideration. From the bulk data, we obtain that at low temperature the 3 rhombohedral phase has a lower energy than the monoclinic phase, 𝐹𝑅3 − 𝐹𝑐2/𝑚 < 0. But a structural phase transition in atomically thin samples on cooling is prevented by an energy barrier imposed by the capping layers and/or substrates. The application of pressure P could facilitate such a structural phase transition by further increasing the energy difference, 𝑃(𝑉𝑅3 − 𝑉𝑐2/𝑚) < 0, since the rhombohedral structure has a smaller volume V. The transition is not reversible. Finally, we perform tunneling measurements on 2D CrI3 junctions to study the pressure effect on magnetism in-situ. Figure 4a shows the magnetic-field dependence of the tunnel conductance G of the middle 2-layer flake (Fig. 2b) as a function of pressure. (See Supplementary Fig. S1 for the optical image of the device.) Before applying pressure, G shows a jump at the spin-flip transition field 𝐵𝑠𝑓 ≈ 0.75 T. This is the spin-filtering effect that has been recently reported in few-layer CrI3 (ref 16-20). The electron tunneling rate is higher when spins are aligned with the magnetization of each CrI3 layer. The spin-filtering efficiency can be characterized by TMR (normalized conductance difference above and below 𝐵𝑠𝑓 by conductance below 𝐵𝑠𝑓). Again, hysteresis is clearly visible. The monotonic decrease of G with increasing field is a response of the graphene electrodes. At 1 GPa, we observe three major changes: the spin-filtering efficiency decreases, 𝐵𝑠𝑓 shifts to a higher value, and the overall G increases. With a further increase of pressure to 1.8 GPa, the spin-filtering effect nearly disappears. After removing pressure, the behavior of the junction does not revert back to that before applying pressure. A similar trend is observed in a 4-layer tunnel junction under 0 -- 1.4 GPa in Fig. 4b (See Supplementary Sections 4 for results from additional devices). There is a second 𝐵𝑠𝑓 that corresponds to spin flip in the interior layers of CrI3. Both transitions behave similarly as a function of pressure. After removing the pressure, in this case the spin-filtering effect is still clear: the efficiency is reduced, but the 𝐵𝑠𝑓's revert back to the values before applying pressure. We summarize the pressure dependence of 𝐵𝑠𝑓 in the form of relative enhancement over the zero pressure value in Fig. 4c. It increases monotonically with pressure for samples of different thicknesses. The rate is roughly the same for all samples except that for the bilayers, which is about twice as large. The spin-filtering efficiency is determined by fraction of the AF area in the junction. The significantly suppressed spin-filtering effect under high pressure (1.8 GPa) in the bilayer sample is consistent with a nearly complete structural phase transition. The residual effect under intermediate pressure (0.9-1.4 GPa) in the 4-layer sample can be attributed to a spatially mixed AF-FM phase in CrI3. On the other hand, spin flip occurs when the Zeeman splitting energy becomes comparable to the interlayer AF exchange energy 𝐽⊥~𝑔𝜇𝐵𝑚𝑆𝐵𝑠𝑓, where 𝑔 ≈ 2 is the 3 electron g-factor, 𝜇𝐵 is the Bohr magneton, and 𝑚𝑆 ≈ is the spin magnetic quantum number of 2 Cr3+ cations31. In bilayer CrI3, 𝐽⊥ is about 0.25 meV with 𝐵𝑠𝑓 ≈ 0.75 T under zero pressure, and it nearly doubles under 1.5 GPa. Our first-principles calculations show that for bilayer CrI3 in space group C2/m, 𝐽⊥ remains AF, and its magnitude increases nearly linearly with decreasing layer separation by a few percent (Supplementary Sect. 6). Result in Fig. 4c is thus consistent with the picture of layer compression without a stacking order change on application of pressure. The reduction in layer separation also contributes to the observed increase of the overall conductance. But future theoretical studies are needed for a more quantitative comparison between experiment and theory and to understand the unusually large pressure effect on the interlayer exchange interaction in bilayer CrI3. 4  In conclusion, we have experimentally demonstrated control of the interlayer magnetic order through the crystal stacking order in atomically thin van der Waals magnets by application of hydrostatic pressure. Pressure can also be employed to strengthen the interlayer exchange interaction. Our findings not only shed light on the physical origin of the thickness-dependent magnetic ground state in atomically thin CrI3, but the correlation between stacking and magnetic ordering also paves the path for engineering moiré magnetism in double-layer heterostructures of van der Waals magnets23,28. References 1. Castro Neto, A. H., Guinea, F, Pere, N. M. R., Novoselov, K. S. & Geim, A. K. The electronic properties of graphene. Rev. Mod. Phys. 81, 109-162 (2009). 2. Gong, C. et al. Discovery of intrinsic ferromagnetism in two-dimensional van der Waals crystals. Nature 546, 265-269 (2017). 3. Huang, B. et al. Layer-dependent ferromagnetism in a van der Waals crystal down to the monolayer limit. Nature 546, 270-273 (2017). 4. Bonilla, M. et al. Strong room-temperature ferromagnetism in VSe2 monolayer on van der Waals substrates. Nature Nano. 13, 289-293 (2018). 5. Fei, Z. et al. Two-dimensional itinerant ferromagnetism in atomically thin Fe3GeTe2. Nature Mat. 17, 778-782 (2018). 6. Deng, Y. et al. Gate-tunable room-temperature ferromagnetism in two-dimensional Fe3GeTe2. Nature 563, 94-99 (2018). 7. Burch, K. S., Mandrus, D. & Park, J. G. Magnetism in two-dimensional van der Waals materials. Nature 563, 47-52 (2018). 8. Gong, C. & Zhang, X. Two-dimensional magnetic crystals and emergent heterostructure devices. Science 363, 706 (2019). 9. Gibertini, M., Koperski, M., Morpurgo, A. F. & Novoselov, K. S. Magnetic 2D materials and heterostructures. Nature Nano. 14, 408-419 (2019). 10. Cortie, D. L et al. Two-Dimensional Magnets: Forgotten History and Recent Progress towards Spintronic Applications. Adv. Funct. Mater. 1901414 (2019). 11. Li, H., Ruan, S. & Zeng, Y. J. Intrinsic Van Der Waals Magnetic Materials from Bulk to the 2D Limit: New Frontiers of Spintronics. Adv. Mater. 1900065 (2019). 12. Jiang, S., Shan J. & Mak, K. F. Electric-field switching of two-dimensional van der Waals magnets. Nature Mat. 17, 406-410 (2018). 13. Huang, B. et al. Electrical control of 2D magnetism in bilayer CrI3. Nature Nano. 13, 544-548 (2018). 14. Jiang, S., Li, L., Wang, Z., Mak, K. F. & Shan J. Controlling magnetism in 2D CrI3 by electrostatic doping. Nature Nano. 13, 549-553 (2018). 15. Wang, Z. et al. Electric-field control of magnetism in a few-layered van der Waals ferromagnetic semiconductor. Nature Nano. 13, 554-559 (2018). 16. Song, T. et al. Giant tunneling magnetoresistance in spin-filter van der Waals heterostructures. Science 360, 1214-1218 (2018). 17. Klein, D. R. et al. Probing magnetism in 2D van der Waals crystalline insulators via electron tunneling. Science 360, 1218-1222 (2018). 18. Kim, H. H. et al. One Million Percent Tunnel Magnetoresistance in a Magnetic can der Waals Heterostructure. Nanolett. 18, 4885-4890 (2018). 5 19. Song, T. et al. Voltage Control of a van der Waals Spin-Filter Magnetic Tunnel Junction. Nanolett. 19, 915-920 (2019). 20. Wang, Z. et al. Very large tunneling magnetoresistance in layered magnetic semiconductor CrI3. Nature Comm. 9, 2516 (2018). 21. Jiang, S., Li, L., Wang, Z., Mak, K. F. & Shan J. Spin transistor built on 2D van der Waals heterostructures. Nature Electronics 2, 159-163 (2019). 22. Thiel, L. et al. Probing magnetism in 2D materials at the nanoscale with single-spin microscopy. Science 10.1126/science.aav6926 (2019). 23. Sivadas, N., Okamoto, S., Xu, X., Fennie, C. J. & Xiao D. Stacking-Dependent Magnetism in Bilayer CrI3. Nanolett. 18, 7658-7664 (2018). 24. Jiang, P. et al. Stacking tunable interlayer magnetism in bilayer CrI3. Phys. Rev. B 99, 144401 (2019). 25. Soriano, D., Cardoso, C. & Fernández-Rossier, J. Interplay between interlayer exchange and stacking in CrI3 bilayers. arXiv 1807.00357. 26. Jang, S. W., Jeong, M. Y., Yoon, H., Ryee, S. & Han, M. J. Microscopic understanding of magnetic interactions in bilayer CrI3. Phys. Rev. Materials 3, 031001 (2019). 27. Lei, C. et al. Magnetoelectric Response of Antiferromagnetic Van der Waals Bilayers. arXiv 1902.06418. 28. Tong, Q., Liu, F., Xiao, J. & Yao, W. Skyrmions in the Moiré of van der Waals 2D Magnets. Nanolett. 18, 7194-7199 (2018). 29. Klein, D. R. et al. Giant enhancement of interlayer exchange in an ultrathin 2D magnet. arXiv 1903.00002. 30. Sun, Z. et al. Giant and nonreciprocal second harmonic generation from layered antiferromagnetism in bilayer CrI3. arXiv 1904.03577. 31. McGuire, M. A., Dixit, H., Cooper, V. R. & Sales, B. C. Coupling of Crystal Structure and Magnetism in the Layered Ferromagnetic Insulator CrI3. Chem. Mater. 27, 612-620 (2015). 32. Djurdjić-Mijin, S. et al. Lattice dynamics and phase transition in CrI3 single crystals. Phys. Rev. B 98, 104307 (2018). 33. Yankowitz, M. et al. Dynamic band-structure tuning of graphene moiré superlattices with pressure. Nature 557, 404-408 (2018). 34. Yankowitz, M. et al. Tuning superconductivity in twisted bilayer graphene. Science 363, 1059-1064 (2019). Competing interests The authors declare no competing interests Data availability The data supporting the plots within this paper and other findings of this study are available from the corresponding authors upon request. 6 Methods Growth of CrI3 bulk crystals. Bulk CrI3 crystals were synthesized by chemical vapor transport, as described previously35. The elements were sealed in an evacuated ampoule (109.0 mg Cr chunks, 1 eq., Alfa Aesar, 99.999 % purity; 798.0 mg I2 crystals, 1.5 eq, Alfa Aesar, 99.999 % purity). The ampoule was heated to 650 °C at the feed and 550 °C at the growth zone. After a week, the ampoule was cooled and the reaction yielded crystals with 1-2 mm edge length. CrI3 produced by this method crystallized in the C2/m space group with typical lattice parameters of a = 6.904 Å, b = 11.899 Å, c = 7.008 Å and β = 108.74 ° at room temperature. In a typical sample, the crystals exhibit a Curie temperature of 61 K as well as a structural phase transition to the space group R3 at 210-220 K31. Device fabrication. CrI3 tunnel junctions were fabricated by the layer-by-layer dry-transfer method36. Atomically thin flakes of hexagonal boron nitride (hBN), graphite, and CrI3 were mechanically exfoliated from bulk crystals onto silicon substrates covered by a 300-nm thermal oxide layer. The flakes were picked up in sequence by a stamp made of a thin film of polycarbonate (PC) on polydimethylsiloxane (PDMS). The entire stack was then released onto a substrate with prepatterned titanium/gold (Ti/Au 5 nm/35 nm) electrodes. The residual PC on the device surface was dissolved in chloroform before measurements. Both the exfoliation and the transfer processes were performed in a nitrogen-filled glovebox to avoid degradation of CrI3. The thickness of atomically thin CrI3 samples was initially estimated from their optical reflectance contrast, and later verified by MCD or atomic force microscopy measurements. High-pressure experiments. A piston high-pressure cell compatible with electrical transport at low temperature (easyCell 30, Almax easyLab) was employed. The cylindrical cell has an inner bore diameter of about 3.9 mm. Hydrostatic pressure is applied on the samples through pressure transmission medium, pentane and iso-pentane mixture (1:1), which freezes around 160 K. The typical range of pressure that can be accessed at low temperature is 1 - 2 GPa. Pressure was applied (or changed) at room temperature. The cell was then cooled to low temperature for measurements. Two types of manometers were used for pressure calibration. At room temperature, pressure was calibrated by measuring the resistance of a manganin wire (~ 10 cm long) wound into a coil37. At low temperature, pressure was calibrated by measuring the superconducting transition temperature of a tin wire37. The pressure is generally about 0.1 - 0.3 GPa lower at low temperature than at room temperature due to the freezing of the pressure transmission medium. MCD microscopy. Magnetic circular dichroism (MCD) measurements were performed in an attoDry1000 cryostat with a base temperature of 3.5 K. For non-imaging measurements, a HeNe laser at 633 nm was used. The laser beam was coupled into and out of the cryostat using free- space optics. It was focused onto the sample by a cold objective. The beam size on the sample was about 1 µm2, and the power, no more than 10 µW. The incident beam was modulated between left and right circular polarization by a photoelastic modulator at 50.1 kHz. The reflected beam was collected by the same objective and detected by a photodiode. The MCD signal was determined as a ratio of the ac component at 50.1 kHz (measured by the lock-in amplifier) and the dc component (measured by a digital multimeter) of the reflected light intensity. 7  In MCD imaging, a broad-field illumination and a nitrogen-cooled charge-coupled device (CCD) were used. The light source was selected from an incoherent white light using a band pass filter centered at 632 nm. A linear polarizer and a quarter wave plate were used to generate circularly polarized light. The MCD image was calculated as the normalized difference between the left and right circularly polarized light reflection by the total reflection. The spatial resolution is diffraction limited (~ 500 nm). Polarized Raman spectroscopy. Raman spectroscopy was performed using a home-built microscope in the back-scattering geometry. A solid-state laser at 532 nm was employed as the excitation source. The laser beam was focused onto the CrI3 samples along the c-axis by a 40X objective. The scattered light was collected by the same objective and detected by a spectrometer with a 1800-grooves/mm diffraction grating and a nitrogen-cooled CCD. The polarized Raman modality was used to characterize the crystal structure. To this end, the laser beam passed through a linear polarizer and a half-wave plate. The latter varies the polarization angle in the crystal a-b plane. The scattered light passed through the same half-wave plate, and a second linear polarizer, which selects the component either parallel or perpendicular to the incident beam polarization (referred to as the parallel- and crossed-polarization configuration, respectively). The polarized Raman spectroscopy in crystals of both the R3 and C2/m groups has been analyzed previously29. In particular, for the doubly degenerate 1Eg and 2Eg modes (in the rhombohedral phase) and the non-degenerate Ag and Bg modes (in the monoclinic phase) that have been examined in this study, the Raman tensors are derived as 1 = ( 𝐸𝑔 𝑑 𝑐 𝑒 𝑑 −𝑐 𝑓 0 𝑒 𝑓 ), 𝐸𝑔 2 = ( 𝑑 −𝑐 −𝑓 −𝑐 −𝑑 𝑒 0 𝑒 −𝑓 ), 𝐴𝑔 = ( 𝑎 0 𝑑 0 𝑐 0 𝑑 0 𝑏 ), 𝐵𝑔 = ( 0 𝑒 0 𝑒 0 𝑓 0 𝑓 0 ). 1 ) 𝑐𝑟𝑜𝑠𝑠 = 𝑐 sin(2𝜃) − 𝑑 cos(2𝜃)2 and 𝐼( 𝐸𝑔 In the crossed-polarization configuration, the Raman intensity for the two Eg modes is given as 𝐼( 𝐸𝑔 is the polarization angle with respect to the a-axis. The total intensity of the two degenerate modes (𝑐2 + 𝑑2) is polarization independent. Similar result can be derived for the parallel- polarization configuration. = 𝑐 cos(2𝜃) + 𝑑 sin(2𝜃)2, where 𝜃 𝑐𝑟𝑜𝑠𝑠 2 ) The result is different for the non-degenerate Ag and Bg modes in the monoclinic phase. The Raman intensity of each mode shows a four-fold polarization dependence: 𝐼(𝐴𝑔) 𝑎2sin2(2𝜃), 𝐼(𝐵𝑔) since the symmetry of 𝐴𝑔 + 𝐵𝑔 should match with the symmetry of 𝐸𝑔 total intensity of these two spectrally unresolved modes shows a four-fold polarization dependence for the peak energy. Similar result can be derived for the parallel-polarization configuration. = 𝑒2cos2(2𝜃), but they are off by π/4. Here we have assumed 𝑎 ~ − 𝑐, 1 + 𝐸𝑔 (Ref. 29). The 𝑐𝑟𝑜𝑠𝑠 𝑐𝑟𝑜𝑠𝑠 = 2 References 35. Shcherbakov, D. et al. Raman Spectroscopy, Photocatalytic Degradation, and Stabilization of Atomically Thin Chromium Tri-iodide. Nano Lett. 18, 4214-4219 (2018). 8 36. Wang, L. et al. One-dimensional electrical contact to a two-dimensional material. Science 342, 614-617 (2013). 37. Eremets, M. High Pressure Experimental Methods (Oxford University Press, Oxford, 1996). Figure captions Fig. 1 Crystal structure of CrI3. a, Top: atomic structure of monolayer CrI3. Cr atoms (blue balls) form a honeycomb lattice structure. Each Cr atom is surrounded by six I atoms (purple balls) in an octahedral coordination. Middle: top view of stacking order in the monoclinic (left) and rhombohedral phase (right) for three layers. The Cr network shifts in the order of blue, red and black. Bottom: side view of the corresponding stacking order for two layers. a, b, and c are the crystallographic axes. b, Polarized Raman spectrum of a bulk CrI3 crystal at 300 K (upper) and 90 K (lower) in the crossed-polarization configuration. c, Polarization angle dependence of the spectrum inside the dashed box in b. The peak energy of the mode around 107 cm-1 shows a four-fold dependence on the polarization angle in the monoclinic phase (left), and no dependence in the rhombohedral phase (right). Fig. 2 Pressure-induced interlayer AF-to-FM transition in atomically thin CrI3. a, Schematic side view of CrI3 tunnel junctions employed in this study. Atomically thin CrI3 serves as a tunnel barrier, and few-layer graphene (G) (connected to Au contacts) is used as tunnel electrodes. The entire device is encapsulated with hBN. b, Optical micrograph of a thin flake of CrI3 before being integrated into a tunnel junction. The flake has regions of different thicknesses identified by dashed lines of different colors. The scale bar is 5 𝜇m. c, MCD image of the CrI3 flake before (left) and after (right) applying pressure of 1.8 GPa. The data were recorded at 3.5 K under zero magnetic field. d, Magnetic-field dependence of MCD at 3.5 K for two 2-layer and two 5-layer regions before (left) and after (right) applying pressure. The color of the lines matches the color of the dots identified on the image in b. e, MCD vs. field at varying temperature for a 5-layer (left) and a 2-layer region (right) reveals that the pressure-induced ferromagnetism persists to ~ 60 K. Fig. 3 Pressure-induced structural phase transition in atomically thin CrI3. Polarization angle dependence of Raman spectrum in the crossed-polarization configuration for a 5-layer CrI3 sample before (a, c) and after (b, d) applying pressure of 1.8 GPa; at 300 K (a, b) and at 90 K (c, d). The change from a four-fold dependence to a no-dependence for the peak energy of the mode around 107 cm-1 is indicative of a monoclinic-to-rhombohedral phase change. No structural phase change is observed on cooling in the absence of pressure regardless of the initial structural phase. Fig. 4 Spin-filtering effect in atomically thin CrI3 as a function of pressure. a, b, Tunnel conductance G versus applied magnetic field B of a bilayer CrI3 tunnel junction at 1.7 K under varying pressure in the order of 0, 1, 1.8 and 0 GPa. b, Same measurement of a 4-layer CrI3 tunnel junction under varying pressure in the order of 0, 0.9, 1.4, and 0 GPa. The bias voltage for all measurements was 10 mV. c, Enhancement of the spin-flip transition field 𝐵𝑠𝑓(𝑃)−𝐵𝑠𝑓(0) 𝐵𝑠𝑓(0) as a function of pressure P from 6 different devices. The dashed straight lines are a guide to the eye. 9 The spin-flip transition field for each field sweep direction was estimated from the field corresponding to the steepest change of G. Its uncertainty was estimated from the width of the rise/fall of G with field in the transition region. 𝐵𝑠𝑓 in c was calculated as the average for the two sweep directions. 10 a b c a Monoclinic (C2/m) b s t n u o C s t n u o C 1500 1000 500 0 1000 500 0 Rhombohedral ( ) Bulk CrI3, 300 K Cross polarized configuration Crossed-polarization Bulk CrI3, 90 K Cross polarized configuration Crossed-polarization 100 150 Raman shift (cm-1) 200 250 a b c c a b c ) ° ( l e g n a n o i t a z i r a o P l 360 270 180 90 0 Bulk CrI3, cross polarized Counts T ~ 300 K 80 100 120 Raman shift (cm-1) 140 1200 900 600 300 0 ) ° ( l e g n a n o i t a z i r a o P l 360 270 180 90 0 Bulk CrI3, cross polarized Counts 800 T ~ 90 K 400 80 100 120 Raman shift (cm-1) 0 140 Figure 1 G Au hBN CrI3 hBN SiO2/Si G Au Before applying pressure 5 μm B = 0 T, T ~ 3.5 K, 0 GPa b 5 μm MCD 0.06 0.06 60.0 40.0 0.04 0.02 20.0 0.00 After applying pressure 5 μm B = 0 T, T ~ 3.5 K, 0 GPa Before applying pressure After applying pressure a c d D C M 0.03 0.00 -0.03 0.03 0.00 -0.03 0.05 0.00 -0.05 0.05 0.00 -0.05 e 0.05 D C M 0.00 -0.05 Bilayer 2L 2L Bilayer 5L 5-layer 5-layer 5L -2 3.5 K 18.5 K 27 K 36 K 44 K 61 K 69 K 0 B (T) 2 5-layer region After applying pressure 5L 2L Bilayer 2L Bilayer 5L 5-layer 5-layer 5L -2 3.5 K 14 K 27 K 36 K 44 K 53 K 61 K D C M 0.03 0.00 -0.03 0.03 0.00 -0.03 0.05 0.00 -0.05 0.05 0.00 -0.05 D C M 0.04 0.02 0.00 -0.02 -0.04 -2 -1 0 B (T) 1 2 -2 -1 Figure 2 0 B (T) 2 Bilayer region 2L After applying pressure 1 2 0 B (T) a ) ° ( l e g n a n o i t a z i r a o P l c ) ° ( l e g n a n o i t a z i r a o P l 360 270 180 90 0 360 270 180 90 0 Before appyling pressure Native 5-layer, cross polarized Counts T ~ 300 K 80 100 120 Raman shift (cm-1) 140 900 600 300 0 Before applying pressure Native 5-layer, cross polarized Counts 800 T ~ 90 K 400 80 100 120 Raman shift (cm-1) 0 140 b ) ° ( l e g n a n o i t a z i r a o P l d ) ° ( l e g n a n o i t a z i r a o P l 360 270 180 90 0 360 270 180 90 0 Pressurized 5-layer, cross polarized After applying pressure Counts T ~ 300 K 80 100 120 Raman shift (cm-1) 140 900 600 300 0 Pressurized 5-layer, cross polarized After applying pressure T ~ 90 K Counts 600 300 80 100 120 Raman shift (cm-1) 0 140 Figure 3 a 25 0 GPa 15 40 1 GPa ) S μ ( G 20 40 1.8 GPa 20 45 25 -3 0 GPa -2 -1 0 B (T) 1 2 3 b ) S μ ( G 0.05 0.00 0.1 0.0 0.1 0.0 0 GPa 0.9 GPa 1.4 GPa 0.02 0.00 -4 0 GPa -2 c 100 ) % ( f s B f o t n e m e c n a h n E 80 60 40 20 0 Trilayer-1 Trilayer-2 4-layer, (interior layer) 4-layer, (surface layer) 9 layer, (interior layer) 9 layer, (surface layer) Bilayer-1 Bilayer-2 0 B (T) 2 4 0.0 0.5 1.0 1.5 2.0 P (GPa) Figure 4 

1809.08612

1

1809

Transport signatures of relativistic quantum scars in a graphene cavity

We study a relativistic quantum cavity system realized by etching out from a graphene sheet by quantum transport measurements and theoretical calculations. The conductance of the graphene cavity has been measured as a function of the back gate voltage (or the Fermi energy) and the magnetic field applied perpendicular to the graphene sheet, and characteristic conductance contour patterns are observed in the measurements. In particular, two types of high conductance contour lines, i.e., straight and parabolic-like high conductance contour lines, are found in the measurements. The theoretical calculations are performed within the framework of tight-binding approach and Green's function formalism. Similar characteristic high conductance contour features as in the experiments are found in the calculations. The wave functions calculated at points selected along a straight conductance contour line are found to be dominated by a chain of scars of high probability distributions arranged as a necklace following the shape of cavity and the current density distributions calculated at these point are dominated by an overall vortex in the cavity. These characteristics are found to be insensitive to increasing magnetic field. However, the wave function probability distributions and the current density distributions calculated at points selected along a parabolic-like contour line show a clear dependence on increasing magnetic field, and the current density distributions at these points are characterized by the complex formation of several localized vortices in the cavity. Our work brings a new insight into quantum chaos in relativistic particle systems and would greatly stimulate experimental and theoretical efforts towards this still emerging field.

cond-mat.mes-hall

cond-mat

Transport signatures of relativistic quantum scars in a graphene cavity G. Q. Zhang,1, ∗ Xianzhang Chen,2, ∗ Li Lin,3 Hailin Peng,3 Zhongfan Liu,3 Liang Huang,2, † N. Kang,1, ‡ and H. Q. Xu1, 4, § 1Beijing Key Laboratory of Quantum Devices, Key Laboratory for the Physics and Chemistry of Nanodevices and Department of Electronics, Peking University, Beijing 100871, P. R. China 2School of Physical Science and Technology and Key Laboratory for Magnetism and Magnetic Materials of MOE, Lanzhou University, Lanzhou, Gansu 730000, P. R. China 3Center for Nanochemistry, Beijing Science and Engineering Center for Nanocarbons, Beijing National Laboratory for Molecular Sciences, College of Chemistry and Molecular Engineering, Peking University, Beijing 100871, P. R. China 4Division of Solid State Physics, Lund University, Box 118, S-221 00 Lund, Sweden (Dated: September 25, 2018) 8 1 0 2 p e S 3 2 ] l l a h - s e m . t a m - d n o c [ 1 v 2 1 6 8 0 . 9 0 8 1 : v i X r a 1 Abstract Wave function scars refer to localized complex patterns of enhanced wave function probability distributions in a quantum system. Existing experimental studies of wave function scars concen- trate nearly exclusively on nonrelativistic quantum systems. Here we study a relativistic quantum cavity system realized by etching out from a graphene sheet by quantum transport measurements and theoretical calculations. The conductance of the graphene cavity has been measured as a function of the back gate voltage (or the Fermi energy) and the magnetic field applied perpen- dicular to the graphene sheet, and characteristic conductance contour patterns are observed in the measurements. In particular, two types of high conductance contour lines, i.e., straight and parabolic-like high conductance contour lines, are found in the measurements. The theoretical calculations are performed within the framework of tight-binding approach and Green's function formalism. Similar characteristic high conductance contour features as in the experiments are found in the calculations. The wave functions calculated at points selected along a straight conductance contour line are found to be dominated by a chain of scars of high probability distributions ar- ranged as a necklace following the shape of cavity and the current density distributions calculated at these point are dominated by an overall vortex in the cavity. These characteristics are found to be insensitive to increasing magnetic field. However, the wave function probability distributions and the current density distributions calculated at points selected along a parabolic-like contour line show a clear dependence on increasing magnetic field, and the current density distributions at these points are characterized by the complex formation of several localized vortices in the cavity. Our work brings a new insight into quantum chaos in relativistic particle systems and would greatly stimulate experimental and theoretical efforts towards this still emerging field. 2 In mesoscopic chaotic structures, the wave functions of eigenstates can coalesce in partic- ular coordinate space to form scars with enhanced probability distributions1 -- 18. Scars could be related to classical trajectories in semiclassical limit2,3,8 or result from a superposition of several regular eigenstates in a closed structure4. It is feasible to measure such quantum scars in an open quantum system via transport measurements. In spite of that coupling to an environment will wash out the characteristics of quantum states, a few eigenstates of the corresponding closed structure, which can effectively mediate the transport, can be selected and visualized by transport measurements4,8,19,20. Scars have been experimentally observed in different systems, including microwave billiard14,15 and mesoscopic cavities1,9,16,17,21,22. For example, at low magnetic fields, certain characteristic patterns in the conductance of an open mesoscopic system are shown to be related to the underlying energy spectrum of the corre- sponding closed structure22 -- 24. But, so far, experimental studies of mesoscopic cavities are nearly exclusively performed in nonrelativistic systems described by the Schrodinger equa- tion with a quadratic energy dispersion in a corresponding infinite system. An interesting question is whether such scars can generally appear with new characteristics in relativistic quantum systems described by the Dirac equation which, for an infinite system, gives a linear energy dispersion. Theoretical works have predicted25 -- 29 unequivocal evidence of quantum scars in relativistic systems. But, on the experimental side, observation of quantum scars has only been achieved very recently in a mesoscopic graphene ring device using scanning gate technique.30 Graphene is a two-dimensional material consisting of carbon atoms in the honeycomb lattice31. Graphene exhibits rich interesting physics properties32 -- 37, such as lin- ear energy dispersion in the vicinity of the Dirac point, chiral carriers of zero mass, and extremely high mobility. Thus, a graphene cavity is an excellent candidate for the study of quantum scars in a finite relativistic system. In this work, we study the transport properties of an open graphene cavity and demon- strate the observation of transport signatures of quantum scars in the relativistic particle system. The cavity is made of chemical vapor deposition (CVD) grown graphene on a sub- strate of n-doped Si covered by a thin layer of SiO2. The conductance of the graphene cavity is measured as a function of the Fermi energy and magnetic field. Characteristic patterns are found in the measurements and are analyzed in terms of the underlying energy spectra. The system is also studied by theoretical calculations based on Green's function method. The calculated conductance map (i.e., a plot of the conductance as a function of the Fermi en- 3 ergy and the magnetic field) is found to be in good agreement with the experiment. In both the measured and the calculated conductance maps, two distinct types of high conductance contour lines, i.e., straight and parabolic-like lines, are found. From the calculated local density of states (LDOS) and the current density distributions, we find that along a straight high conductance contour line the scar pattern remains almost unchanged with increasing magnetic field. The current in the cavity is found to circulate clockwise at a finite magnetic field, when the contour line has a positive slope, and anti-clockwise, when the contour line has a negative slope. On contrast, along a parabolic-like high conductance contour line, the charge density distribution displays a much more complex scar pattern and the current density distribution exhibits the formation of several local vortices. At zero magnetic field, the total effective areas enclosed by the vortices circulating in opposite directions are the same. However, with increasing magnetic field, this balance is broken and the difference in the total effective area enclosed by the vortices of opposite directions increases, leading to the observation of the parabolic-like high conductance contour lines in the conductance maps. Our graphene cavity device was fabricated on a Si/SiO2 substrate from monolayer graphene grown via CVD. The fabrication was started by transferring CVD grown graphene on a substrate of n-doped Si covered by a 300-nm-thick layer of SiO2 38. After transferring, a standard 16-µm-long and 3-µm-wide Hall-bar structure with a cavity inside was fabricated by electron beam lithography (EBL) and reactive ion etching with oxygen plasma. Contacts were subsequently fabricated by an additional step of EBL and deposition of a bilayer of Ti/Au (10 nm/90 nm) by electron beam evaporation. Figure 1(a) displays a false-color atomic force microscope (AFM) image of the fabricated device and a schematic for the measurement setup. The highlighted dark red region is the graphene current channel. The small green regions are graphene flakes which are isolated by narrow trenches from the cavity structure. A zoom-in look of the cavity structure is shown in Figure 1(b). The graphene cavity structure is of octagonal shape with ∼1 µm in size and is connected to bulk graphene via two 400-nm-wide constrictions. The device also consists of a region without a fine cavity structure. This arrangement enables us to directly compare the transport measurements of the cavity structure with bulk graphene on the same device. The measurements were carried out by applying a constant current I through the two most distant contacts, that is, the source and the drain, and recording voltage drop V1 over the cavity structure and 4 voltage drop V2 over the graphene bulk at the same time. The conductance of the cavity region is obtained as G1 = I/V1 and the conductance of the reference bulk region is ob- tained as G2 = I/V2. The magnetotransport measurements were performed in a 3He/4He dilution refrigerator with magnetic field B applied perpendicular to the graphene plane, using a standard ac lock-in technique (with a current bias of 10-100 nA at a frequency of 13 Hz). Before comparative studies of the cavity with the graphene bulk, the graphene sheet was characterized by standard Hall measurements. Figure 1(c) shows the measured Hall resistance Rxy and longitudinal resistance Rxx in the bulk graphene region at magnetic field B = 5 T at 60 mK. Here, well developed quantized Hall plateaus are observed, demonstrat- ing the high quality of the graphene32. The mobility extracted from the measurements is around 17,000 cm2V−1s−1 at carrier density n ≈ 1.0 × 1011 cm2 and the mean free path le derived from semi-classical relation37 le = (¯h/e) · µ · (πn)1/2 is about 100 nm. Figures 2(a) and 2(b) show the conductance G2 of the bulk graphene region and the conductance G1 of the cavity region measured as a function of the magnetic field at different back gate voltages. Here and after, the longitudinal bulk graphene conductance is denoted by G2. In Figure 2(a), the feature of universal conductance fluctuations (UCFs)39, i.e., aperiodic fluctuations with fluctuation amplitude δg2D ≈ 0.2e2/h are observable. Through the theoretical prediction39 of δg2D ≈ Lϕ(W 1/2/L3/2)δg0, where W and L are the width and the length of the bulk graphene Hall bar, Lϕ is the phase coherence length, and δg0 ∼ e2/h, the phase coherence length in our graphene sample is estimated to be on the order of 1 µm, which is consistent with previous experiments40. In Figure 2(b), instead of showing UCFs, the conductance curves are smoother. In addition, We can observe that conductance peaks (indicated by black arrows) appear at some particular magnetic fields. These characteristics remind us about the conductance enhancements via transport through the eigenstates of the corresponding closed system4. To explore further about the evolution of these conductance peaks we performed the conductance map measurements, i.e., the measurements of the conductance as a function of the back gate voltage and the magnetic field, for both the bulk and cavity regions. Figure 2(c) and 2(d) show the measured conductance maps (in a color scale) for the bulk and cavity regions in a back gate voltage window of VBG =8 V to 12 V and a magnetic field window of −0.2 T to 0.2 T. The red regions in Figure 2(c) and 2(d) represent the regions with high conductance. It is seen in Figure 2(d) that the measured conductance map exhibits a characteristic "monkey face" pattern of enhanced conductance. 5 However, the measured conductance map shown in Figure 2(c) displays a nearly structure- less distribution of the conductance. In early experiments, similar characteristic features as seen in Figure 2(d) were found and were linked to the underlying energy spectra of closed quantum cavities4,22 -- 24,41. Furthermore, because the carriers injected through a quantum point contact will be in a collimated form3,5,22,42 -- 45, carriers can enter the cavity with a sufficiently large probability only in certain angles and thus only some particular eigenstates can be preferentially excited and can contribute to the carrier transport. Figure 3(a) is the conductance map measured in the same magnetic field window of −0.2 to 0.2 T but a back gate voltage window of VBG=0 V to 5.8 V. Red arrows beside the graph indicate the direction towards the Dirac point. Here, as we expected, the overall conductance is found to decrease with increasing back gate voltage, i.e., when the Fermi level moves towards the Dirac point. But, more importantly, the measured conductance map is found to exhibit complex contour patterns. To better understand these complex features, full quantum-mechanical transport calculations were performed for the graphene cavity structure within the Landauer formalism46, which relates the zero-temperature two- terminal conductance G of the device to the transmission coefficient T in the form of G = 2e2 h T . The transmission coefficient T were calculated in the Green's function scheme within the tight-binding framework25,27,47, which we will only briefly describe here (for further details of the calculations, see Supporting Information). The cavity device can be split into three parts: left lead, cavity and right lead. The two leads are set to be semi-infinite to simulate the open boundaries39. The Green's function of the device is given by GD(E) = (EI − HD − ΣL − ΣR)−1, where ΣL and ΣR are the self-energies caused by the left and right leads, and HD is the tight-binding Hamiltonian of the graphene cavity with hopping terms up to the third-nearest-neighbor atoms included. The hopping energies are 2.8 eV, 0.28 eV, and 0.07 eV for the nearest, the second-nearest, and the third-nearest neighbors, respectively48 -- 50. The coupling matrices between the leads and the cavity, ΓL(E) and ΓR(E), are given in terms of self-energies ΓL,R = i(ΣL,R − Σ† T (E) = Tr(ΓLGDΓRG† current flow is given by Ji→j = 4e L,R). The transmission T is given by π Im[diag(GD)]. The local D is the electron D). The LDOS can be obtained by ρ = − 1 h Im[HD,ijC n ji(E)]39, where C n = GDΓLG† correlation function. Figure 3(b) shows the results of the calculations, which clearly succeed in reproducing the main features of the experiment results shown in Figure 3(a). The satisfactory agree- 6 ment between the experiment and the theory inspires us to get further understanding of the characteristic patterns of high conductance contour lines by taking a close comparison between Figure 3(a) and Figure 3(b). Let us focus on the regions marked by two dashed rectangles in Figure 3(a), which we label as regions I and II. We note that region I is closer to the Dirac point than region II. Figure 4(a) is a close-up plot of the measurements in region I of Figure 3(a), while Figure 4(b) is a close-up plot of the calculations in the corresponding region shown in Figure 3(b). Here, more featured high conductance contour lines are observable. Surprisingly, the patterns observed in the measurements and the calculations are still well matched. Both Figure 4(a) and Figure 4(b) show similar straight high conductance contour lines (see, e.g., the lines marked by yellow dashed lines A and B) and parabolic-like high contour lines (see, e.g., the lines marked by green dashed lines C). To get the physical insights into these characteristic high conductance contour lines, we have computed the LDOS and the current density distribution at a few selected points along the lines. In the tight-binding formulation (see Supporting Information for details), the LDOS provides the wave function probability distribution (or charge density distribution) contributed by all the states at energy E and a given magnetic field B, while the current density distribution provides the information about current paths for carriers with energy E to pass through the cavity at magnetic field B. Figures 4(c) and 4(d) show the calculated LDOS and current density distribution at five selected points, denoted by an with n =1, 2, 3, 4, and 5, along the straight high conductance contour line A in Figure 4(b). Red regions in Figure 4(c) correspond to the regions with high charge density probability distributions. Note that here the color scales in different panels are different. The patterns seen in Figure 4(c) are highly reminiscent of scars of enhanced wave function probabilities in coordinate space4,11. For example, at point a1, the wave functions are highly localized to the regions close to the boundary of the cavity, looking as a chain of pearls (scars) arranged in a peanut shell like structure. At point a2 where a finite magnetic field is applied, the wave functions remain localized to the regions close to the boundary of the cavity. The same localization characteristics are also seen in the wave function probability distributions at points a3 to a5, although the scars become slightly smeared. In Figure 4(d), the corresponding current density distributions calculated at the same five selected points along the straight high conductance contour line A are plotted. 7 Here it is seen that at zero magnetic field, i.e., at point a1, the current density distribution is symmetric with respect to the horizontal axis (marked by a dot-dashed red line). On both the upper and the lower side of the axis, we see an overall current flow from the left to the right, although the several sharp current turns inside the cavity are observable. At finite magnetic fields, i.e., at points a2 to a5, the current density distributions are no longer symmetric with respect to the horizontal axis. Here on the upper side the current flows from the left to the right, while on the lower side the current flows from the right to the left, leading to the formation of an overall clockwise current vortex in the cavity. Note that here a net current passing through the cavity still remains to be from the left to the right. Note also that although we find the wave function probability distribution patterns at points a2 to a5 are very similar in Figure 4(c), their corresponding current density distributions shown in Figure 4(d) do exhibit small but noticeable differences. For example, although very similar current density distribution patterns are found at points a2 and a3 or at points a4 and a5, a small difference can be seen when the current density distribution patterns at points a3 and a4 are compared. This difference is most likely caused by the difference in mixing of the scarring states with other states, since the two points lie on the two sides of another high conductance contour line. Similar localization characteristics have been found in the calculated wave function prob- ability distributions and current density distributions at points bm with m =1, 2, 3, and 4 selected along high conductance contour line B. Here, the wave functions are again highly localized to form a chain of scars in the regions close to the boundary of the cavity and the current density distributions are seen to form an overall vortex in the cavity. However, it is interesting to note that the current vortex found at each of these points is to rotate anti-clockwise, in difference from the results obtained at points a2 to a5. This difference is consistent with the fact that high conductance contour line B has a negative slope, which is on contrast to line A (line A has a positive slope). In a semiclassical description, along a high conductance contour lines, the energy of a state in the cavity at a magnetic field is approximately proportional to the magnetic flux penetrating through an effective area S enclosed in the current paths of an effective total length L, i.e., E = E0 ± (vF eS/L)B,29 where the sign depends on the orientation of the local current circulating the magnetic flux (see Supporting Information for details). As we have shown above, with increasing magnetic field along a straight high conductance contour 8 line, the wave function probability distributions and the current density distributions remain roughly the same. Thus, the effective area enclosed in the current vortex is approximately unchanged with increasing magnetic field. As a result, the energy of the states increases linearly with increasing magnetic field as seen in line A in Figures 4(a) and 4(b). The negative slope seen in line B is because here the current vortices are anti-clockwise and thus the magnetic flux penetrating through the effective area enclosed in each of these vortices carries an opposite sign. Figures 5(a) and 5(b) show the calculated LDOS and current density distributions at selected points ck with k =1, 2, 3, and 4 along parabolic-like high conductance contour line C shown in Figure 4(b). Again, the charge density distributions shown in Figures 5(a) are all symmetric with respect to the horizontal axis. But, such symmetry is found in the current density distribution only at zero magnetic field as seen in panel c1 of Figure 5(b). However, when comparing to the results shown in Figure 4(c) to 4(f), significant differences are found. First, high density spots localized in the middle of the cavity and arranged as vertically elongated X patterns are found in the charge density distributions. Thus, no closed orbit-like structures are seen. Second, much more complex structures are seen in the current density distributions. In particular, several small current vortices are present in the current density distributions and are spread over the cavity. Third, at zero magnetic field, clockwise and anti-clockwise orientated current vortices are symmetrically localized in the cavity and the areas enclosed by all clockwise and all anti-clockwise oriented vortices are equal. But, with increasing magnetic field, the area enclosed by all vortices oriented in one direction (say clockwise) grows slowly and the area enclosed by all vortices oriented in the opposite direction (say anti-clockwise) shrinks. This difference in the area enclosed by the vortices of two different directions at a finite magnetic field is in contrast to the results shown in Figures 4(d) and 4(f), i.e., instead of being a constant, the effective circulating area S in this case increases with increasing magnetic field, which could be the origin of the observed parabolic-like magnetic field dependence of the energy as revealed by the high conductance contour lines C shown in Figures 4(a) and 4(b). Figure 6(a) shows a close-up plot of the measured conductance map in region II of Figure 3(a) and Figure 6(b) shows a plot of the calculated conductance map in the corresponding region. By comparison of the results shown in Figures 6(a)and 6(b), similar features can again be found in the measurements and calculations. Straight high conductance contour 9 lines can be recognized and are marked by dashed yellow lines in the figures. Figures 6(c) and 6(d) show the calculated LDOS and the current density distributions at four selected points dl with l =1, 2, 3, and 4 along the yellow dashed line D shown in Figure 6(b). Here, as we have seen in Figures 4(c) to 4(f), the charge density distributions and the current density distributions shown in Figures 6(c) and 6(d) display similar ring-like orbit structures, and exhibit little changes with increasing magnetic field. This result is consistent with the linear dependence of the state energy on the magnetic field as we discussed above. However, comparing to the results calculated for the region I shown in Figure 4, here we can recognize clearly that additional current paths appear along the edges of the cavity. This might manifests the higher conductance observed in this far from Dirac point region. In summary, we have studied the quantum transport properties of a relativistic quantum cavity. The cavity was made from a CVD grown graphene sheet on a Si/SiO2 substrate. The low-temperature measurements of the conductance map, i.e., the conductance in the linear response regime as a function of the back gate voltage and the magnetic field applied perpendicular to the graphene plane, have been carried out for the cavity device. The com- plex characteristic features were found in the measured conductance map. To analyze the underlying physics revealed in these measurements, the graphene cavity device was modeled by a third-nearest-neighbor tight-binding Hamiltonian, and the conductance, charge den- sity distribution and current density distribution were calculated based on Green's function formalism. The calculated conductance map exhibits similar complex characteristics as ob- served in the measurements. The calculated charge density distributions show the formation of scars and the current density distributions display the formation of complex vortices in the cavity. In particular, both straight and parabolic-like high conductance contour lines were found in the calculated and measured conductance maps. It has been found that along a straight high conductance contour line, the scar pattern remains almost unchanged with increasing magnetic field, while the circulating direction of the current in the cavity at a finite magnetic field is closely related to the slope of the contour line -- it circulates clockwise when the contour line has a positive slope but anti-clockwise when the contour line has a negative slope. Here it should be emphasized that the straight high conductance contour line and the associated characteristics found in the scar pattern and the current density dis- tribution are inherit to a relativistic quantum cavity. However, along a parabolic-like high conductance contour line, it has been found that the charge density distribution displays a 10 complex scar pattern and the current density distribution exhibits the formation of several local vortices. Furthermore, although at zero magnetic field the total effective areas enclosed by the vortices circulating in opposite directions are the same, this balance is broken at a finite magnetic field and the difference in the total effective area enclosed by the vortices of opposite directions changes with increasing magnetic field. Such a parabolic-like high conductance contour line has been commonly observed for nonrelativistic quantum system at low Fermi energy. But here we show it can also been observed in a relativistic quantum cavity. We expect that our work would stimulate experimental and theoretical studies of quantum chaos in relativistic quantum systems. Acknowledgments We acknowledge financial supports by the Ministry of Science and Technology of China (MOST) through the National Key Research and Development Program of China (No. 2016YFA0300601 and 2017YFA0303304), the National Natural Science Foundation of China (Nos. 11874071, 11774005, 11775101, 91221202 and 91421303, 11374019), and the Swedish Research Council (VR). ∗ These authors contributed equally to this work. † Corresponding author: huangl@lzu.edu.cn ‡ Corresponding author: nkang@pku.edu.cn § Corresponding author: hqxu@pku.edu.cn 1 Bird, J. P.; Ishibashi, K.; Aoyagi, Y.; Sugano, T.; Akis, R.; Ferry, D. K.; Pivin JR, D. P.; Connolly, K. M.;Taylor, R. P.; Newbury, R.; Olatona, D. M.; Micolich, A.; Wirtz, R.;Ochiai, Y.; Okubo, Y. Chaos, Solitons & Fractals 1997, 8, 1299-1324. 2 Bird, J. P.; Ferry, D. K.; Akis, R.; Ochiai, Y.; Ishibashi, K.; Aoyagi, Y.; Sugano, T. Europhys. Lett. 1996, 35, 529-534. 3 Akis, R.; Ferry, D. K.; Bird, J. P. Phys. Rev. B: Condense Matter Mater. Phys. 1996, 54, 17705. 4 Zozoulenko, I. V.; Berggren, K.-F. Phys. Rev. B: Condense Matter Mater. Phys. 1997, 56, 6931. 11 5 Bird, J. P.; Akis, R.; Ferry, D. K.; Aoyagi, Y.; Sugano, T. J. Phys.: Condens. Matter 1997, 9, 5935. 6 Ferry, D. K.; Akis, R. A.; Pivin Jr., D. P.; Bird, J. P.; Holmberg, N.; Badrieh, F.; Vasileska, D. Physica E 1998, 3, 137-144. 7 Ferry, D. K.; Akis, R.; Bird, J. P. Phys. Rev. Lett. 2004, 93, 026803. 8 Ferry, D. K.; Akis, R.; Bird, J. P. J. Phys.: Condens. Matter 2005, 17, S1017. 9 Akis, R.; Ferry, D. K.; Bird, J. P. Phys. Rev. Lett. 1997, 79, 123. 10 Akis, R.; Bird, J. P.; Ferry, D. K. Appl. Phys. Lett. 2002, 81, 129-131. 11 Heller, E. J. Phys. Rev. Lett. 1984, 53, 1515. 12 Bogomolny, E. B. Physica D 1988, 31, 169-189. 13 Berry, M. V. Proc. R. Soc. Lond. A 1989, 423, 219-231. 14 Sridhar, S. Phys. Rev. Lett. 1991, 67, 785. 15 Stein, J.; Stockmann, H. -J. Phys. Rev. lett. 1992, 68, 2867. 16 Marcus, C. M.; Rimberg, A. J.; Westervelt, R. M.; Hopkins, P. F.; Gossard, A. C. Phys. Rev. Lett. 1992, 69, 506. 17 Fromhold, T. M.; Wilkinson, P. B.; Sheard, F. W.; Eaves, L.; Miao, J.; Edwards, G. Phys. Rev. Lett. 1995, 75, 1142. 18 Xu, H.-Y.; Huang, L.; Lai, Y.-C.; Grebogi, C. Phys. Rev. Lett. 2013, 110, 064102. 19 Zurek, W. H. Phys. Rev. D 1982, 26, 1862. 20 Zurek, W. H. Rev. Mod. Phys. 2003, 75, 715-775. 21 Bird, J. P.; Olatona, D. M.; Newbury, R.; Taylor, R. P.; Ishibashi, K.; Stopa, M.; Aoyagi, Y.; Sugano, T.; Ochiai, Y. Phys. Rev. B: Condense Matter Mater. Phys. 1995, 52 R14336. 22 Akis, R.;Ferry, D. K.; Bird, J. P.; Vasileska, D. Phys. Rev. B: Condense Matter Mater. Phys. 1999, 60, 2680. 23 Persson, M.; Pettersson, J.; Von Sydow, B.;Lindelof, P. E.;Kristensen, A.; Berggren, K. F. Phys. Rev. B: Condense Matter Mater. Phys. 1995, 52, 8921. 24 Zozoulenko, I. V.; Sachrajda, A. S.; Gould, C.; Berggren, K. -F.; Zawadzki, P.; Feng, Y.; Wasilewski, Z. Phys. Rev. Lett. 1999, 83, 1838. 25 Huang, L.; Lai, Y.-C.; Ferry, D. K.; Akis, R.; Goodnick, S. M. J. Phys.: Condens. Matter 2009, 21 344203. 26 Huang, L.; Lai, Y.-C.; Ferry, D. K.; Goodnick, S. M.; Akis, R. Phys. Rev. Lett. 2009, 103 12 054101. 27 Ferry, D. K.; Huang, L.; Yang, R.; Lai, Y.-C.; Akis, R. J. Phys. Conf. Ser. 2010, 220 012015. 28 Huang, L.; Lai, Y.-C.; Grebogi, C. Chaos 2011, 21, 013102. 29 Ying, L.; Huang, L.; Lai, Y.-C.; Grebogi, C. Phys. Rev. B: Condense Matter Mater. Phys. 2012, 85, 245448. 30 Cabosart, D.; Felten, A.; Reckinger, N.; Iordanescu, A.; Toussaint, S.; Faniel, S.; Hackens, B. Nano Lett. 2017, 17, 1344-1349. 31 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. 32 Zhang, Y.; Tan, Y.-W.; Stormer, H. L.; Kim, P. Nature 2005, 438, 201. 33 Yang, L.; Cohen, M. L.; Louie, S. G. Nano Lett. 2007, 7, 3112-3115. 34 McCann, E.; Kechedzhi, K.; Fal'ko, V. I.; Suzuura, H.; Ando, T.; Altshuler, B. L. Phys. Rev. Lett. 2006, 97, 146805. 35 Chen, C.; Rosenblatt, S.; Bolotin, K. I.; Kalb, W.; Kim, P.; Kymissis, I.; Stormer, H. L.; Heinz, T. F.; Hone, J. Nat. Nanotechnol. 2009, 4, 861-867. 36 Castro Neto, A. H.; Guinea, F.; Peres, N. M. R.; Novoselov, K. S.; Geim, A. K. Rev. Mod. Phys. 2009, 81, 109-162. 37 Bolotin, K. I.; Sikes, K. J.; Jiang, Z.; Klima, M.; Fudenberg, G.; Hone, J.; Kim, P.; Stormer, H. L. Solid State Commun 2008, 146, 351-355. 38 Lin, L.; Li, J.; Ren, H.; Koh, A. L.; Kang, N.; Peng, H.; Xu, H. Q.; Liu, Z. ACS Nano 2016, 10, 2922-2929. 39 Datta, S. Electronic Transport in Mesoscopic Systems (Cambridge University Press, Cambridge, England, 1995). 40 Chen, Y.-F.; Bae, M. -H.; Chialvo, C.; Dirks, T.; Bezryadin, A.; Mason, N. J. Phys.: Condens. Matter 2010, 22, 205301. 41 Akis, R.; Ferry, D. K. Semicond. Sci. Technol. 1998, 13, A18-A20. 42 Molenkamp, L. W.; Staring, A. A. M.;Beenakker, C. W. J.; Eppenga, R.; Timmering, C. E.; Williamson, J. G.; Harmans, C. J. P. M.; Foxon, C. T. Phys. Rev. B: Condense Matter Mater. Phys. 1990, 41, 1274. 43 Beenakker C. W. J.; van Houten, H. Phys. Rev. B: Condense Matter Mater. Phys. 1989, 39, 10445. 13 44 Beenakker C. W. J.; van Houten, H. Solid State Physics 1991, 44, 1-228. 45 Zozoulenko, I. V.; Schuster, R.; Berggren, K. -F.; Ensslin, K. Phys. Rev. B: Condense Matter Mater. Phys. 1997, 55, R10209. 46 Landauer, R. Philos. Mag. 1970, 21, 863-867. 47 Li, T. C.; Lu, S. -P. Phys. Rev. B: Condense Matter Mater. Phys. 2008, 77, 085408. 48 Wallace, P. R. Phys. Rev. 1947, 71, 622-634. 49 Reich, S.; Maultzsch, J.; Thomsen, C.; Ordej´on, P. Phys. Rev. B: Condense Matter Mater. Phys. 2002, 66, 035412. 50 Kretinin, A.; Yu, G. L.; Jalil, R.; Cao, Y.; Withers, F.; Mishchenko, A.; Katsnelson, M. I.; Novoselov, K. S.; Geim, A. K.; Guinea, F. Phys. Rev. B: Condense Matter Mater. Phys. 2013, 88, 165427. 14 (a) (b) (c) 2 2 Bulk 6 10 14 10 6 2 0.6 0.4 0.2 0.0 -0.2 -0.4 ) 2 e / h ( y x R ) : k ( x x R 8 6 4 2 0 -20 -10 0 10 VBG-VDrift (V) 20 -0.6 30 FIG. 1. (a) False-colored atomic force microscope (AFM) image of the graphene device and schematics for the device structure and measurement setup. Yellow parts are electrodes. Red region is graphene. Green highlighted regions are graphene pieces isolated from the cavity by etched trenches, which could be used as side gates but are not used in this work. The device is made on a Si/SiO2 substrate which is used as a back gate. V1 and V2 denote the voltage drops over the cavity and a graphene bulk region, respectively, and VH is the Hall voltage generated in the graphene bulk region. (b) Zoom-in AFM image of the cavity structure in (a). (c) Hall resistance Rxy and longitudinal resistance Rxx of the graphene bulk region measured at perpendicularly ap- plied magnetic field B = 5 T and temperature T = 60 mK. VBG is the applied back gate voltage and VDrift is the drifting gate voltage of the Dirac point. 15 FIG. 2. (a) Longitudinal conductance of the graphene bulk region measured as a function of the magnetic field B at different back gate voltages VBG and temperature T = 60 mK. Curves are successively vertically offset by 1e2/h for clarity. (b) Conductance of the graphene cavity measured as a function of B at different back gate voltages VBG and temperature T = 60 mK. Curves are successively vertically offset by 0.15e2/h. Black arrows in the figure denote the positions of conductance peaks which evolute with increasing back gate voltage. (c) Conductance map of the bulk graphene region, i.e., longitudinal conductance measured for the bulk graphene region as a function of VBG and B at T = 60 mK. (d) Conductance map of the graphene cavity at T = 60 mK. 16 (a) (b) : ; FIG. 3. (a) Conductance map of the graphene cavity measured at temperature T = 60 mK over a large range of VBG. (b) Simulated conductance map of the graphene cavity. Here, similar characteristic conductance patterns are found in the measurements and the calculations. The red arrows point to the direction towards the Dirac point. 17 (a) (b) A C B A (cid:135)(cid:222) (cid:135)(cid:221) C (cid:136)(cid:220)(cid:136)(cid:221) B (cid:135)(cid:220) (cid:135)(cid:219) (cid:136)(cid:218)(cid:136)(cid:219) (cid:135)(cid:218) (cid:137)(cid:221) (cid:137)(cid:219) (cid:137)(cid:218) (cid:137)(cid:220) (c) =5 =6 =7 =8 =9 (d) =5 =6 =7 =8 =9 (e) >5 >6 >7 >8 (f) >5 >6 >7 >8 FIG. 4. (a) and (b) Zoom-in plot of the measurements shown in rectangular region I of Figure 3(a) and zoom-in plot of the calculations in the corresponding region shown in Figure 3(b). Straight (parabolic-like) high conductance contour lines are highlighted with yellow (green) dashed lines. (c) and (d) Calculated wave function probability distributions and current density distributions at points an, where n =1, 2, 3, 4, and 5, selected along straight line A. Here, it is seen that the scar pattern does not show a significant change with increasing magnetic field and the current density distribution in each panel shows only one clockwise current vortex. (e) and (f) Calculated wave function probability distributions and current density distributions at points bm, where m =1, 2, 3, and 4, selected along straight contour line B. Similar characteristic features in the wave function probability distributions and the current density distributions as in (c) and (d) are observed, except that the current vortex seen in each current density distribution is anti-clockwise. 18 (a) ?5 ?6 ?7 ?8 (b) ?5 ?6 ?7 ?8 FIG. 5. (a) and (b) Calculated wave function probability distributions and current density dis- tributions at points ck, where k =1, 2, 3, and 4, selected along parabolic-like contour line C in figure 4(b). Here the scar distribution pattern shows sensitive change with change in magnetic field, and the current density distribution exhibits formation of a complex structure consisting of several localized clockwise and anti-clockwise current vortices. 19 (a) (b) D (cid:138)(cid:221) (cid:138)(cid:220) (cid:138)(cid:219) D (cid:138)(cid:218) (c) @5 @6 @7 @8 (d) @5 @6 @7 @8 FIG. 6. (a) and (b) Zoom-in plot of the measurements shown in rectangular region II of Figure 3(a) and zoom-in plot of the calculations in the corresponding region shown in Figure 3(b). Note that region II is far from the Dirac point compared with region I. Yellow dashed lines highlights straight high conductance contour lines observable in the measurements and the calculations. (c) and (d) Calculated wave function probability distributions and current density distributions at points dl, where l =1, 2, 3, and 4, selected along straight line D in (b). Similar characteristic features in the charge density distributions and the current density distributions as in Figure 4 are observed, except that a well-defined additional current path along the edge of the cavity is observable in each current density distribution panel. 20 

1907.00137

1

1907

Terahertz Faraday and Kerr rotation spectroscopy of Bi$_{1-x}$Sb$_x$ films in high magnetic fields up to 30 Tesla

We report results of terahertz Faraday and Kerr rotation spectroscopy measurements on thin films of $\text{Bi}_{1-x}\text{Sb}_{x}$, an alloy system that exhibits a semimetal-to-topological-insulator transition as the Sb composition $x$ increases. By using a single-shot time-domain terahertz spectroscopy setup combined with a table-top pulsed mini-coil magnet, we conducted measurements in magnetic fields up to 30~T, observing distinctly different behaviors between semimetallic ($x < 0.07$) and topological insulator ($x > 0.07$) samples. Faraday and Kerr rotation spectra for the semimetallic films showed a pronounced dip that blue-shifted with the magnetic field, whereas spectra for the topological insulator films were positive and featureless, increasing in amplitude with increasing magnetic field and eventually saturating at high fields ($>$20~T). Ellipticity spectra for the semimetallic films showed resonances, whereas the topological insulator films showed no detectable ellipticity. To explain these observations, we developed a theoretical model based on realistic band parameters and the Kubo formula for calculating the optical conductivity of Landau-quantized charge carriers. Our calculations quantitatively reproduced all experimental features, establishing that the Faraday and Kerr signals in the semimetallic films predominantly arise from bulk hole cyclotron resonances while the signals in the topological insulator films represent combined effects of surface carriers originating from multiple electron and hole pockets. These results demonstrate that the use of high magnetic fields in terahertz magnetopolarimetry, combined with detailed electronic structure and conductivity calculations, allows us to unambiguously identify and quantitatively determine unique contributions from different species of carriers of topological and nontopological nature in Bi$_{1-x}$Sb$_x$.

cond-mat.mes-hall

cond-mat

Terahertz Faraday and Kerr rotation spectroscopy of Bi1−xSbx films in high magnetic fields up to 30 Tesla Xinwei Li,1, ∗ Katsumasa Yoshioka,2, ∗ Ming Xie,3, ∗ G. Timothy Noe II,1 Woojoo Lee,3 Nicolas Marquez Peraca,4 Weilu Gao,1 Toshio Hagiwara,2 Handegard S. Orjan,5 Li-Wei Nien,5 Tadaaki Nagao,5 Masahiro Kitajima,2, 5 Hiroyuki Nojiri,6 Chih-Kang Shih,3 Allan H. MacDonald,3 Ikufumi Katayama,2 Jun Takeda,2 Gregory A. Fiete,3, 7, 8 and Junichiro Kono1, 4, 9 1Department of Electrical and Computer Engineering, Rice University, Houston, Texas 77005, USA 2Department of Physics, Graduate School of Engineering, Yokohama National University, Yokohama 240-8501, Japan 3Department of Physics and Center for Complex Quantum Systems, The University of Texas at Austin, Austin, Texas 78712, USA 4Department of Physics and Astronomy, Rice University, Houston, Texas 77005, USA 5National Institute for Materials Science, Tsukuba, Ibaraki 305-0044, Japan 6Institute for Materials Research, Tohoku University, Sendai 980-8577, Japan 7Department of Physics, Northeastern University, Boston, Massachusetts 02115, USA 8Department of Physics, Massachusetts Institute of Technology, Cambridge, Massachusetts 02139, USA 9Department of Material Science and NanoEngineering, Rice University, Houston, Texas 77005, USA (Dated: July 2, 2019) We report results of terahertz Faraday and Kerr rotation spectroscopy measurements on thin films of Bi1−xSbx, an alloy system that exhibits a semimetal-to-topological-insulator transition as the Sb composition x increases. By using a single-shot time-domain terahertz spectroscopy setup combined with a table-top pulsed mini-coil magnet, we conducted measurements in magnetic fields up to 30 T, observing distinctly different behaviors between semimetallic (x < 0.07) and topological insulator (x > 0.07) samples. Faraday and Kerr rotation spectra for the semimetallic films showed a pronounced dip that blue-shifted with the magnetic field, whereas spectra for the topological insulator films were positive and featureless, increasing in amplitude with increasing magnetic field and eventually saturating at high fields (>20 T). Ellipticity spectra for the semimetallic films showed resonances, whereas the topological insulator films showed no detectable ellipticity. To explain these observations, we developed a theoretical model based on realistic band parameters and the Kubo formula for calculating the optical conductivity of Landau-quantized charge carriers. Our calculations quantitatively reproduced all experimental features, establishing that the Faraday and Kerr signals in the semimetallic films predominantly arise from bulk hole cyclotron resonances while the signals in the topological insulator films represent combined effects of surface carriers originating from multiple electron and hole pockets. These results demonstrate that the use of high magnetic fields in terahertz magnetopolarimetry, combined with detailed electronic structure and conductivity calculations, allows us to unambiguously identify and quantitatively determine unique contributions from different species of carriers of topological and nontopological nature in Bi1−xSbx. I. INTRODUCTION One of the most appealing methods proposed for prob- ing topologically protected surface states in a three- dimensional (3D) topological insulator (TI) is observ- ing the topological magnetoelectric (TME) effect [1 -- 8]. When time-reversal symmetry is broken by a mag- netic field, a gap appears in the surface Dirac state(s), and the Hall conductivity of the sample surface is half- integer quantized, which in turn leads to a quantized bulk magnetoelectric response. This phenomenon can be described through an electromagnetic Lagrangian anal- ogous to the theory of axion electrodynamics in par- ticle physics [9]. Experimentally, the most common and established experimental techniques for studying 3D TIs are angle-resolved photoemission spectroscopy ∗ These authors contributed equally. (ARPES) [10 -- 18] and electronic transport [19 -- 25] mea- surements, both of which are not suited for detecting the TME effect. Specifically, a magnetic field cannot be ap- plied in ARPES experiments, and there are limitations due to nontopological edge states in transport experi- ments. Magneto-optical spectroscopy at low photon ener- gies, such as the microwave and far-infrared spectral ranges, has been recognized to be an ideal probe for the TME effect [3 -- 5]. By using electromagnetic radiation whose photon energy is comparable to or smaller than the magnetic-field-induced surface gap, previous tera- hertz (THz) magneto-optical spectroscopy experiments have detected changes of the surface optical conductiv- ity induced by an applied small or moderate magnetic field [26 -- 32]. Applying a much stronger magnetic field on the sample is, generally speaking, expected to be more advantageous because the field can induce a gap much larger than the photon energy and bring the surface car- rier states toward the quantum limit [33]. 9 1 0 2 n u J 9 2 ] l l a h - s e m . t a m - d n o c [ 1 v 7 3 1 0 0 . 7 0 9 1 : v i X r a In this article, we describe results of THz Faraday and Kerr rotation spectroscopy measurements on thin films of Bi1−xSbx. This alloy system is known to show a semimetal (SM)-to-TI transition as a function of x, and its phase diagram (Fig. 1) is well established [10, 13, 34 -- 36], although some controversy remains as to whether surface states in the SM region are topologically nontriv- ial [37]. We performed measurements using a single-shot THz time-domain polarimetry setup combined with a 30- T pulsed magnet system [38 -- 40]. The wide tunability of the band structure of Bi1−xSbx with x makes this mate- rial system suitable for identifying and distinguishing the uniquely different magneto-optical responses of samples in the topologically trivial and nontrivial phases. A THz beam normally incident on the sample surface in the Faraday geometry exhibited polarization rotations due to the field-induced Hall conductivities of the surface and/or bulk carriers. Faraday rotation spectra for the SM films had a pronounced dip, which blue-shifted with the magnetic field, while the Faraday rotations in the TI films were positive and spectrally featureless, increas- ing and then saturating with increasing magnetic field. Using a theoretical model incorporating realistic band parameters and the Kubo formula for calculating the op- tical conductivity, we found that the optical Hall signals in the SM (TI) samples can be attributed to carriers in the bulk (surface) bands. The model suggested that the magneto-optical signal from the SM films was dominated by the cyclotron resonance of bulk high-mobility holes, while that from the TI film resulted from the summed contributions of multiple electron and hole pockets asso- ciated with the surface bands. FIG. 1. Schematic phase diagram of Bi1−xSbx alloys. The energies of different band edges at high symmetry points (T and L points) are plotted as a function of Sb composition x. The five arrows pointing to the horizontal axis mark the nominal values of x of the five film samples we studied. See Table I for more details of the characteristics of the samples. 2 II. SAMPLES AND METHODS A. Bi1−xSbx films We studied Bi1−xSbx films on silicon substrates grown by molecular beam epitaxy using the methods described in Refs. 14 and 41. Five samples with different x values were studied; see Table I. The nominal x values were 0, 0.04, 0.08, 0.1, and 0.15, respectively, while the thick- ness, t, was nominally 40 nm for all films. According to Fig. 1, Samples 1 and 2 were in the SM regime while Samples 3 -- 5 were in the TI regime. We used a combina- tion of structural and chemical characterization methods to precisely determine the actual values of x, t, and crys- tal orientation of the films; in-situ reflection high-energy electron diffraction (RHEED) patterns determined the crystal orientation, while ex-situ x-ray diffraction (XRD), x-ray fluorescence (XRF), and atomic force microscopy (AFM) experiments provided information on the crystal structure, chemical composition, and film thickness, re- spectively. The obtained parameters of the samples are summarized in Table. I. For Sample 1 (x = 0), RHEED determined that the film orientation was (cid:104)001(cid:105). No Sb was incorporated in this sample, so that chemical analysis was not needed. AFM determined that t = 68 nm. For Sample 2 (nom- inal x = 0.04), XRF and XRD measurements were per- formed. An obtained XRF spectrum was fit with a model built in the measurement software, using x and t as ad- justable parameters, and the parameters that gave the best fit were x = 0.03 and t = 60 nm. XRD showed a dominating diffraction peak due to the (001) plane of the film, confirming that the crystal orientation was (cid:104)001(cid:105). For Sample 4 (nominal x = 0.10), fitting analysis on an XRF spectrum allowed us to determine x = 0.136 and t = 54 nm. An XRD curve showed diffraction peaks from the (001) and (012) planes, suggesting that the film possibly had some spatial inhomogeneity in terms of orientation; however, the ARPES data shown in Fig. 2(a) confirmed that the film area on which our magneto-optical measure- ments were performed was dominated by the (cid:104)001(cid:105) ori- entation. Transport measurements were also performed on the film; see Fig. 2(b) for the resistance-temperature (R-T ) characteristic. The increasing R with decreasing T in the 120 K < T < 250 K region can be attributed to the decreasing number of thermal carriers in the in- sulating bulk, while the subsequent decrease of R in the T < 120 K region is likely due to surface metallicity; such behavior has been observed in R-T curves measured for TI systems with minimal bulk doping [24, 25]. For Samples 3 (nominal x = 0.08) and 5 (nominal x = 0.15), we found that the uncertainties given by model fits to their XRF spectra were too large to determine both x and t accurately. However, as the amount of Sb incorporation was progressively increased in growing the five films, from Sample 1 to Sample 5, it is certain that Sample 3 had an x value that is between those of Samples 2 and 4 while Sample 5 had x > 0.136. 𝑥Energy (a.u.)Semimetal(SM)Topological Insulator(TI)TLsLa𝑥=0𝑥=0.04𝑥=0.08𝑥=0.1𝑥=0.15Sample No. Nominal x Actual x Thickness t (nm) 1 2 3 4 5 0 0.04 0.08 0.10 0.15 0 0.03 (XRF) 0.03 < x < 0.136 68 (AFM) 60 (XRF) 70 (AFM) 0.136 (XRF) 54 (XRF) x > 0.136 77 (AFM) Orientation (cid:104)001(cid:105) (RHEED) (cid:104)001(cid:105) (XRD) N/A (cid:104)001(cid:105) and (cid:104)012(cid:105) (XRD) (cid:104)001(cid:105) (ARPES) (cid:104)001(cid:105) (RHEED) 3 Character SM SM TI TI TI TABLE I. Characteristics of the five Bi1−xSbx thin film samples studied. The Sb content (x), film thickness (t), and crystal orientation of the films are shown. The experimental methods used to determine the parameters are indicated in the parentheses. SM = semimetal. TI = topological insulator. FIG. 3. THz polarimetry setup probing the magneto-optical response of the Bi1−xSbx films. Both Faraday and Kerr rota- tion angles can be measured with the transmission geometry. The incident THz beam was linearly polarized in the x direction, but both the x and y components of the electric field (Ex and Ey) of the transmitted field were measured by electro-optic sampling in the time domain to quantify the THz polarization state that was affected by the car- rier Hall effect. Because the THz waveforms were both amplitude- and phase-resolved, we were able to detect polarization rotations as small as 1 mrad as well as the corresponding ellipticity change. In addition, the pulse- based technique allowed us to measure both Faraday and Kerr rotations using only the transmission geometry. As shown in the upper right schematic in Fig. 3, the THz pulse that directly passes through the film and the substrate gives the Faraday rotation. However, a por- tion of the pulse experiences additional reflection events at the vacuum-substrate interface and the film-vacuum interface, and can still be measured as a back reflection pulse in the same alignment geometry; the back reflec- tion pulse appears later in time than the main pulse that is directly transmitted due to its additional path inside the substrate, but it contains both Faraday and Kerr ro- FIG. 2. Electronic properties of Sample 4. (a) Fermi surface mapping using ARPES. Black solid line marks the fitting of Fermi energy by assuming a Dirac velocity of 2600 meVA. (b) Resistance versus temperature characteristic. B. THz polarimetry Figure 3 schematically shows the THz polarimetry technique we used to probe the magneto-optical response of the Bi1−xSbx films; similar techniques have previ- ously been used to study systems other than TIs [42 -- 44]. Our laser system was a Ti:sapphire regenerative amplifier (1 kHz, 150 fs, 775 nm, Clark-MXR, inc.), which gener- ated and detected THz probe pulses with ZnTe crystals. Intensity (a.u.)(a)(b)0.30.20.10-0.1-0.2-0.3ky(Å−1)1.61.41.21.00.80.60.40.200.10.20.30.40.50.6kx(Å−1)9.08.98.88.78.68.58.4Resistance ()25020015010050Temperature (K)BFaraday Faraday + Kerr (cid:2016)Bi1-xSbxfilmSi substrateBi1-xSbxfilm(Si substrate)Linearly polarizedTHz probe pulse(cid:1876)(cid:1877)tation signals. The Kerr rotation signal can be isolated out by subtracting the Faraday rotation from the total polarization rotation of the back reflection pulse. Below, we show the process of determining the polar- ization state of a pulse. We start with measurements of the Ex(t) and Ey(t) of a pulse in the time domain. Since the Ey(t) signal is usually much smaller than Ex(t), we changed the polarity of the magnetic field B and took the average of the subtraction [Ey(+B) − Ey(−B)]/2 as the real Ey(t) signal that excludes any effect of imperfect linear polarization of the incident beam. We Fourier- transformed Ex(t) and Ey(t) into complex-valued Ex(ω) and Ey(ω), which are functions of frequency ω. We then defined the modes in the circular polarization basis, E±(ω) = [Ex(ω) ± iEy(ω)]. 1√ 2 (1) Then, the rotation of the polarization plane and change of ellipticity can be quantified, respectively, as θ(ω) = η(ω) = arg[E+(ω)] − arg[E−(ω)] E−(ω) − E+(ω) E−(ω) + E+(ω) . 2 , (2) (3) In addition, the real and imaginary parts of the complex- valued longitudinal and Hall conductivity of the film can be determined if a reference pulse signal Eref is collected as a THz pulse passed through the bare substrate without the film. We first calculated the complex-valued conduc- tivity in the circular basis σ±(ω) = (1 + nSi)(1 − t±) Z0dt± , (4) where nSi = 3.5 is the refractive index of the silicon sub- strate, Z0 = 377 Ω is the vacuum impedance, d is the thickness of the film, and t± = E±(ω)/Eref,±. Then, the longitudinal and Hall conductivities of the film can be obtained, respectively, as σxx = σxy = σ+(ω) + σ−(ω) , −σ+(ω) + σ−(ω) 2 2i . (5) (6) Just like the complex-valued longitudinal and Hall con- ductivity (σxx and σxy), the combination of the polariza- tion rotation angle and ellipticity change (θ and η) con- tains complete information on the THz magneto-optical response of the film, with the only difference that a ref- erence signal is not needed in determining θ and η. Electronic properties of a sample, irrespective of whether it is a SM or TI, contain a mixture of contri- butions from electrons and holes. Sometimes one carrier type can be dominant, and determining the carrier type gives valuable information on the origin of the optical conductivity signal. We can use the sign of the Hall con- ductivity to determine whether the dominant carriers are 4 FIG. 4. Surface Hall conductivity of a standard GaAs 2DEG sample at 3 T. The electron cyclotron resonance appears as a positive peak, which was used to determine the dominant carrier type in the Bi1−xSbx samples. electrons or holes. We performed THz transmission mea- surements on a standard GaAs two-dimensional electron gas (2DEG) sample at 3 T, where the electron cyclotron resonance peak appeared in the middle of the THz spec- trum, and calculated its optical Hall conductivity. As shown in Fig. 4, both the real and imaginary parts of the Hall conductivity are positive at the cyclotron resonance peak. We used this sign as a reference to determine the dominant carrier type in the Bi1−xSbx samples. C. Single-shot THz spectroscopy in pulsed high magnetic fields In addition to using a standard delay-stage-based step- scan THz setup combined with a 10-T superconducting magnet [45 -- 52], we used a unique single-shot THz setup to perform measurements in pulsed magnetic fields up to 30 T [40]. Pulsed magnets typically generate higher mag- netic fields with a smaller magnet and cryostat size than DC magnets, but the challenge is that the peak magnetic field only lasts for a short time (approximately 400 µs in our case [38]), so that the time for a delay stage to step through the THz waveform is insufficient. We developed a single-shot THz time-domain detector, which is capable of measuring the full THz time-domain waveform using just one laser pulse within the 400-µs-long time window during which the sample is experiencing the peak of the pulsed magnetic field. Single-shot THz detection relied on a reflective echelon mirror tilting the pulse front of the optical gate beam by forming time-delayed beamlets, which encoded time de- lay information across the beam intensity profile [53 -- 56]; see the inset in Fig. 5. The temporal and spatial overlap of the gate beam and the THz beam was achieved at the 1.2x10-21.00.80.60.40.20.0-0.2Surface Hall conductivity (1/W)2.52.01.51.00.5Frequency (THz) Real part Imaginary part2DEG sample, 4 K, 3 Tesla5 FIG. 5. Schematic diagram of the single-shot THz spec- troscopy system. At the peak of the pulsed magnetic field, an optical gate pulse incident onto a reflective echelon mirror is converted into time-delayed beamlets, which enable map- ping out the time-domain THz electric field waveform in a single shot without involving any delay stage. electro-optical crystal. A combination of a quarter-wave plate, a Wollaston prism, and a silicon complementary metal-oxide-semiconductor (CMOS) camera was used to detect the THz-induced change of the gate beam polar- ization. The CMOS camera captured the full image of the gate beam, which contained information about the THz electric field at various time delays. The difference between the two images spatially separated on the CMOS camera by the Wollaston prism gave the time-domain THz electric field signal. III. RESULTS A. Semimetallic samples: Samples 1 and 2 Figures 6(a) and (b) display Faraday rotation (θF) and Faraday ellipticity (ηF) spectra, respectively, for Sam- ple 1 (x = 0) at T = 2 K in B up to 10 T. The curves are intentionally offset vertically for clarity, and the zero baselines are shown as dashed colored lines. A resonance feature that shifts higher in frequency with increasing B is clearly observed. We then calculated the real and imaginary parts of the optical Hall conductivity (σxy) of the sample by taking into account the reference signal; see Eq. (6). Figure 7 shows the calculated σxy spectra. The signs of the real and imaginary parts of σxy are both opposite to that obtained from the standard 2DEG sam- ple shown in Fig. 4. This indicates that holes are the major contributors to the magneto-optical signal. Figures 8(a) and (b) display θF and ηF spectra, re- spectively, for Sample 2 at T = 2 K in B up to 10 T. The major resonance feature that shifts with B in a sim- FIG. 6. (a) Faraday rotation and (b) Faraday ellipticity spec- tra for Sample 1 at a temperature of 2 K at different magnetic fields up to 10 T. Curves at different magnetic fields are ver- tically offset for clarity, and the baselines for the different spectra are indicated by dashed lines. ilar manner to that in Sample 1 is observed, except that the linewidth is broader. This suggests that the major contributors to the magneto-optical signal have the same carrier origin as in the Bi film, and the Sb incorporation reduces the carrier mobility. Figures 9(a) and (b) show Kerr rotation (θK) and Kerr ellipticity (ηK) spectra, re- spectively, obtained for the same sample under the same conditions as in Fig. 8. The resonance feature induces a drastic response in the θK spectra; an abrupt π phase shift of the rotation angle appears at the resonance fre- quency at each B. The ηK spectra, on the other hand, do not show a clear trend, likely due to an insufficient signal-to-noise ratio. We also performed measurements on Sample 2 in B up to 30 T at T = 21 K, using the single-shot THz detection setup described in Section II C. Figures 10(a) and (b) show, respectively, the extracted θF and θK versus B for a fixed THz frequency (0.7 THz). Resonance behavior is again clearly observed in both θF and θK. θF shows a dip, while θK shows a π phase shift at the resonance magnetic field. At higher B, both quantities decrease in magnitude and tend to zero. EchelonTime (ps)EfieldGate pulseTime (ms)Magnetic Field (T)30 TTime-delayedbeamlets1.51.00.50.0F 2.01.0Frequency (THz)(a)(b)1 T2 T3 T4 T5 T6 T7 T8 T9 T10 T1 T2 T3 T4 T5 T6 T7 T8 T9 T10 T(cid:1876)=01.51.00.50.0F (rad)2.01.0Frequency (THz)6 FIG. 7. (a) Re(σxy) and (b) Im(σxy) spectra for Sample 1 at a temperature of 2 K at different magnetic fields up to 10 T. Curves at different magnetic fields are vertically off- set for clarity, and the baselines for the different spectra are indicated by dashed lines. FIG. 8. (a) Faraday rotation and (b) Faraday ellipticity spec- tra for Sample 2 at a temperature of 2 K at different magnetic fields up to 10 T. Curves at different magnetic fields are ver- tically offset for clarity, and the baselines for the different spectra are indicated by dashed lines. B. Topological insulator sample: Sample 4 Faraday rotation, Kerr rotation, Faraday ellipticity, and Kerr ellipticity spectra for Sample 4 (nominal x = 0.1 at T = 2 K in B up to 10 T are shown in Figs. 11(a)- (d). No curve is intentionally offset. The Faraday and Kerr rotations are featureless as a function of frequency, but both clearly show an increasing trend as B increases. θF is smaller than θK for a given B, but its signal-to- noise ratio is higher because the back reflection pulse from which θK is derived has a smaller amplitude than the main pulse. Neither ηF nor ηK shows any signal that can be clearly distinguished from the noise floor for all B, suggest- ing that the carriers in the TI film induce pure rota- tions without any ellipticity change in the THz probe light. This behavior can be directly observed in the time- domain THz waveform data, as shown in Fig. 11(e). The incident THz pulse is polarized in the x direction. It is clear that the amplitude of the transmitted Ex does not vary much between 1 T and 10 T; at both magnetic fields, the main pulse appears at 3.5 ps, together with a smaller back reflection pulse at 12.5 ps. However, the Ey wave- form at 10 T is much larger than that at 1 T, suggesting a much larger polarization rotation; this is consistent with Figs. 11(a) and (b). In addition, as marked by the ver- tical dashed lines at the pulse peaks, the phase of Ey also matches well with Ex for both the main and back reflection pulses. This suggests that both pulses remain linearly polarized. Only the polarization plane is rotated with respect to the incident beam, and the ellipticity does not change. We further performed measurements in B up to 30 T on the same sample at T = 21 K. Figures 12(a) and (b) show, respectively, θF and θK versus THz frequency and B. The data obtained using the 10-T magnet system (Fig. 11) is also included in Fig. 12, agreeing with the data obtained using the 30 T system. We extracted the evolutions of θF and θK with B for a fixed THz frequency (0.7 THz), and the results are plotted in Figs. 12(c) and (d), respectively. We observe that θF and θK increase with increasing B until 15 T, above which both quantities saturate with further increasing B. C. All samples Figure 13 displays θF versus B at T = 2 K for a fixed THz frequency (0.7 THz) for all samples of Bi1−xSbx films. As x increases, the system moves from the SM to the TI regime in the phase diagram (Fig. 1). We can -1.5x106-1.0-0.50.0Re(xy) (-1m-1)2.01.0Frequency (THz)-1.5x106-1.0-0.50.0Im(xy) (-1m-1)2.01.0Frequency (THz)(a)(b)1 T2 T3 T4 T5 T6 T7 T8 T9 T10 T1 T2 T3 T4 T5 T6 T7 T8 T9 T10 T(cid:1876)=00.80.60.40.20.0F (rad)2.01.0Frequency (THz)0.80.60.40.20.0F2.01.0Frequency (THz)(a)(b)1 T2 T3 T4 T5 T6 T7 T8 T9 T10 T1 T2 T3 T4 T5 T6 T7 T8 T9 T10 T(cid:1876)=0.047 bulk bands of Bi, which allowed us to determine the ori- gin of the experimentally observed magneto-optical sig- nal for Sample 1. The reason for choosing the Bi sample instead of Bi0.96Sb0.04 for this analysis is because the Bi sample showed a much sharper resonance feature. The bulk band structure we considered is schemati- cally depicted in Fig. 14 [57, 58]. There is an indirect negative band gap between the valence band maximum FIG. 9. (a) Kerr rotation and (b) Kerr ellipticity spectra for Sample 2 at a temperature of 2 K at different magnetic fields up to 10 T. Curves at different magnetic fields are vertically offset for clarity, and the baselines for the different spectra are indicated by dashed lines. see a clear trend of magneto-optical response accompa- nying this transition. From the Bi and Bi0.96Sb0.04 films in which a bulk hole cyclotron resonance appears as neg- atively valued dips within 10 T, to the Bi0.9Sb0.1 film where the low-mobility surface electrons lead to an in- creasing positive θF with increasing B, the change of θF curves with increasing x is monotonic and smooth. FIG. 10. (a) θF and (b) θK versus magnetic field for a fixed THz frequency of 0.7 THz obtained for Sample 2 at T = 21 K in B up to 30 T. IV. DISCUSSION A. Theory of magneto-optical response of Bi1−xSbx in the semimetallic regime In order to understand the THz magneto-optical re- sponse of Bi1−xSbx films in the SM regime, we developed a detailed theoretical model. We took into account the at the T point and the conduction band minima at the three equivalent L points (later we refer to these as a single point, L). For a (001) Bi film, the hole pocket at the T point has an isotropic in-plane effective mass and a parabolic dispersion relation, while the bands at the L point host Dirac electrons with a hyperbolic dispersion, but a small gap 2∆ exists at the L point. The Hamiltonian for the T-point holes and the L-point electrons are 302520151050K (rad)2.01.0Frequency (THz)86420K2.01.0Frequency (THz)(a)(b)1 T2 T3 T4 T5 T6 T7 T8 T9 T10 T1 T2 T3 T4 T5 T6 T7 T8 T9 T10 T(cid:1876)=0.041.51.00.50.0-0.5-1.0-1.5K (rad)302520151050Magnetic Field (T)-8x10-2-6-4-20F (rad)302520151050Magnetic Field (T)(a)@ 0.7 THz(b)𝑥=0.04@ 0.7 THz8 FIG. 11. Magneto-optical response of Sample 4 (nominal x = 0.1) at T = 2 K in B up to 10 T. (a) Faraday rotation, (b) Kerr rotation, (c) Faraday ellipticity, and (d) Kerr ellipticity spectra at B from 1 to 10 T are displayed. No curve is intentionally offset. (e) Time-domain waveforms of Ex and Ey at 1 T and 10 T. Ey curves are multiplied by a factor of 25 and offset vertically for clarity. Black dashed lines are guides to the eye for identifying the pulse peak positions. Hh = E0 He = h − ¯h2(k2 x + k2 y) 2Mc ∆ 0  −i¯hvzkz −i¯hv(kx + iky) − ¯h2k2 z 2Mz 0 ∆ −i¯hv(kx − iky) i¯hvzkz i¯hv(kx + iky) i¯hvzkz −∆ 0  , i¯hv(kx − iky) −i¯hvzkz 0 −∆ (7) (8) where E0 h = 38.5 meV is the T-point band edge off- set, k represents the wave vector, Mc = 0.0677m0 and Mz = 0.721m0 are, respectively, the in-plane and out-of- plane hole effective masses, m0 is the free electron mass, ∆ = 7.65 meV is half of the L-point gap, and v and vz are the in-plane and out-of-plane electron Dirac velocities, respectively. Because our film has a finite thickness, we considered the quantum confinement effect on kz along the growth direction for both the hole and electron pock- ets. As schematically shown in Fig. 14, many hole sub- bands described by discrete kz's with quantum number Nh are above the Fermi energy EF, while only two elec- tron subbands (described by quantum number Ne) are filled at the L point; the electrons are more strongly in- fluenced by quantum confinement than the holes because of the lighter electron mass. 1.6x10-41.20.80.40.0Electric field (a.u.)151050Time (ps)5x10-243210F (rad)2.52.01.51.00.5Frequency (THz)0.250.200.150.100.050.00K (rad)2.52.01.51.00.5Frequency (THz)1-10 T1-10 T(a)𝐸𝑦×25, 1 T𝐸𝑦×25, 10 T𝐸𝑥, 1 T𝐸𝑥, 10 T(b)(c)(d)(e)𝑥=0.1-0.04-0.020.000.020.04F2.52.01.51.00.5Frequency (THz)-0.2-0.10.00.10.2K2.52.01.51.00.5Frequency (THz)9 FIG. 12. Magneto-optical response of Sample 4 (nominal x = 0.1) up to 30 T. (a) Faraday rotation and (b) Kerr rotation maps versus THz frequency and magnetic field. (c) Faraday rotation and (d) Kerr rotation versus magnetic field at a fixed THz frequency of 0.7 THz, corresponding to the cuts marked by the red dashed lines in (a) and (b). The solid red lines in (c) and (d) are calculated curves using the theoretical model described later in the Discussions section. The key for the magneto-optical response of carriers is to calculate the optical conductivity tensor given by the Kubo formula: (cid:88) m,n σαβ(ω) = i¯h S fm − fn Em − En (cid:104)ΨmjαΨn(cid:105)(cid:104)ΨnjβΨm(cid:105) ¯hω + Em − En + iγ , (9) where α and β take choices between x and y, S is the sample area, fm (fn), Em (En), and Ψm(cid:105) (Ψn(cid:105)) are, re- spectively, the occupation factor calculated by the Fermi- Dirac distribution function, energy, and eigenfunction of the mth (nth) eigenstate of the system Hamiltonian, jα and jβ are current operators, and γ is the scattering rate responsible for transition line broadening. We calculated the conductivity tensors contributed by each pocket and later added all their contributions. In a magnetic field, the Landau-level eigenenergies for the hole and electron pockets are (cid:18) (cid:19) ¯heB Mc Eh,n = Eh (cid:115) 0 − n + 1 2 (cid:18) Ee,n = ∆2 + 2∆ n + − ¯h2k2 z 2Mz (cid:19) ¯heB + shgµBB (10) 1 2 + se + ¯h2k2 z 2mz , (11) mc where sh = ±1/2 and se = ±1/2 are, respectively, the hole and electron spin quantum number, n represents the Landau-level index, g = 62.6 is the hole g factor, µB is the Bohr magneton, mc and mz are, respectively, the in-plane and out-of-plane electron effective masses. The calculated spin-resolved Landau-level energies for elec- trons and holes are plotted in Fig. 15(a) and (b), respec- tively. The results are displayed up to Ne = 2 and n = 2 (Nh = 10 and n = 2) for electrons (holes), where Ne and Nh are the electron and hole subband index, respectively. Then we calculated the matrix elements of the current operators, (cid:104)ΨmjαΨn(cid:105), for both electron and hole pock- ets. For the electron pocket, the situation is complicated because the eigenspinor of the Dirac Hamiltonian above takes separate forms for spin-up and spin-down electrons  An,↓n − 1(cid:105)  ,  An,↑n(cid:105) Bn,↓n(cid:105) Cn,↓n − 1(cid:105) Dn,↓n(cid:105) Bn,↑n + 1(cid:105) Cn,↑n(cid:105) Dn,↑n + 1(cid:105) Ψn,↓ = Ψn,↑ = (12) (13) where ↓ and ↑ denote electron spin, and An,↓, Bn,↓, Cn,↓, Dn,↓, An,↑, Bn,↑, Cn,↑, and Dn,↑ are the coefficients to be determined by the eigenvalue equation. Therefore, there are eight matrix elements of the current operator describ- ing possible transitions between Landau levels across Frequency (THz)𝜃𝐹𝜃𝐾(a)(b)(c)(d)01020300102030Magnetic Field (T)0.070.060.050.040.030.020.0100.250.20.150.10.0501.21.110.90.80.70.60.51.21.110.90.80.70.60.58x10-26420F (rad)3020100Magnetic Field (T) Experiment Theory0.300.250.200.150.100.050.00K (rad)3020100Magnetic Field (T) Experiment Theory@ 0.7 THz@ 0.7 THz𝑥=0.110 FIG. 13. Magneto-optical response of all samples. θF is plot- ted versus B at T = 2 K for a fixed THz frequency (0.7 THz). See Table I for the x values for the samples. different spin channels. (cid:104)Ψm,↑jxΨn,↑(cid:105), (cid:104)Ψm,↓jyΨn,↓(cid:105), (cid:104)Ψm,↑jyΨn,↓(cid:105). These are (cid:104)Ψm,↓jxΨn,↓(cid:105), (cid:104)Ψm,↑jxΨn,↓(cid:105), and (cid:104)Ψm,↓jyΨn,↑(cid:105), (cid:104)Ψm,↓jxΨn,↑(cid:105), (cid:104)Ψm,↑jyΨn,↑(cid:105), We calculated optical conductivity tensors by plugging the matrix elements of the current operator above into the Kubo formula. The total conductivity of each car- rier pocket was obtained by summing up transitions be- tween all possible states. Each state has five indices, i.e., the Landau-level index, spin index, band index describ- ing valence or conduction band, subband index, and the pocket index due to the existence of multiple equivalent pockets. As an example, we show the transitions consid- ered for the hole pocket within the Nhth subband and the electron pocket within the Neth subband in Fig. 16. The hole pocket contains only the spin-conserving inter- Landau-level transitions. The electron pocket contains intraband inter-Landau-level transitions and interband magneto-optical transitions across different spin channels and Landau-level indices. Note that intersubband transi- tions are not allowed in either the hole or electron pocket due to the polarization selection rule. Finally, we summed up the conductivities of the elec- tron and hole pockets to obtain the total optical con- ductivity tensor of the Bi film. We adjusted three pa- rameters to try to match the theoretically calculated magneto-optical response with the experiments. They are the Fermi energy EF, the scattering rate of holes at FIG. 14. Band alignment in the theoretical model used to calculate the magneto-optical response of the Bi film. The bottom panel emphasizes quantum confinement effects in the film case compared to the bulk case shown in the top panel. the T point γT h , and the scattering rate of electrons at the L point γL e . The optimized parameters that achieve the best agreement between theory and experiment are EF = 24 meV, γT e = 0.9 meV. The calculated θF and ηF spectra for the Bi film sample at 2 K up to 10 T are shown in Fig. 17; the curves show good agreement with the experimental data in Fig. 6. h = 0.75 meV, and γL We provide some comments on the optimized param- eters. First, the extracted Fermi energy EF = 24 meV is lower than the value (28 meV) used in previous stud- ies [57, 58]. The reason can be that the bands are modi- fied compared to the true bulk Bi due to the band quan- tizations resulting from the finite thickness of the film. Second, γT e = 0.9 meV are both much smaller than the bandwidth of our THz setup. This therefore allows cyclotron resonance signals to be ob- served. From the calculated results in Fig. 17, we found h = 0.75 meV and γL -0.10-0.050.000.05qF (rad)1086420Magnetic field (T)Sample 1Sample 2Sample 3Sample 4Sample 5@ 0.7 THzT(Hole)L(Electron)−2∆0EnergyBi 𝐸ℎ0𝐸F−2∆0Energy…T(Hole)……L(Electron)Bi film 𝐸ℎ0𝐸F11 FIG. 16. Transitions considered for the hole pocket within the Nhth subband and the electron pocket within the Neth subband in Bi. States are labeled by their Landau-level in- dex and spin polarization. The black arrows indicate allowed magneto-optical transitions. B. Theory of magneto-optical response of Bi1−xSbx in the topological insulator regime We used a surface band model to analyze the ex- perimental observations made in Sample 4, a TI sam- ple [17, 36]; see Fig. 19. The (001) surface of Bi0.9Sb0.1 has two spin-polarized surface bands, S1 and S2. The dispersion relation along the ¯Γ- ¯M line has the following features. The bottom of the electron pocket formed by the S1 band is located at the ¯Γ point; the band has a linear dispersion, and its spin texture is similar to that of typical TI surface states with spin-momentum locking. Away from the ¯Γ point, the S1 band bends up and then down to intersect with the Fermi surface again, forming an anisotropic electron pocket near the ¯M point (but not enclosing it); later we refer to it as the ¯M-point electron pocket. On the other hand, the S2 surface band forms an anisotropic hole pocket in the middle of the ¯Γ- ¯M line. Both the ¯M-point electron pocket and the hole pocket have six replicas in the first Brillouin zone, which are related to each other by a six-fold rotational symmetry with respect to the ¯Γ point. We denote the ¯M-point elec- tron pocket and the hole point along the ¯Γ- ¯M line to be e1 and h1, respectively, and their replicas as e2··· , e6, and h2··· , h6. The model Hamiltonians for the ¯Γ-point electron pocket, the h1 hole pocket, and the e1 electron pocket FIG. 15. Landau-level energies of the (a) electron and (b) hole pockets in Bi. The results are displayed up to Ne = 2 and n = 2 (Nh = 10 and n = 2) for electrons (holes), where Ne and Nh are the subband indices for the electron and hole pocket, respectively, and n represents the Landau-level index. that the dip feature that moves to higher frequency with increasing magnetic field is mainly due to hole cyclotron resonance. The electron cyclotron resonance feature, on the other hand, appears very close in frequency with the hole cyclotron resonance feature, but it is inhomoge- neously broadened because cyclotron transition frequen- cies are different for different electron subbands. The effect of electron cyclotron resonance can still be identi- fied in the calculated results in Fig. 17. We plotted the θF and ηF spectra calculated by only taking into account the hole pocket in Fig. 18. It can be easily seen that the hole pocket alone cannot explain the asymmetric lineshapes of θF and ηF in Fig. 17 and in the experimental data in Fig. 6. electron ↑electron ↓40(a)00246810302010Energy (meV)hole ↑hole ↓4000246810Magnetic Field (T)302010(b)Energy (meV)Magnetic Field (T)…Nhthsubband0↑1↑2↑…3↑Nethsubband1↓, 0↑2↓, 1↑3↓, 2↑0↓1↓, 0↑2↓, 1↑3↓, 2↑0↓4↓, 3↑…………EF0↓1↓12 FIG. 17. Calculated (a) Faraday rotation and (b) Faraday ellipticity spectra for the Bi film at T = 2 K in B up to 10 T. Curves at different B are vertically offset intentionally for clarity, and zero baselines are marked by dashed lines. FIG. 18. Calculated (a) Faraday rotation and (b) Faraday ellipticity spectra for the Bi film at T = 2 K in B up to 10 T by considering only the hole pocket. Curves at different B are vertically offset intentionally for clarity, and zero baselines are marked by dashed lines. are H¯Γ e = vDσ · p Hh1 = E0 H ¯M e1 = E0 e + h − (px − ph0)2 (px − pe0)2 2mh x 2me x (14) (15) (16) − p2 y 2mh y p2 y 2me y + , where vD = 2600 meVA is the Dirac velocity, p rep- resents the carrier momentum, E0 h = 170 meV and E0 e = 129.4 meV are the band-edge energies of the hole pocket and the ¯M-point electron pocket, respectively, measured from the Dirac point, ph0 = ¯h × 0.25 meV/A and pe0 = ¯h× 0.67 meV/A are the displacements of band centers of the h1 and e1 pockets, respectively, and in- plane anisotropic effective masses can be described as mh y = 0.05m0, where m0 is the free electron mass. The surface band dispersions calculated by applying these parame- ters are shown in Fig. 19(a). The Fermi surfaces of the pockets within one quadrant of the first Brillouin zone are shown in Fig. 19(b); the Fermi level is obtained from a fit to the Fermi surface of the ¯Γ-point electron pocket measured by ARPES, as shown in Fig. 2(a). x = 1.5m0, and me x = 1.2m0, mh y = 0.2m0, me First, in a magnetic field, the Hamiltonian for the ¯Γ-  , 2¯hvD (cid:96)B a point electron pocket is given by √ He = 0 √ where lB =(cid:112)¯h/eB is the magnetic length, and a (a†) 2¯hvD (cid:96)B a† 0 is the Landau-level raising (lowering) operator. Its eigen- state spinor takes the form (17) Ψn(cid:105) = , (18) Bnn(cid:105) (cid:18)Ann − 1(cid:105) (cid:19) (cid:18)An (cid:19) Bn n √ (cid:18)An (cid:19) Bn . (19) = En  (cid:115) where n is the Landau-level index, and An and Bn are coefficients to be determined by the following eigenvalue problem: √ 0 2¯hvD (cid:96)B √ 2¯hvD (cid:96)B 0 √ n The Landau-level eigenenergies are obtained as En = sgn(n) 2¯h2v2 D (cid:96)2 n. n = 0, 1, 2,··· (20) (a)(b)1 T2 T3 T4 T5 T6 T7 T8 T9 T10 T1 T2 T3 T4 T5 T6 T7 T8 T9 T10 T(cid:1876)=01.51.00.50.0F (rad)2.01.0Frequency (THz)1.51.00.50.0F 2.01.0Frequency (THz)(a)(b)1 T2 T3 T4 T5 T6 T7 T8 T9 T10 T1 T2 T3 T4 T5 T6 T7 T8 T9 T10 T(cid:1876)=01.51.00.50.0F (rad)2.01.0Frequency (THz)1.51.00.50.0F 2.01.0Frequency (THz)See Fig. 20 for the calculated eigenenergies versus mag- netic field up to the n = 10th Landau level. The corre- sponding coefficients are found to be (cid:114) 1 2 An = sgn(n) (cid:114) 1 . 2 Bn = 13 (21) (22) (23) Given that the current operator is given by j = [He, x], ie ¯h FIG. 19. Surface band model used in the theoretical analysis for Bi1−xSbx in the topological insulator regime. (a) Band dispersions along the ¯Γ- ¯M line. The black dashed line marks the position of the Fermi energy. The black dashed-dotted lines mark the positions of the ¯Γ and ¯M points. (b) Fermi surface within one of the quadrants of the first Brillouin zone. Solid black lines are the boundaries of the first Brillouin zone. its matrix elements can be calculated as (cid:104)ΨmjxΨn(cid:105) = evD(AmBnδm−1,n + BmAnδm,n−1) (24) (cid:104)ΨmjyΨn(cid:105) = ievD(−AmBnδm−1,n + BmAnδm,n−1), (25) where δm−1,n and δm,n−1 are Kronecker's δ. By plugging the matrix elements above into the Kubo formula, we obtained the longitudinal and Hall conductivities for the ¯Γ-point electron pocket σ(e,¯Γ) xx = σ(e,¯Γ) xy = i¯he2v2 D S ¯he2v2 D S (cid:88) (cid:88) m m fm − fm+1 Em − Em+1 fm − fm+1 Em − Em+1 (cid:18) (cid:18) A2 m+1B2 m 1 ¯hω + Em − Em+1 + iγ ¯Γ e + 1 ¯hω + Em+1 − Em + iγ ¯Γ e (26) A2 m+1B2 m 1 ¯hω + Em − Em+1 + iγ ¯Γ e − 1 ¯hω + Em+1 − Em + iγ ¯Γ e , (27) (cid:19) (cid:19) where γ ¯Γ ¯Γ-point electron pocket. e represents the scattering rate of electrons in the Second, the Hamiltonian for the hole pocket in a mag- netic field is spatial metric [57]. We scaled the current operator as Hh1 = E0 h − ¯hωh c (a†a + 1/2), (28) where ωh c is the hole cyclotron frequency. The form is simple because the pocket can be assumed to have a parabolic band dispersion. Landau-level energies depend linearly on the magnetic field; see Fig. 20. Regarding the treatment of effective mass anisotropy for the hole pocket, it has been found that the system can be transformed into an isotropic model, which is easier for calculation, by performing a scaling procedure on the jx = η−1jx jy = ηjy, (29) (30) where η = (mh x/mh y )1/4 is the scaling factor. Then the Energy (meV)S1S2(a)𝑒1ℎ1 Γ𝑘𝑥(Å−1)S1S2ℎ1𝑒1𝑒2ℎ2 Γ M M00.20.40.60.80-0.1-0.2-0.3-0.4-0.5-0.6-0.7𝑘𝑦(Å−1)(b)-1000100200 Mmatrix elements of jα can be written as (cid:104)mjxn(cid:105) = − e¯h√ = − e¯h√ 2M (cid:96)B 2M (cid:96)B (cid:104)ma + a†n(cid:105) √ nδm,n−1 + ( √ n + 1δm,n+1) (cid:104)mjyn(cid:105) = = ie¯h√ 2M (cid:96)B ie¯h√ 2M (cid:96)B (cid:104)ma† − an(cid:105) √ n + 1δm,n+1 − √ ( nδm,n−1), (34) 14 (31) (32) (33) FIG. 20. Landau-level energies of the three carrier pockets calculated for Bi0.9Sb0.1 within the surface band model. The ¯Γ and ¯M points are depicted in Fig. 19. (cid:113) mh xmh where M = y . Substituting the matrix elements above into the Kubo formula gives the scaled conductiv- ity tensor σ(h1) xx (ω) = σ(h1) xy (ω) = ie2¯h3 2M 2(cid:96)2 BS e2¯h3 2M 2(cid:96)2 BS (cid:18) (cid:18) (cid:88) (cid:88) m m fm − fm+1 Em − Em+1 fm − fm+1 Em − Em+1 m + 1 ¯hω + Em − Em+1 + iγh m + 1 ¯hω + Em − Em+1 + iγh + − m + 1 ¯hω + Em+1 − Em + iγh m + 1 ¯hω + Em+1 − Em + iγh (cid:19) (cid:19) (35) , (36) where γh represents the scattering rate of the hole pocket. The actual conductivity tensor is related to the scaled conductivity tensor by (37) yy (ω) σ(h1) xx (ω) = η2 σ(h1) xx (ω) yy (ω) = η−2 σ(h1) σ(h1) σ(h1) xy (ω) = σ(h1) xy (ω) yx (ω) = −σ(h1) xy (ω) = −σ(h1) σ(h1) The optical conductivity tensor for the rest of hole pockets h2, h3, ··· , h6 can be obtained by six-fold ro- tations. The total conductivity as a sum of contributions from all hole pockets is obtained as xy (ω) = σ(h1) yx (ω). (38) (40) (39) 6(cid:88) (cid:32) i=1 σ(h)(ω) = σ(hi)(ω) = 3 xx (ω) + σ(h1) σ(h1) xy (ω) −2σ(h1) yy (ω) (41) (cid:33) . 2σ(h1) xy (ω) σ(h1) xx (ω) + σ(h1) yy (ω) Finally, the procedure for calculating the conductivity (42) tensor for the ¯M-point electron pocket, σ(e, ¯M) is similar to that of the hole pocket. xx and σ(e, ¯M) xy , We summed up the contributions from all pockets to obtain the total conductivity tensor elements, σtot xx and σtot xy , and calculated the Faraday and Kerr rotations fol- lowing the process discussed in Section II B. All band parameters are either given in the literature or can be obtained through fits to predetermined band structures. The three free parameters we can tune to match the the- oretical results with experimental data are the ¯Γ-point electron scattering rate γ ¯Γ e , the hole scattering rate γh, and the ¯M-point electron scattering rate γ ¯M e . We found that γ ¯Γ e = 10 meV, γh = 60 meV, and γ ¯M e = 10.5 meV give the best fit with the experimental data, as shown by the polarization rotation curves at 0.7 THz in Figs. 12(c) and (d). Maps of calculated Faraday and Kerr rotations as a function of THz frequency and magnetic field using the optimized parameters are shown in Fig. 21. These re- sults can be compared to the experimental maps shown in Figs. 12(a) and (b), and there is agreement between the experimental and theoretical results. The three optimized scattering rates, γ ¯Γ e , γh, and γ ¯M e , electron @ Γholeelectron @ MMagnetic Field (T)2502001501005000102030Energy (meV)15 FIG. 21. Calculated magneto-optical response of Bi0.9Sb0.1 at T = 21 K in B up to 30 T. (a) Faraday rotation and (b) Kerr rotation maps versus THz frequency and magnetic field. The cuts marked by the red dashed lines are plotted together with experimental data in Figs. 12(c) and (d). are all larger than the bandwidth of our THz probe, sug- gesting that the surface carriers do not have high enough mobility for their cyclotron resonance peaks to appear in our measurements. In addition, γh is much larger than γ ¯Γ e and γ ¯M e . This observation is not surprising as we exam- ine the linewidths of surface bands measured by ARPES in previous studies [17], but its effect is that the contri- bution of surface holes in the THz polarization rotation signal is negligibly small compared to that of the surface electrons. We provide some additional comments on the satu- ration behavior of Faraday and Kerr rotations in the B > 15 T region in Figs. 12(c) and (d). It might be tempting to explain these features as the quantized opti- cal Hall effect observed in several recent studies on other TI systems [31, 32, 44]. However, as shown in the cal- culation of θF and θK in a much wider magnetic field range in Fig. 22, we found that the saturation behavior observed within the 30 T magnetic field range arises from the summation of the broadened cyclotron resonance sig- nals contributed by the two electron pockets. Theoretical calculations predict a major quantum Hall plateau in the 60 T < B < 110 T range as the filling factors of the two electron pockets are both small, but experimental obser- FIG. 22. Calculated magneto-optical response of the nominal Bi0.9Sb0.1 film up to 180 T. (a) Faraday rotation and (b) Kerr rotation versus magnetic field at a fixed THz frequency of 0.7 THz. The total rotation signal is plotted together with the separate contributions from the three carrier pockets. vation might still be challenging because the short carrier localization length in optical experiments (compared to DC experiments) significantly shrinks the quantum Hall plateaus [59]. When B > 150 T, the filling factors of the carrier pockets with parabolic dispersions tend to zero, but the Dirac electron pocket at the ¯Γ point gives a fi- nite signal due to the nontrivial Berry's phase created by electrons circling around in momentum space. V. SUMMARY In summary, we performed THz Faraday and Kerr rotation spectroscopy measurements on Bi1−xSbx thin films. This alloy system exhibits a semimetal-to- topological-insulator transition as a function of x. By using single-shot time-domain THz spectroscopy com- bined with a 30-T table-top pulsed magnet, we ob- served distinctly different behaviors between semimetal- (a)(b)0.060.050.040.030.020.010.250.20.150.10.05𝜃𝐹𝜃𝐾Frequency (THz)1.21.110.90.80.70.60.51.21.110.90.80.70.60.501020300102030Magnetic Field (T)𝑥=0.1(a)totalelectron @ Γholeelectron @ M0.080.060.040.020-0.02𝜃𝐹(rad)050100150(b)0501001500.30.250.20.100.150.05-0.05𝜃𝐾(rad)Magnetic Field (T)totalelectron @ Γholeelectron @ MMagnetic Field (T)16 lic (x < 0.07) and topological insulator (x > 0.07) sam- ples. We were able to distinguish the origins of the magneto-optical responses of these films by comparing experimental data with predictions from our theoretical models. We found that a surface (bulk) band model in- cluding some material parameters established in previous studies can completely explain the THz Hall signal of all samples. The combined effort of the THz polarimetry experiments performed in high magnetic fields and the detailed theoretical analysis can be applied to other topo- logical materials to investigate surface and bulk carrier contributions to the optical conductivity. ACKNOWLEDGMENTS This research was primarily supported by the National Science Foundation through the Center for Dynamics and Control of Materials: an NSF MRSEC under Co- operative Agreement No. DMR-1720595. I.K. acknowl- edges support from the Japan Society for the Promotion of Science (JSPS) through the Bilateral Joint Research Project. I.K. and J.T. thank the Ministry of Education, Culture, Sports, Science and Technology (MEXT)/JSPS for support through KAKENHI Grant Nos. 16H06010, 17H06124, and 18H04288. G.A.F. gratefully acknowl- edges support from a Simons Fellowship. [1] X.-L. Qi, T. L. Hughes, and S.-C. Zhang, Phys. Rev. B 47005 (2017). 78, 195424 (2008). [19] B. Lenoir, M. Cassart, J.-P. Michenaud, H. Scherrer, and [2] A. M. Essin, J. E. Moore, and D. Vanderbilt, Phys. Rev. S. Scherrer, J. Phys. Chem. Solids 57, 89 (1996). Lett. 102, 146805 (2009). [20] A. A. Taskin and Y. Ando, Phys. Rev. B 80, 085303 [3] W.-K. Tse and A. H. MacDonald, Phys. Rev. Lett. 105, (2009). 057401 (2010). [21] A. A. Taskin, K. Segawa, and Y. Ando, Phys. Rev. B 82, [4] J. Maciejko, X.-L. Qi, H. D. Drew, and S.-C. Zhang, 121302 (2010). Phys. Rev. Lett. 105, 166803 (2010). [5] W.-K. Tse and A. H. MacDonald, Phys. Rev. B 84, 205327 (2011). [6] J. Wang, B. Lian, X.-L. Qi, and S.-C. Zhang, Phys. Rev. B 92, 081107 (2015). [7] T. Morimoto, A. Furusaki, and N. Nagaosa, Phys. Rev. B 92, 085113 (2015). [22] J. G. Analytis, J.-H. Chu, Y. Chen, F. Corredor, R. D. McDonald, Z. X. Shen, and I. R. Fisher, Phys. Rev. B 81, 205407 (2010). [23] N. P. Butch, K. Kirshenbaum, P. Syers, A. B. Sushkov, G. S. Jenkins, H. D. Drew, and J. Paglione, Phys. Rev. B 81, 241301 (2010). [24] Z. Ren, A. A. Taskin, S. Sasaki, K. Segawa, and Y. Ando, [8] D. Zhang, M. Shi, T. Zhu, D. Xing, H. Zhang, and Phys. Rev. B 82, 241306 (2010). J. Wang, Phys. Rev. Lett. 122, 206401 (2019). [9] F. Wilczek, Phys. Rev. Lett. 58, 1799 (1987). [10] D. Hsieh, D. Qian, L. Wray, Y. Xia, Y. S. Hor, R. J. Cava, and M. Z. Hasan, Nature 452, 970 (2008). [11] D. Hsieh, Y. Xia, L. Wray, D. Qian, A. Pal, J. H. Dil, J. Osterwalder, F. Meier, G. Bihlmayer, C. L. Kane, Y. S. Hor, R. J. Cava, and M. Z. Hasan, Science 323, 919 (2009). [12] Y. Xia, D. Qian, D. Hsieh, L. Wray, A. Pal, H. Lin, A. Bansil, D. Grauer, Y. S. Hor, R. J. Cava, and M. Z. Hasan, Nat. Phys. 5, 398 (2009). [13] D. Hsieh, Y. Xia, D. Qian, L. Wray, J. H. Dil, F. Meier, J. Osterwalder, L. Patthey, J. G. Checkelsky, N. P. Ong, A. V. Fedorov, H. Lin, A. Bansil, D. Grauer, Y. S. Hor, R. J. Cava, and M. Z. Hasan, Nature 460, 1101 (2009). [14] T. Hirahara, Y. Sakamoto, Y. Saisyu, H. Miyazaki, S. Kimura, T. Okuda, I. Matsuda, S. Murakami, and S. Hasegawa, Phys. Rev. B 81, 165422 (2010). [15] A. Nishide, A. A. Taskin, Y. Takeichi, T. Okuda, A. Kak- izaki, T. Hirahara, K. Nakatsuji, F. Komori, Y. Ando, and I. Matsuda, Phys. Rev. B 81, 041309 (2010). [16] F. Nakamura, Y. Kousa, A. A. Taskin, Y. Takeichi, A. Nishide, A. Kakizaki, M. D'Angelo, P. Lefevre, F. Bertran, A. Taleb-Ibrahimi, F. Komori, S.-i. Kimura, H. Kondo, Y. Ando, and I. Matsuda, Phys. Rev. B 84, 235308 (2011). [17] H. M. Benia, C. Strasser, K. Kern, and C. R. Ast, Phys. Rev. B 91, 161406 (2015). [18] Lee, Hwangho, Ko, Kyung-Tae, Park, Byeong-Gyu, Lee, Seungseok, and Park, Jae-Hoon, Europhys. Lett. 118, [25] D.-X. Qu, Y. S. Hor, J. Xiong, R. J. Cava, and N. P. Ong, Science 329, 821 (2010). [26] G. S. Jenkins, A. B. Sushkov, D. C. Schmadel, N. P. Butch, P. Syers, J. Paglione, and H. D. Drew, Phys. Rev. B 82, 125120 (2010). [27] A. D. LaForge, A. Frenzel, B. C. Pursley, T. Lin, X. Liu, J. Shi, and D. N. Basov, Phys. Rev. B 81, 125120 (2010). [28] J. N. Hancock, J. L. M. van Mechelen, A. B. Kuzmenko, D. van der Marel, C. Brune, E. G. Novik, G. V. Astakhov, H. Buhmann, and L. W. Molenkamp, Phys. Rev. Lett. 107, 136803 (2011). [29] R. Vald´es Aguilar, A. V. Stier, W. Liu, L. S. Bilbro, D. K. George, N. Bansal, L. Wu, J. Cerne, A. G. Markelz, S. Oh, and N. P. Armitage, Phys. Rev. Lett. 108, 087403 (2012). [30] L. Wu, W.-K. Tse, M. Brahlek, C. M. Morris, R. V. Aguilar, N. Koirala, S. Oh, and N. P. Armitage, Phys. Rev. Lett. 115, 217602 (2015). [31] L. Wu, M. Salehi, N. Koirala, J. Moon, S. Oh, and N. P. Armitage, Science 354, 1124 (2016). [32] V. Dziom, A. Shuvaev, A. Pimenov, G. V. As- takhov, C. Ames, K. Bendias, J. Bottcher, G. Tkachov, E. M. Hankiewicz, C. Brune, H. Buhmann, and L. W. Molenkamp, Nat. Commun. 8, 15197 (2017). [33] Z. Zhu, J. Wang, H. Zuo, B. Fauqu´e, R. D. McDon- ald, Y. Fuseya, and K. Behnia, Nat. Commun. 8, 15297 (2017). [34] L. Fu and C. L. Kane, Phys. Rev. B 76, 045302 (2007). [35] J. C. Y. Teo, L. Fu, and C. L. Kane, Phys. Rev. B 78, 045426 (2008). 17 [36] H.-J. Zhang, C.-X. Liu, X.-L. Qi, X.-Y. Deng, X. Dai, S.- C. Zhang, and Z. Fang, Phys. Rev. B 80, 085307 (2009). [37] S. Ito, B. Feng, M. Arita, A. Takayama, R.-Y. Liu, T. Someya, W.-C. Chen, T. Iimori, H. Namatame, M. Taniguchi, C.-M. Cheng, S.-J. Tang, F. Komori, K. Kobayashi, T.-C. Chiang, and I. Matsuda, Phys. Rev. Lett. 117, 236402 (2016). [38] G. T. Noe II, H. Nojiri, J. Lee, G. L. Woods, J. L´eotin, and J. Kono, Rev. Sci. Instrum. 84, 123906 (2013). [39] G. T. Noe II, Q. Zhang, J. Lee, E. Kato, G. L. Woods, H. Nojiri, and J. Kono, Appl. Opt. 53, 5850 (2014). [40] G. T. Noe II, I. Katayama, F. Katsutani, J. J. Allred, J. A. Horowitz, D. M. Sullivan, Q. Zhang, F. Sekiguchi, G. L. Woods, M. C. Hoffmann, H. Nojiri, J. Takeda, and J. Kono, Opt. Exp. 24, 30328 (2016). [41] I. Katayama, H. Kawakami, T. Hagiwara, Y. Arashida, Y. Minami, L.-W. Nien, O. S. Handegard, T. Nagao, M. Kitajima, and J. Takeda, Phys. Rev. B 98, 214302 (2018). [42] Y. Ikebe, T. Morimoto, R. Masutomi, T. Okamoto, H. Aoki, and R. Shimano, Phys. Rev. Lett. 104, 256802 (2010). [43] R. Shimano, G. Yumoto, J. Y. Yoo, R. Matsunaga, S. Tanabe, H. Hibino, T. Morimoto, and H. Aoki, Nat. Commun. 4, 1841 (2013). [44] K. N. Okada, Y. Takahashi, M. Mogi, R. Yoshimi, A. Tsukazaki, K. S. Takahashi, N. Ogawa, M. Kawasaki, and Y. Tokura, Nat. Commun. 7, 12245 (2016). [45] X. Wang, D. J. Hilton, L. Ren, D. M. Mittleman, [47] T. Arikawa, X. Wang, D. J. Hilton, J. L. Reno, W. Pan, and J. Kono, Phys. Rev. B 84, 241307(R) (2011). [48] T. Arikawa, X. Wang, A. A. Belyanin, and J. Kono, Opt. Exp. 20, 19484 (2012). [49] Q. Zhang, T. Arikawa, E. Kato, J. L. Reno, W. Pan, J. D. Watson, M. J. Manfra, M. A. Zudov, M. Tokman, M. Erukhimova, A. Belyanin, and J. Kono, Phys. Rev. Lett. 113, 047601 (2014). [50] Q. Zhang, M. Lou, X. Li, J. L. Reno, W. Pan, J. D. Watson, M. J. Manfra, and J. Kono, Nat. Phys. 12, 1005 (2016). [51] X. Li, M. Bamba, Q. Zhang, S. Fallahi, G. C. Gard- ner, W. Gao, M. Lou, K. Yoshioka, M. J. Manfra, and J. Kono, Nat. Photon. 12, 324 (2018). [52] X. Li, M. Bamba, N. Yuan, Q. Zhang, Y. Zhao, M. Xiang, K. Xu, Z. Jin, W. Ren, G. Ma, S. Cao, D. Turchinovich, and J. Kono, Science 361, 794 (2018). [53] I. Katayama, H. Sakaibara, and J. Takeda, Jpn. J. Appl. Phys. 50, 102701 (2011). [54] Y. Minami, Y. Hayashi, J. Takeda, and I. Katayama, Appl. Phys. Lett. 103, 051103 (2013). [55] Y. Minami, K. Horiuchi, K. Masuda, J. Takeda, and I. Katayama, Appl. Phys. Lett. 107, 171104 (2015). [56] G. Mead, I. Katayama, J. Takeda, and G. A. Blake, Rev. Sci. Instr. 90, 053107 (2019). [57] P. J. de Visser, J. Levallois, M. K. Tran, J.-M. Poumirol, I. O. Nedoliuk, J. Teyssier, C. Uher, D. van der Marel, and A. B. Kuzmenko, Phys. Rev. Lett. 117, 017402 (2016). J. Kono, and J. L. Reno, Opt. Lett. 32, 1845 (2007). [58] Z. Zhu, B. Fauqu´e, Y. Fuseya, and K. Behnia, Phys. Rev. [46] X. Wang, A. A. Belyanin, S. A. Crooker, D. M. Mittle- B 84, 115137 (2011). man, and J. Kono, Nat. Phys. 6, 126 (2010). [59] H. Aoki, Rep. Prog. Phys. 50, 655 (1987). 

1506.01947

4

1506

Theory of Two-Dimensional Spatially Indirect Equilibrium Exciton Condensates

We present a theory of bilayer two-dimensional electron systems that host a spatially indirect exciton condensate when in thermal equilibrium. Equilibrium bilayer exciton condensates (BXCs) are expected to form when two nearby semiconductor layers are electrically isolated, and when the conduction band of one layer is brought close to degeneracy with the valence band of a nearby layer by varying bias or gate voltages. BXCs are characterized by spontaneous inter-layer phase coherence and counterflow superfluidity. The bilayer system we consider is composed of two transition metal dichalcogenide monolayers separated and surrounded by hexagonal boron nitride. We use mean-field-theory and a bosonic weakly interacting exciton model to explore the BXC phase diagram, and time-dependent mean-field theory to address condensate collective mode spectra and quantum fluctuations. We find that a phase transition occurs between states containing one and two condensate components as the layer separation and the exciton density are varied, and derive simple approximate expressions for the exciton-exciton interaction strength which we show can be measured capacitively.

cond-mat.mes-hall

cond-mat

Theory of Two-Dimensional Spatially Indirect Equilibrium Exciton Condensates Feng-Cheng Wu†,1 Fei Xue†,1 and A.H. MacDonald1 1Department of Physics, The University of Texas at Austin, Austin, TX 78712, USA (†These authors contributed equally to this work.) (Dated: September 2, 2018) We present a theory of bilayer two-dimensional electron systems that host a spatially indirect exciton condensate when in thermal equilibrium. Equilibrium bilayer exciton condensates (BXCs) are expected to form when two nearby semiconductor layers are electrically isolated, and when the conduction band of one layer is brought close to degeneracy with the valence band of a nearby layer by varying bias or gate voltages. BXCs are characterized by spontaneous inter-layer phase coherence and counterflow superfluidity. The bilayer system we consider is composed of two tran- sition metal dichalcogenide monolayers separated and surrounded by hexagonal boron nitride. We use mean-field-theory and a bosonic weakly interacting exciton model to explore the BXC phase diagram, and time-dependent mean-field theory to address condensate collective mode spectra and quantum fluctuations. We find that a phase transition occurs between states containing one and two condensate components as the layer separation and the exciton density are varied, and derive simple approximate expressions for the exciton-exciton interaction strength which we show can be measured capacitively. PACS numbers: 71.35.-y, 73.21.-b I. INTRODUCTION Recent advances in the study of two-dimensional van der Waals materials1 have opened up new horizons in condensed matter physics by allowing familiar proper- ties, including those of metals, superconductors, gap- less semiconductors, semiconductors, and insulators, to be combined in new ways simply by designing stacks of atomically thick layers. In this article we consider condensation of spatially indirect excitons in the case of a two-dimensional semiconductor bilayer formed by two group-VI transition metal dichalcogenides (TMD) that are separated and surrounded by an insulator, for ex- ample hexagonal boron nitride (hBN). The TMDs are in their 2H structure monolayer form. Two-dimensional material stacks of this type are promising hosts for exci- ton condensation, both because they host strongly bound excitons,2 -- 8 and because of recent progress in realizing flexible high quality TMD heterostructures.9 -- 14 In van der Waals heterostructures it is possible15,16 to tune the positions of the Fermi levels in individual layers over wide ranges while maintaining overall charge neutrality, either by applying a gate voltage between sur- rounding electrodes or a bias voltage between the semi- conductor layers. When the indirect band gap between the conduction band of one layer and the valence band of the other layer is reduced to less than the indirect ex- citon binding energy, charge will be transferred between layers in equilibrium. At low densities, the transferred charges form spatially indirect excitons, and these are expected17 -- 24 to form Bose condensates. The bilayer ex- citon condensate (BXC) state has spontaneous interlayer phase coherence and supports dissipationless counterflow supercurrents25,26 that could enable the design of low- dissipation electronic devices.27 Exciton condensates in TMD heterostructures are sim- ilar to atomic spinor Bose-Einstein condensates because of the presence of both spin and valley degrees of free- dom. The spin-valley coupling of conduction band elec- trons and valence band holes that are specific to TMD heterostructures28 enriches the excitonic physics. In this paper we study the interplay between the exciton conden- sation and spin and valley internal degrees of freedom to construct an exciton condensate zero temperature phase diagram as a function of effective layer separation d and exciton chemical potential µ, or equivalently exciton den- sity. We demonstrate that there are two distinct conden- sate phases with different number of condensate flavors, as shown in Fig. 1. Our paper is organized as follows. In Sec. II, we ex- plain how we model the heterostructure, and present the mean-field phase diagram implied by Hartree-Fock the- ory. In Sec. III, we derive an effective boson model that incorporates exciton-exciton interaction effects and can be used to describe excitons in the low density limit. The difference between the strengths of the repulsive interac- tions between excitons with the same internal label and between excitons with different internal labels changes sign as the layer separation increases. This change drives the transition from phase-II, a phase with two conden- sate flavors present, to phase-I, a phase with only one condensate flavor. Both phases spontaneously break the symmetry of the model Hamiltonian, and the symme- try breaking pattern of each phase is analyzed. In this section we also explain how capacitance measurements can be used to study the exciton phase diagram experi- mentally and to extract the value of the exciton-exciton interaction strength within each phase. In Sec. IV, we use a time-dependent Hartree-Fock theory to study the stability of phase-I against small fluctuations, and to cal- culate the collective mode spectra of these exciton con- densates. Finally in Sec. V, we present a brief summary, 7 1 0 2 n a J 7 2 ] l l a h - s e m . t a m - d n o c [ 4 v 7 4 9 1 0 . 6 0 5 1 : v i X r a 2 conduction bands with labels c = 1, 2, 3, 4 correspond- ing to spin and valley. We assume that exciton bind- ing energies and densities are small enough to justify a parabolic band approximation for all band extrema. Our mean-field ansatz allows up to two types of excitons to be present; for examples pairs formed from holes in band v = 1 and electrons, selected by spin-splitting, in band c = 1, can condense, along with pairs formed from holes in band v = 2 and electrons in band c = 2. Although the unpaired conduction bands c = 3, 4 are only slightly higher in energy, this pairing ansatz is fully self-consistent at low exciton density, because of the substantial exciton binding energy. Our pairing ansatz is also justified by an interacting boson model, described in Sec. III, which allows for the most general possible pairing scenario. The ansatz leads to the mean-field Hamiltonian: (cid:48)(cid:88) (cid:88) (cid:126)k(vc) † (a c(cid:126)k † , a v(cid:126)k (cid:0)2k2 2me (cid:126)k,c=3,4 HM F = + )(ζ(cid:126)k + ξ(vc) (cid:126)k σz − ∆(vc) (cid:126)k σx) µ(cid:1)a − 1 2 † c(cid:126)k ac(cid:126)k, (cid:19) (cid:18)ac(cid:126)k av(cid:126)k (1) where the prime in the first summation restricts the pair index (vc) to (11) and (22) contributions. a† and a are fermionic creation and annihilation operators. The ki- netic term ζ(cid:126)k = 2k2(1/(4me) − 1/(4mh)) accounts for the difference between conduction and valence band ef- fective masses, me and mh, and σz,x are Pauli matrices. The dressed energy difference between conduction and valence bands, ξ(vc) , and the coherence induced effective interlayer tunneling amplitude, ∆(vc) , are defined as: (cid:126)k ξ(vc) (cid:126)k = (cid:126)k (cid:88) µ − 1 A − 1 2 (cid:126)k(cid:48) † c(cid:126)k(cid:48)av(cid:126)k(cid:48)(cid:105), U ((cid:126)k − (cid:126)k(cid:48))(cid:104)a 2k2 4m (cid:88) (cid:126)k(cid:48) † V ((cid:126)k − (cid:126)k(cid:48))(cid:104)a c(cid:126)k(cid:48)ac(cid:126)k(cid:48)(cid:105), (2) where (cid:104)...(cid:105) is the expectation value in the mean-field ground state, II. MEAN-FIELD PHASE DIAGRAM We consider two monolayer TMD semiconductors sep- arated and surrounded by hBN (Fig. 1). Many of the points we make apply with minor modification, how- ever, to any bilayer two-dimensional semiconductor sys- tem. Monolayer TMDs are direct-gap semiconductors with band extrema located at valleys K and K(cid:48). Because these TMD layers lack inversion symmetry, spin degener- acy in the TMD bands is lifted by spin-orbit interactions. Because of differences between the orbital character of conduction and valence band states,28 it turns out that spin splitting is large at the valence band maxima and small at the conduction band minima. As illustrated in Fig. 1, we therefore retain in our theory the two valley- degenerate valence bands labeled by v = 1, 2, and four (1 − ξ(vc) (cid:126)k /E(vc) (cid:126)k ), /(2E(vc) (cid:126)k ), (3) 1 2 † (cid:104)a c(cid:126)k † (cid:104)a c(cid:126)k E(vc) ac(cid:126)k(cid:105) = (cid:113) av(cid:126)k(cid:105) = ∆(vc) ξ(vc)2 (cid:126)k = (cid:126)k (cid:126)k In Eq. (2), m = memh/(me +mh) is the reduced mass, and A is the area of the system. The paramter µ is: µ = µ − 4πe2nd/
, n = 1 A † (cid:104)a c(cid:126)k ac(cid:126)k(cid:105), (4) + ∆(vc)2 . (cid:126)k (cid:88) (cid:88) (cid:126)k c=1,2 where µ is the chemical potential for excitons, and n is the total charge density transferred between layers. FIG. 1: (Color online) Transferred charge density and ex- citon condensate phase as a function of effective layer sep- aration d and chemical potential µ. The solid orange line marks the second-order phase transition from the phase in which there is no charge transfer between the bilayers i.e. the phase with no excitons present, to the bilayer exciton conden- sate (BXC) phase. The blue dashed line separates phase-II in which two condensate flavors are present from phase-I in which the ground state has a single condensate flavor. The two red dotted lines are contours at the density values na∗2 B = B (cid:38) 0.1 the exciton-condensate ground 0.01 and 0.1. (For na∗2 state is expected to be superseded by an electron-hole plasma state. See text for a more complete discussion.) The ar- rows mark the values of d examined in Fig. 2(a) and 2(b), and the crosses(×) indicate the parameter values examined in Fig. 2(c) and 2(d). The left inset is a schematic experi- mental setup for BXC studies in which the spatially indirect gap is tuned by an interlayer bias potential, and the right in- set illustrates the bilayer band structure in the absence of the bias potential Vb. discuss issues related to experiments, and comment on the relationship between our work and previous studies. ∆(vc) (cid:126)k = 1 A 𝑑/𝑎𝐵∗𝜇/Ry∗III𝑛𝑎𝐵∗21,31K2K'2,4𝐸𝑔MoTe2𝐷MoS2𝑉𝑏0.010.13 TABLE I: Parameter values for different combinations of monolayer layer 2H-TMDs. m0 is the electron bare mass.
= 5 for hBN. Electrons reside in MoS2, and holes in other TMDs. The listed energy gaps Eg apply in the absence of a gate voltage. me/m0[32] mh/m0[32] a∗ MoS2/MoTe2 MoS2/WSe2 MoS2/WTe2 0.47 0.47 0.47 0.62 0.36 0.32 B(A) Ry∗(meV) Eg(eV)[29] 9.89 12.97 13.89 145 111 104 1.1 1.4 0.8 (cid:126)k (cid:126)k (cid:126)k (cid:126)k = ∆(22) (cid:54)= 0 and I, only one type of exciton condenses (e.g. ∆(11) ∆(22) = 0). In phase-II, excitons associated with both valence bands condense and have equal population (e.g. (cid:54)= 0). Both phases are allowed by Eq. (2). ∆(11) We obtain the phase diagram in Fig. 1 by comparing the total energy of phase-I and II as a function of (d, µ). Below a critical layer separation dc ≈ 0.25a∗ B, phase-II always has a lower energy, as illustrated in Fig. 2(a). Above dc, a transition from phase-I to phase-II occurs as the chemical potential µ increases(Fig. 2(b)). Typical quasiparticle energy bands in phase-II and I are depicted in Fig. 2(c) and (d), and show that the system is an excitonic insulator with a charge gap. In our mean-field theory, condensation of one type or the other always occurs at T = 0 when excitons are present. It is well known however that at high electron and hole densities a first-order Mott transi- tion occurs33 -- 37 from the gapped exciton condensate phase to an ungapped electron-hole plasma state. The electron-hole plasma state is preferred energetically be- cause it can achieve better correlations between like- charge particles, reducing the probability that they are close together, while maintaining good correlations be- tween oppositely-charged particles. The density at which the Mott transition occurs is most reliably estimated via a non-perturbative approaches.34 No estimate is cur- rently available for the TMD case, for which the valley degeneracy and the small spin-splitting in the conduction band will tend to favor plasma states over exciton con- densate states. Based on existing estimates34 we can con- B ∼ 0.3 clude that the Mott transition density is below na∗2 B ∼ 1. Cor- as d/a∗ rections to mean-field theory which go in the direction of favoring plasma states can be partially captured by accounting for screening of the electron-hole interaction which becomes stronger as exciton sizes increase and ex- citons correspondingly become more polarizable. The re- sults reported here are intended to be reliable only in the low exciton density limit. B → 0 and below na∗2 B ∼ 0.05 for d/a∗ III. INTERACTING BOSON MODEL To understand the phase diagram more deeply, we em- ploy a boson Hamiltonian designed to describe weakly- interacting excitons in the low density limit. Our strategy B = 0.1 (a) and d/a∗ (Color online) (a)-(b) Energy difference (δE ≡ FIG. 2: EII − EI) per area between phase-I and II states as a func- tion µ at d/a∗ B = 0.5 (b). (c)-(d) Typi- cal quasiparticle energy bands in phase-II (c), and in phase- I (d) with (blue) and without (black) interlayer coherence. (d/a∗ B, µ/Ry∗) is (0.1,-2.5) in (c) and (0.5,-1.1) in (d), corre- sponding to the two crosses(×) in Fig. 1. These results were calculated with me = mh. Equations (2), (3) and (4) form a set of mean-field equa- tions that can be solved self-consistently. Note that the (11) and (22) pairing channels are coupled through the dependence of µ on the total transferred density n. The exciton chemical potential can be tuned electri- cally by applying a bias potential Vb between the electri- cally isolated layers: µ = Vb − Eg where Eg is the spa- tially indirect band gap between the conduction band of the electron layer and the valence band of the hole layer. The band gap Eg can be adjusted to a conveniently small value by choosing two-dimensional materials with favor- able band alignments29,30. V ((cid:126)q) = 2πe2/(
q) and U ((cid:126)q) = V ((cid:126)q)e−qd are the Coulomb interaction potentials within and between lay- ers. The forms of Coulomb potentials are determined by solving the Poisson equation for our schematic ex-
⊥
(cid:107), where
⊥ and
(cid:107) perimental setup(Fig. 1).
= are hBN dielectric constants perpendicular and parallel to the z-axis, is the effective dielectric constant due to insulator layer(hBN) between electron and hole layers. d = D(cid:112)
⊥/
(cid:107), where D is the geometric layer separa- √ tion between electron and hole layers, is the effective layer separation and slightly larger than D.31 Below we express lengths and energies in terms of B =
2/(me2), and Ry∗ = B). Typical values for different material combi- the characteristic scales a∗ e2/(2
a∗ nations are listed in Table I. The indirect exciton binding energy Eb determines the value for µ at which excitons first appear. When µ < −Eb no excitons are present. In this state each layer is electrically neutral and there is no interlayer coherence. Eq. (2) has nontrivial (n, ∆(cid:126)k (cid:54)= 0) solutions only for µ > −Eb. We find two distinct types of BXC phase. In phase- 10120.0202101221012KK'121234122,413(c)(d)𝐸/Ry∗KK'2101210.50δ𝐸/𝐴[Ry∗/𝑎𝐵∗2]𝑑/𝑎𝐵∗=0.1𝑑/𝑎𝐵∗=0.5(a)(b)𝜇/Ry∗𝜇/Ry∗4 pendix A. We focus here on their zero-momentum limits gH = gH (0) and gX = gX (0, 0), which are more easily inter- preted and capture much of the exciton-exciton interac- tion physics. For the case in which the exciton conden- sate is populated by a single flavor we find that for low exciton densities µ = −Eb + gn. (6) X − g(2) where g = gH + gX is the total exciton-exciton in- teraction, as expected from the mean-field theory for weakly interacting bosons. This behavior is illustrated in Fig. 3(a). We have verified that the interaction pa- rameter obtained by examining the dependence of µ on n in the fermion mean-field theory agrees with the an- alytic expression in App. A, as illustrated in Fig. 3(b) which plots g as a function of layer separation d. We find that gH = 4πe2d/
and that gX = g(1) X , where g(1) X and g(2) X are both positive and originate from inter and intra layer fermionic exchange interactions respectively. The binding energy of isolated excitons is due micro- scopically to attractive inter layer exchange interactions. When excitons overlap and interact with each other, co- herence between layers is reduced weakening inter layer exchange, but strengthening intra layer exchange. This explains the signs of the two contributions to gX . The overall sign of gX is positive at d = 0 because the loss of interlayer exchange energy when excitons overlap is greater than the gain in intralayer exchange energy. In Fig. 3(c), we show that gX becomes negative beyond a critical layer separation dc ∼ 0.25a∗ B. It turns out that X and g(2) although both g(1) X increase with layer separation d, the rate of increase of g(1) X is smaller than for g(2) X . The difference in behavior can be traced to the exponential decrease in the momentum space inter-layer Coulomb in- teraction with layer separation d as shown in Eq. (A19) and (A20). To find the ground state in the realistic multi-flavor case, we assume that all excitons condense into (cid:126)Q = 0 states and introduce the following matrix: (cid:32)(cid:104)B(11)(cid:105) (cid:104)B(12)(cid:105) (cid:104)B(13)(cid:105) (cid:104)B(14)(cid:105) (cid:33) (cid:104)B(21)(cid:105) (cid:104)B(22)(cid:105) (cid:104)B(23)(cid:105) (cid:104)B(24)(cid:105) . (7) F = 1√ A Neglecting the small spin-orbit splitting of conduction band states, the total energy per area can be written in a compact form, (cid:104)HB(cid:105) A (TrT )2 + gH 2 TrT 2, gX 2 = −(Eb + µ)TrT + (8) where T = F†F. In Eq. (8) TrT is the total density of excitons,summed over all flavors,and TrT 2−(TrT )2 mea- sures the flavor polarization of the exciton condensate. This energy functional is invariant under the following transformation: F (cid:55)→ eiθU2FU† 4 . (9) FIG. 3: (Color online)(a)Chemical potential µ as a function of density n obtained from Hartree-Fock calculations in which a single exciton flavor is condensed. The line is a linear fit of the numerical data to Eq. (6). (b)Exciton-exciton interaction strength extracted from the self-consistent Hartree-Fock equa- tion solutions (blue dashed line), and exciton-exciton inter- action strength calculated from the interacting boson model (red solid line). The black dashed line is the Hartree contri- bution gH to the total interaction strength. (c)The blue and red lines separate the interlayer (g(1) X ) ex- change contributions to the total exciton-exciton interaction strength. X ) and intralayer (g(2) to obtain the boson Hamiltonian is to construct a La- grangian based on a al wavefunction which parametrizes a family of states with electron-hole coherence. The Berry phase part of the Lagrangian has the same form as that in the field-theory functional integral representation of a standard interacting boson model.38 Appealing to this property, we promote variational parameters in the wavefunction to bosonic operators. The details of the derivation are presented in Appendices A and B. The boson Hamiltonian is: − Eb − µ)B HB = (cid:88) (cid:48)(cid:88)(cid:8)gH ( (cid:126)Q14)B 2Q2 2M ( + 1 2A + gX ( (cid:126)Q13, (cid:126)Q14)B † (v(cid:48)c(cid:48)) (cid:126)Q2 B † (vc) (cid:126)Q1 † (v(cid:48)c(cid:48)) (cid:126)Q2 B † (vc) (cid:126)Q1 † (vc) (cid:126)Q B(vc) (cid:126)Q B(v(cid:48)c(cid:48)) (cid:126)Q3 B(vc) (cid:126)Q4 B(v(cid:48)c) (cid:126)Q3 B(vc(cid:48)) (cid:126)Q4 (cid:9), (5) where B(vc) (cid:126)Q is a bosonic operator for an exciton with a hole in valence band v, an electron in conduction band c, and total momentum (cid:126)Q. (cid:126)Qab is the momen- tum transfer (cid:126)Qa − (cid:126)Qb. For the TMD system, there are 8 possibilities for the composite index (vc). The quadratic term in Eq. (5) accounts for exciton kinetic energy (M = me + mh) and chemical potential. The quartic terms describe exction-exciton interactions. The prime on the quartic term summation enforces momen- tum conservation (cid:126)Q1 + (cid:126)Q2 = (cid:126)Q3 + (cid:126)Q4. The two types of exciton interaction arise from the fermionic Hartree and exchange interactions respectively. In the exchange interaction, two excitons swap con- stituent electrons or holes. Analytic expressions for the coupling strength gH ((cid:126)q) and gX ((cid:126)q (cid:48), (cid:126)q) are given in Ap- 0.0.10.20.30.40.5202530354000.0050.012.582.562.54𝑛𝑎𝐵∗2𝜇/Ry∗𝑦=−2.58+5.80𝑥(a)𝑔𝑋(1)𝑔𝑋(2)d/𝑎𝐵∗𝑔𝑋(1,2)/(Ry∗𝑎𝐵∗2)(c)00.10.20.30.40.502468(b)𝑔/(Ry∗𝑎𝐵∗2)𝑔=𝜕𝜇/𝜕𝑛𝑛→0𝑔=𝑔𝐻+𝑔𝑋(1)-𝑔𝑋(2)𝑔𝐻=4𝜋𝑒2𝑑/𝜖d/𝑎𝐵∗d/𝑎𝐵∗= 0.1Here eiθ captures the U(1) symmetry which originates from separate charge conservation in the individual lay- ers. U2 and U4 are respectively 2 × 2 and 4 × 4 spe- cial unitary matrices, which capture the SU(2) symme- try of the valence bands and the SU(4) symmetry present in the conduction bands when their spin-splitting is ne- glected. The overall symmetry group of the system is U(1)×SU(2)×SU(4). When the conduction band spin- orbit splitting is included, the higher energy conduction band states in each valley are not occupied and the sym- metry group is reduced to U(1)×SU(2)×SU(2), corre- sponding to separate charge conservation and rotations in both conduction and valence band valley spaces. F acquires a nonzero value in the ground state only if µ > −Eb, . By minimizing the energy functional, we verify that the sign of gX determines the position of a phase boundary between two different classes of exciton condensate which we refer to as phase-I and II. When gX < 0, phase-I is energetically favorable and a repre- sentative realization of the ground state is, (cid:32) FI = √ nI 1 0 0 0 0 0 0 0 , (10) where nI = (µ + Eb)/(gH + gX ) is the exciton density. FI is invariant under the transformation: e−iφ 0 V† 0 eiφ 0 0 e−iφ = FI, (cid:32) (cid:32) (cid:33) FI (11) where V3 is a 3 × 3 unitary matrix. Therefore, phase-I spontaneously breaks the U(1)×SU(2)×SU(4) symmetry down to U(1)×U(3) symmetry. When gX > 0 phase-II is realized. Energy minimiza- tion shows that a representative realization of the ground state in phase-II is, (cid:32) FII =(cid:112)nII/2 (cid:33) 1 0 0 0 0 1 0 0 (cid:33) (cid:33) 3 5 FIG. 4: (Color online) (a) Four representative degenerate ground states in phase-I. For illustration purpose, only two conduction bands are shown. (The other two are slightly higher in energy because of spin-orbit splitting and do not participate in the ground state manifold.) The dashed black oval highlights the two bands with spontaneous phase co- herence in phase-I. Charge is transferred from the valence band partner to the conduction band partner. In the upper panels (↑↑) and (↓↓), coherence is established between like spins and the ground state is not spin-polarized. In the lower panels (↑↓) and (↓↑), charge is transferred between opposite spins and the ground state is spin-polarized, with a finite spin-polarization that is proportional to the transfered charge density. Because spin and valley are locked by spin-orbit cou- pling, spin-polarized states are stabilized by an infinitesimal Zeeman field. (b) In the presence of an infinitesimal Zee- man field that favors spin up, the spin-polarized state (↓↑) is selected as the unique ground state in phase-I. (c) Two rep- resentative degenerate ground states in phase-II. Both states have zero spin-polarization, and remain degenerate in an in- finitesimal Zeeman field as shown in (d). , (12) distinct set of broken symmetries compared to the usual spin rotational symmetry breaking. where nII = (µ + Eb)/(gH + gX /2) is the total exciton density in phase-II. FII is invariant under the transfor- mation: (cid:32)U† (cid:33) U2FII 2 0 0 V† 2 = FII, (13) where V2 is a 2 × 2 unitary matrix. Phase-II sponta- neously breaks the U(1)×SU(2)×SU(4) symmetry down to SU(2)×U(2) symmetry. A similar analysis can be ap- plied to identify the symmetry breaking pattern when the spin splitting of the conduction bands is consid- ered. In phase-I, an application of an infinitesimal ex- ternal Zeeman field lifts both conduction and valence band valley degeneracies, and selects a unique conden- sate ground state with a finite spin-polarization, as illus- trated in Fig. 4. Phase-I therefore satisfies the definition of a ferromagnet, defined as a system with a finite spin- polarization in an infinitesimal Zeeman field, but has a Based on this mean-field calculation, we conclude that although the system has 8 types of excitons in total, only one or two flavors condense in the ground state. The number of condensed flavors is in general limited by the number of distinct valence or conduction bands, which ever is smaller in number. Although the boson model correctly captures the phase transition position as a func- tion of d, it is important to emphasize that it is valid only in the low exciton density limit. For this reason, it fails to accurately predict the µ dependence of the phase boundary. In addition it fails to capture the tendency toward weaker electron-hole pairing at high exciton den- sities, which eventually leads to an electron-hole quantum liquid state with no interlayer coherence. The relationship between the exciton density n, the exciton chemical potential µ and the coupling strength g makes it possible to extract the value of g from ca- pacitance measurement. The differential capacitance per ↑↑↓↓↑↑↓↓↑↑↓↓↑↑↓↓↑↑↓↓↑↑↓↓↑↑↓↓↑↑↓↓B(↑ ↑)(↓↓)(↑↓ )(↓↑ )(↑ ↑)(↓↓)(↑↓ )(↓↑ )(a)(b)↑↑↓↓↑↑↓↓↑↑↓↓↑↑↓↓B(c)(d)area for the heterostructure is: C = e2 ∂n ∂µ . (14) The Hartree coupling strength gH can be identified as the inverse of the geometric capacitance: Cgeo = e2/gH =
/(4πd). (15) 6 Therefore, capacitance measurement provides a simple way to determine the value of gX in the low-exciton den- sity limit: e2(C−1 − C−1 geo) = gX < 0, phase-I 1 2 gX > 0, phase-II . (16) (cid:40) The sign of gX helps to distinguish phase-I and II. IV. FLUCTUATIONS AND STABILITY The bilayer exciton condensate is a state with spon- taneously broken continuous symmetries, and therefore hosts low-energy collective fluctuations. Theoretical studies of fluctuation properties are of interest in part because they can reveal mean-field state39 instabilities. The collective modes can be studied using the interact- ing boson model, which is described in detail in Ap- pendix C. The interacting boson model admits analytic solutions for collective modes associated with exciton density, phase and flavor fluctuations in both phase-I and II. However, it is valid only in the low exciton density limit. Here, we study another approach that can be ap- plied to any exciton density. This approach is based on the following variational wave function which captures exciton density and phase fluctuations in phase-I: (cid:89) (cid:104)Z(cid:126)k + (cid:88) Φ(cid:105) = z(cid:126)k( (cid:126)Q)γ † ((cid:126)k+ (cid:126)Q),1 γ(cid:126)k,0 (17) (cid:105)XC(cid:105). (cid:126)k (cid:126)Q where XC(cid:105) is the phase-I ground state. γ are respectively quasiparticle creation operators for occupied and empty quasiparticle states in XC(cid:105) associated with the ground state condensate and are defined as follows: and γ † (cid:126)k,0 † (cid:126)k,1 † γ (cid:126)k,0 † γ (cid:126)k,1 † =u(cid:126)ka c(cid:126)k † =v(cid:126)ka c(cid:126)k † + v(cid:126)ka v(cid:126)k † − u(cid:126)ka v(cid:126)k , , (18) where u(cid:126)k and v(cid:126)k are parameters determined by self- consistent Hatree-Fock equations (2), (cid:114) 1 2 (cid:114) 1 2 u(cid:126)k = Z(cid:126)k is a normalization factor, v(cid:126)k = (1 − ξ(cid:126)k/E(cid:126)k), (cid:115) 1 −(cid:88) Z(cid:126)k = (cid:126)Q (1 + ξ(cid:126)k/E(cid:126)k). (19) z(cid:126)k( (cid:126)Q)2, (20) (Color online)(a)-(b) Eigenvalues of K(±) for d/a∗ FIG. 5: B = 0.5 as a function of momentum (cid:126)Q at na∗2 B = 0.008 (a) and na∗2 B = 0.08 (b). Red solid and blue dashed lines connect the lowest eigenvalues of the density fluctuation matrix K(+) and the phase fluctuation matrix K(−) respectively. Small red points label higher eigenvalues of K(+). The gray lines mark the lower edge of the quasiparticle electron-hole con- tinua [Min(E(cid:126)k + E(cid:126)k+ (cid:126)Q)]. (c)-(d) Collective mode spectra at na∗2 B = 0.5. Larger points are used for the lowest energy excitations. In (c), the red and black lines are respectively a linear fit at small Q and a quadratic fit at large Q to the collective mode energy. For the results presented here, we assumed me = mh. B = 0.008 (c) and na∗2 B = 0.08 (d) for d/a∗ and z(cid:126)k( (cid:126)Q) are complex parameters. To study fluctuation dynamics, we construct the La- grangian: L = (cid:104)Φi∂t − HΦ(cid:105) ≈ B − δE(2), (21) where B is the harmonic Berry phase, and δE(2) is the harmonic energy variation40: {E(cid:126)k,(cid:126)p( (cid:126)Q)z∗ ( (cid:126)Q)z(cid:126)p( (cid:126)Q) (cid:88) δE(2) = (cid:126)k (cid:126)Q,(cid:126)k,(cid:126)p + 1 2 Γ(cid:126)k,(cid:126)p( (cid:126)Q)[z(cid:126)k( (cid:126)Q)z(cid:126)p(− (cid:126)Q) + z∗ (cid:126)k ( (cid:126)Q)z∗ (cid:126)p(− (cid:126)Q)]}. (22) Explicit forms for the matrices E and Γ are given in App. D. To decouple ± (cid:126)Q contributions in Eq. (22), we perform a change of variables, defining z(cid:126)k( (cid:126)Q) = x(cid:126)k( (cid:126)Q) + iy(cid:126)k( (cid:126)Q) (− (cid:126)Q) = x(cid:126)k( (cid:126)Q) − iy(cid:126)k( (cid:126)Q). z∗ −(cid:126)k (23) (24) Note that x(cid:126)k( (cid:126)Q) and y(cid:126)k( (cid:126)Q) are complex numbers, and that there is a redundancy, x−(cid:126)k(− (cid:126)Q) = x∗ y−(cid:126)k(− (cid:126)Q) = y∗ ( (cid:126)Q), (cid:126)k ( (cid:126)Q). (cid:126)k (25) In terms of the x and y fields, the Berry phase B and Q𝑎𝐵∗density eh continuumphaseQ𝑎𝐵∗densityeh continuumphaseQ𝑎𝐵∗𝜔/Ry∗Linear FitQuadratic FitQ𝑎𝐵∗𝜔/Ry∗(a)(b)(c)(d)()()()()(cid:126)k ( (cid:126)Q), (26) (cid:1)( (cid:126)Q) are real 1 2 energy variation δE(2) are, B = (cid:126)Q,(cid:126)k y∗∂tx + y∂tx∗ − x∗∂ty − x∂ty∗(cid:17) (cid:16) (cid:88) (cid:16) (cid:17) (cid:88) (cid:88) ( (cid:126)Q) =(cid:0)E(cid:126)k,(cid:126)p±Γ(cid:126)k,−(cid:126)p x(cid:126)p + y∗ K(+) (cid:126)k,(cid:126)p K(−) x∗ (cid:126)k (cid:126)k,(cid:126)p y(cid:126)p (cid:126)Q (cid:126)k,(cid:126)p (cid:126)k ( (cid:126)Q). δE(2) = (cid:126)k,(cid:126)p The kernel matrices K(±) and symmetric. The x and y fields in δE(2) can be identi- fied with exciton density and phase respectively, and the Berry phase contribution to the action captures the con- jugate relationship between these fluctuation variables. Stability of the mean-field ground states against small fluctuations requires that the matrices K(±) are nonnega- tive. We have verified that this condition is satisfied out to large d by explicit numerical calculations like those summarized in Fig. 5(a) and (b). At (cid:126)Q = 0, the matrix K(−) always has a zero-energy eigenvalue since, K(−) (cid:126)k,(cid:126)p (0) ∆(cid:126)p E(cid:126)p = 0, (27) (cid:88) (cid:126)p which follows from the fact that ground state energy is independent of global interlayer phase. For low exciton density n (Fig. 5(a)), the lowest eigen- values of K(+) and K(−) have similar behavior and are separated from the continuum. This is expected since K(+) and K(−) are identical in the limit n → 0. Fig. 5(b) demonstrates that the lowest eigenvalues of K(+) are close to the particle-hole continuum when the exciton den- sity becomes large; the interacting boson model discussed above fails qualitatively in this limit. The Euler-Lagrange equation for the Lagrangian in Eq. (21) gives rise to the equation of motion, (cid:17) (cid:17) ∂tx(cid:126)k( (cid:126)Q) = −(cid:16)K(−) (cid:16)K(+) t y(cid:126)k( (cid:126)Q) = −(cid:104)(cid:0)K(+)K(−)(cid:1) ∂ty(cid:126)k( (cid:126)Q) = + (cid:126)k,(cid:126)p (cid:126)k,(cid:126)p x(cid:126)p y(cid:126)p ∂2 ( (cid:126)Q), ( (cid:126)Q). (cid:105) (cid:126)k,(cid:126)py(cid:126)p ( (cid:126)Q). (28) (29) which leads to It follows that the energy of the collective mode is given by the square root of the lowest eigenvalues of the matrix product K(+)K(−), which is plotted in Fig. 5(c) and (d). The lowest energy collective mode is the gapless Gold- stone mode of the exciton condensate. For low exciton density n (Fig. 5(c)), the Goldstone mode has linear dis- persion at small (cid:126)Q, becoming quadratic at large (cid:126)Q. This agrees with the Goldstone mode behavior predicted by the weakly interacting boson model(Eq. (5)). For large n (Fig. 5(d)), the Goldstone mode deviates from quadratic behavior at large (cid:126)Q. The failure of the weakly interact- ing boson model in the high density limit originates from the internal structure of the excitons. When the typical distance between excitons is comparable to exciton size, excitations must be described in terms of the underlying conduction and valence band fermion states.18,41,42 7 V. SUMMARY AND DISCUSSION By combining Hartree-Fock theory and an interacting boson model, we have shown that spatially indirect ex- citon condensates in group-VI TMD bilayers have two distinct phases. We have also studied the dynamics of exciton condensate density and phase fluctuations and calculated the associated collective mode spectra. The topic of exciton condensation in semiconductors has a long history and our work is related to some earlier studies. For example, Berman et al.43 studied exciton condensation in bilayers formed from gapped graphene, although the possibility of two distinct condensate phases was not considered. The phase transition between the two condensate phases as a function of layer separation was studied previously44,45 for the case of quantum well bilayer excitons, and further explored in a very recent publication.46 The TMD layers considered in this pa- per are distinguished from semiconductor quantum well systems by exciton binding energies that are an order of magnitude larger, and by spin-valley coupling which leads to two-fold degenerate valence bands and approx- imately four-fold degenerate conduction bands. Com- pared to Refs.45 and 46, we used a completely differ- ent approach to derive an interacting boson model. Our approach is physically transparent, and is based on a variatonal wavefunctions defined by parameters whose quantum fluctuations are characterized by using a La- grangian formalism. The bosonic nature of excitons is automatically taken into account in the Lagrangian, and there is no need to calculate combinatorial factors aris- ing from the indistinguishability of bosonic particles. Our approach provides a simple yet systematic way to model the exction-exciton interaction. We have also discussed a fermionic Hartree-Fock approach from which the exciton- exciton interaction strengths can be extracted with sim- ilar results, and proposed that the interaction strengths can be experimentally determined by performing capac- itance measurement. Because the hBN dielectric barrier in the systems of interest, must be thick enough to make interlayer tun- neling weak, Fig. 1 implies that phase-I with a single condensate flavor is more likely to be realized in experi- ment than phase-II. Phase-I breaks the invariance of the system Hamiltonian under separate valley rotations in conduction and valence bands, and is ferromagnetic in the sense that infinitesimal Zeeman coupling leads to a spin-polarization that is proportional to the exciton den- sity. In spit of their large gaps, band edge states in TMDs have relatively large Berry curvatures28. In monolayer TMDs momentum space Berry curvatures lead8 to un- usual exictonic spectra in which hydrogenic degeneracies are lifted. Although band Berry curvatures should be less important in spatially indirect exciton systems because weaker binding implies that the exciton states are formed within a smaller region of momentum space. In terms of its influence on quasiparticle bands, exciton condensation has the effect of preventing gaps between conduction and valence band states from closing. Since the host semicon- ductor materials are topologically trivial, and since tran- sitions between trivial and non-trivial states can occur continuously only when the quasiparticle charged excita- tion energy vanishes, we argue that exciton condensation will not result in interaction-driven topologically nontriv- ial states in our system. The critical temperature of spatially indirect exciton condensate is the Berezinskii-Kosterlitz-Thouless transi- tion temperature, given at low exciton densities by the weakly-interacting boson expression kBTBKT ≈ 1.3 2n M = 2.6(na∗2 B ) Ry∗, m M (30) B ∼ 0.3 as d/a∗ where M is the electron-hole pair total mass and m is the reduced mass. In the low exciton density limit kBTBKT scales linearly with exciton density.37,47 For the MoS2/hBN/MoTe2 heterostructure and exciton densities B ≈ 0.01 and TBKT is about n in the 1012cm−2 range, na∗2 10K. The maximum possible transition temperature is closely related to the critical density at which the Mott transition to an electron-hole plasma occurs, and this in- creases as d/a∗ B decreases. Using the variational Monte Carlo estimate of DePalo et al.34 the critical value of B → 0. From this we conclude that na∗2 kBTBKT cannot exceed around 300K. Adjustment of ex- citon density by external bias voltage can be employed to search for the highest transition temperature and to study the Mott transition to an ungapped electron-hole plasma that is expected at high exciton densities. The most interesting regime is likely to be the case of very small layer separations of which current leakage driven through the tunnel barrier by an interlayer bias potential might be appreciable, requiring the bilayer to be treated as a non-equilibrium system. The exciton condensate should be experimentally re- alizable in TMD bilayers provided that samples with sufficiently weak disorder can be achieved. The photo- luminescence line width W of an individual monolayer TMD is a particularly useful characterization of sample quality for this purpose. W is currently dominated48 by the position-dependence of exciton energies. Therefore, the narrower the line width W , the weaker the disor- der, and the better the sample quality. We expect bi- layer exciton condensation to occur only in samples in which W < kBTBKT, since the excitons will otherwise simply localize near positions where they have minimum energy. Note that the inhomogeneous broadening W of spatially indirect excitons will not be experimentally ac- cessible since the corresponding transitions are optically inactive when the interlayer tunneling is negligible, but that it should be similar to the broadening of the readily measurable direct exciton energies. It should therefore be possible to judge on the basis of optical characteri- zation when samples have achieved sufficient quality to study spatially indirect exciton condensate physics. 8 VI. ACKNOWLEDGMENT This work was supported by the SRC and NIST under the Nanoelectronic Research Initiative (NRI) and SWAN, and by the Welch Foundation under Grant No. F1473. Appendix A: Interacting boson model for excitons in the low density limit In the low density limit, excitons can be approximated as interacting bosons. We take a BCS like variational wave function to describe excitons, Ψ(cid:105) = Ω† = 1 (cid:88) N exp(Ω†)vac(cid:105), † λV Ca CaV , V,C (A1) where C and V respectively denote a conduction and valence band state, and include internal indices such as vac(cid:105) is the spin and valley and also momentum label. † V vac(cid:105) = aCvac(cid:105) = 0. Ω† vacuum state defined by a operator creates particle-hole excitations on top of the vacuum. N is a normalization factor so that (cid:104)ΨΨ(cid:105) = 1. λV C is a set of complex variational parameters, which are small when the exciton density is low. The density matrix with respect to Ψ(cid:105) is ραβ = αaβΨ(cid:105), where α and β can be conduction or valence (cid:104)Ψa† states. We expand the density matrix to fourth order in λV C, ρV V (cid:48) ≈ δV V (cid:48) + (−λλ† + λλ†λλ†)V V (cid:48), ρCC(cid:48) ≈ (λ†λ − λ†λλ†λ)CC(cid:48), ρV C ≈ (λ − λλ†λ)V C, (A2) where λ is understood to be a matrix and λ† is its Her- mitian conjugate. We introduce another matrix Λ so that ρCC(cid:48) has a quadratic form without fourth-order correction, λ = Λ + ΛΛ†Λ. 1 2 (A3) FIG. 6: Feynman diagrams for the Hartree exciton-exciton interaction processes. A double line with arrow represents an exciton state, and a single solid (dashed) line with ar- row depicts an electron (hole) in the exciton. Wavy lines are interaction V ((cid:126)q) or U ((cid:126)q). (a) and (b) are the intralayer con- tributions, while (c) is the interlayer contribution. (a), (b) and (c) correspond to the three terms in H(4) H (Eq. (A9)) and also the three terms in gH ((cid:126)q) (Eq. (A17)). (a)(b)(c)𝑈( 𝑞)𝑉( 𝑞)𝑉( 𝑞)Expanding ρ up to fourth order of Λ, we have that ρV V (cid:48) ≈ δV V (cid:48) − (ΛΛ†)V V (cid:48), ρCC(cid:48) ≈ (Λ†Λ)CC(cid:48), ρV C ≈ (Λ − 1 2 ΛΛ†Λ)V C. (A4) The number of excitons is (cid:104)Nex(cid:105) = (cid:80) C ρCC ≈ TrΛ†Λ. Therefore, we verify that Λ acts as the small parameter in the limit of low (cid:104)Nex(cid:105). An important property of the density matrix is that ρ2 − ρ = O(Λ5), (A5) which indicates that Ψ(cid:105) can be approximated as a Slater determinant up to fourth order in Λ.40 Ψ(cid:105) parametrizes a family of states with electron-hole coherence, and also represents low-energy states in the low-exicton density limit. We choose to construct an effective interacting boson model using this variational wavefunction approach rather than a commonly used auxiliary field approach because of the necessity of consis- tently accounting for both exchange and Hartree mean- fields in spatially-indirect exciton systems. (See addi- tional discussion below.) To study low-energy dynamics, we construct a Lagrangian based on Ψ(cid:105) L = (cid:104)Ψi∂t − HΨ(cid:105) = B − H, (A6) and again expand everything to Λ4. This Lagrangian provides an effective field theory for excitons. The Berry phase has the following form, (cid:0)Tr[Λ†∂tΛ] − Tr[(∂tΛ†)Λ](cid:1), B = (cid:104)Ψi∂tΨ(cid:105) ≈ i 2 9 which does not have fourth order corrections. To calculate the energy functional H, we take advan- tage of the Slater determinant approximation40 to Ψ(cid:105) (Eq. (A5)) and obtain that H = (cid:104)ΨHΨ(cid:105) ≈ H(2) + H(4) where H(2) is quadratic in Λ, and H(4) H,X is quartic in Λ with subscript H and X representing Hartree and ex- change contributions. The explicit forms are below. H + H(4) X , (A8) ΛV1C2 , H(4) H = † H(2) = (εC − εV )Λ CV ΛV C † − WV1C1C2V2 Λ C1V2 WC1C2C3C4 (Λ†Λ)C1C4 (Λ†Λ)C2C3 1 2 WV1V2V3V4 (ΛΛ†)V1V4(ΛΛ†)V2V3 1 + 2 − WV1C1C2V2 (ΛΛ†)V1V2(Λ†Λ)C1C2, 1 2 1 + 2 − 1 2 − 1 2 (ΛΛ†Λ)V1C2 WV1C1C2V2Λ WV1C1C2V2(Λ†ΛΛ†)C1V2ΛV1C2 WC1C2C3C4(Λ†Λ)C1C3(Λ†Λ)C2C4 WV1V2V3V4 (ΛΛ†)V1V3(ΛΛ†)V2V4. H(4) X = † C1V2 (A9) Here εC and εV are conduction and valence state energy including self-energy effects. The interaction kernel W has the form (A7) W(n1(cid:126)k1)(n2(cid:126)k2)(n3(cid:126)k3)(n4(cid:126)k4) 1 A δn1n4δn2n3 δ(cid:126)k1+(cid:126)k2,(cid:126)k3+(cid:126)k4 = Wn1n2((cid:126)k1 − (cid:126)k4), (A10) where the momentum dependence is now explicit, and n denotes internal indices. A is the area of the system. Wn1n2((cid:126)q) is the intralayer interaction V ((cid:126)q) if both n1 and n2 represent conduction or valence bands, and the interlayer interaction U (q) otherwise. We now write H(2) in a more concrete form (cid:104)(cid:16)2((cid:126)k + (cid:126)Q)2 (cid:105) 2me Λ(v,(cid:126)k(cid:48))(c,(cid:126)k(cid:48)+ (cid:126)Q). (cid:17) 2k2 2mh + − µ δ(cid:126)k(cid:126)k(cid:48) H(2) =Λ∗ (v,(cid:126)k)(c,(cid:126)k+ (cid:126)Q) − 1 A U ((cid:126)k − (cid:126)k(cid:48)) (A11) Here v and c denote different valence and conduction bands. We approximate εC and εV by parabolic bands, and assume different valence (conduction) bands have the same hole (electron) mass mh (me). In the case of TMDs, these are reasonable approximations, and v and c respec- tively take two and four different values. H(2) can be reduced into a diagonal from by doing the following decomposition, 1√ A Λ(v,(cid:126)k)(c,(cid:126)k+ (cid:126)Q) = f ((cid:126)k + xh (cid:126)Q)B(vc) (cid:126)Q (A12) FIG. 7: Feynman diagrams for the exchange exciton-exciton interaction processes. The convention is the same as in Fig. 6. (a) and (b) are the interlayer contributions, which correspond to the first two terms in H(4) X (Eq. (A9)) and the two terms in g(1) X (Eq. (A19)). (c) and (d) are the intralayer contributions, which correspond to the lase two terms in H(4) X (Eq. (A9)) and the two terms in g(2) X (Eq. (A20)). (a)(b)(c)(d)𝑈( 𝑞)𝑈( 𝑞)𝑉( 𝑞)𝑉( 𝑞)where B(vc) (cid:126)Q is a complex field that depends on momen- tum (cid:126)Q but not on (cid:126)k. xh = mh/M , where the total mass M = me + mh. For notation convenience, we also intro- duce xe = me/M = 1 − xh. f ((cid:126)k) is the 1s wavefunction for a single exciton, δ(cid:126)k(cid:126)k(cid:48) − 1 A U ((cid:126)k − (cid:126)k(cid:48)) f ((cid:126)k(cid:48)) = −Ebf ((cid:126)k), (A13) where the reduced mass m = memh/M , and Eb is the binding energy for 1s state. The normalization condition is that (cid:104)2k2 2m (cid:105) f ((cid:126)k)2 = 1. (A14) (cid:88) (cid:126)k 1 A Here we have chosen f ((cid:126)k) to be real. In Eq. (A12), f ((cid:126)k + xh (cid:126)Q) is the wavefunction for an exciton with center-of- mass momentum (cid:126)Q. By substituting Eq. (A12) into Eq. (A7) and (A9), we obtain that B ≈ i 2 H = (vc) (cid:126)Q (cid:0)B∗ (cid:88)(cid:0)2Q2 (cid:48)(cid:88)(cid:8)gH ( (cid:126)Q14)B∗ 2M (cid:1), ∂tB(vc) (cid:126)Q − B(vc) (cid:126)Q∂tB∗ (vc) (cid:126)Q (A15) − Eb − µ(cid:1)B∗ (vc) (cid:126)Q B(vc) (cid:126)Q 1 2A + +gX ( (cid:126)Q13, (cid:126)Q14)B∗ (vc) (cid:126)Q1 B∗ (v(cid:48)c(cid:48)) (cid:126)Q2 (vc) (cid:126)Q1 B∗ (v(cid:48)c(cid:48)) (cid:126)Q2 B(v(cid:48)c(cid:48)) (cid:126)Q3 B(vc) (cid:126)Q4 B(v(cid:48)c) (cid:126)Q3 B(vc(cid:48)) (cid:126)Q4 (cid:9). (A16) The Berry phase B has the same form as that in the field-theory functional integral representation of a stan- dard interacting boson model, which suggests that the Lagrangian L = B − H is a functional field integral rep- resentation of a boson model38. By replacing the complex numbers (B∗, B) with bosonic creation and annihilation operators (B†, B) in the energy functional H, we arrive at the boson model (5) in the main text. H , and has the following ana- gH ((cid:126)q) is derived from H(4) lytic expression gH ((cid:126)q) = V ((cid:126)q)(cid:2)F (xe(cid:126)q)2 + F (xh(cid:126)q)2(cid:3) − 2U ((cid:126)q)F (xe(cid:126)q)F (xh(cid:126)q), (A17) where (cid:88) (cid:126)k 1 A F ((cid:126)q) = f ((cid:126)k)f ((cid:126)k + (cid:126)q). (A18) For zero-momentum transfer, gH (0) = 2(V (0) − U (0)) = 4πe2d/
. gX ( (cid:126)Q13, (cid:126)Q14) is derived from H(4) decomposed into two parts gX = g(1) X , and can be further X − g(2) X arises X . g(1) from the loss of interlayer exchange energy as more ex- citons condense, while −g(2) X comes from the gain of in- tralayer exchange energy as electron or hole density in- creases. The explicit forms of g(1,2) are below. X = = (cid:126)k,(cid:126)k(cid:48) U ((cid:126)k − (cid:126)k(cid:48))f ((cid:126)k(cid:48))f ((cid:126)k + (cid:126)Qe) g(1) X ( (cid:126)Q13, (cid:126)Q14) 1 A2 (cid:88) ×(cid:2)f ((cid:126)k + (cid:126)Qh)f ((cid:126)k + (cid:126)Qe + (cid:126)Qh) + f ((cid:126)k − (cid:126)Qh)f ((cid:126)k + (cid:126)Qe − (cid:126)Qh)(cid:3) (cid:88) (cid:1)f ((cid:126)k)f ((cid:126)k + (cid:126)Qe) ×(cid:2)f ((cid:126)k + (cid:126)Qh)f ((cid:126)k + (cid:126)Qe + (cid:126)Qh) + f ((cid:126)k − (cid:126)Qh)f ((cid:126)k + (cid:126)Qe − (cid:126)Qh)(cid:3), (cid:0)Eb + 2k2 2m 1 A (cid:126)k 10 (A19) (A20) (cid:88) (cid:8) (cid:126)k,(cid:126)k(cid:48) 1 A2 g(2) X ( (cid:126)Q13, (cid:126)Q14) = V ((cid:126)k − (cid:126)k(cid:48) + (cid:126)Qe)f ((cid:126)k)f ((cid:126)k(cid:48))f ((cid:126)k + (cid:126)Qh)f ((cid:126)k(cid:48) + (cid:126)Qh)+ V ((cid:126)k − (cid:126)k(cid:48) + (cid:126)Qh)f ((cid:126)k)f ((cid:126)k(cid:48))f ((cid:126)k + (cid:126)Qe)f ((cid:126)k(cid:48) + (cid:126)Qe)(cid:9), where (cid:126)Qe = xe (cid:126)Q13 and (cid:126)Qh = xh (cid:126)Q14. According to Eq. (A19), g(1) X can also be interpreted as the increase of kinetic energy due to the increase of the exciton density. The diagrammatic representation of different processes for the exciton-exciton interaction is shown in Fig. 6 and 7. The momentum dependence of the interaction strength gH,X is illustrated in Fig. 8. Both gH and gX have a strong momentum dependence, which indicates that extion-exciton interaction are long-ranged instead of short-ranged. The problem of exciton-exction interaction has been studied using many different methods in the literature45,46,49 -- 52. Here we used an alternative ap- proach based on a variational wave function combined with the Lagrangian formalism. The functional field in- tegral representation of the boson model in Eq. (5) is exactly given by the Lagrangian L with Berry phase B in Eq. (A15) and energy function H in Eq. (A16)38. This property establishes the equivalence between the Lagrangian and the boson model. In fact, all results obtained using the boson model can be equivalently de- rived from the Lagrangian. The phase transition between phase-I and II can be determined by the minimization of the energy functional (A16) with the ansatz that (B∗, B) is spatially uniform and time independent. The collective mode studied in Appendix C can be obtained using the Euler-Lagrange equation of the Lagrangian. There are other approaches in constructing an effective theory of bosonic excitation in an interacting fermion sys- tem. One example is the standard Hubbard-Stratonovich transformation. There is however a certain arbitrari- ness in decomposing electron-electron interactions into 11 To ensure that at each momentum (cid:126)k the two occupied states are orthogonal to each other, we require that s1((cid:126)k)χ((cid:126)k) + f ((cid:126)k)2(cid:88) b∗ (v1c)b(v2c) = 0. (B3) c Similar to the procedure in App. A, we expand the energy functional (cid:104)Ψ0HΨ0(cid:105) to fourth order in b(vc), and then replace complex numbers (b∗ (vc), b(vc)) by op- erators (B A, which gives rise to the same boson model in Eq. (A16) except that all momenta are restricted to be zero. √ , B(vc)(cid:126)0)/ † (vc)(cid:126)0 Appendix C: Collective modes in the interacting boson model The collective modes can be calculated analytically us- ing the interacting boson model. The strategy is to shift a bosonic operator by its mean-field value, B(vc) (cid:126)Q = (cid:104)B(vc) (cid:126)Q(cid:105)δ (cid:126)Q,0 + b(vc) (cid:126)Q, (C1) where b(vc) (cid:126)Q is also a bosonic operator that describes fluctuations around the mean field state. The bosonic Hamiltonian in Eq. (5) is then expanded to second or- der in (b†, b). The first order terms vanish as the en- ergy functional in Eq. (8) is minimized in the mean-field state. The quadratic terms can be diagonalized using the Bogoliubov transformation, giving rise to the collective mode spectra. We present the dispersion and degeneracy of collective modes in phase-I and II without giving the details of the derivation. In phase-I, there are five gapless modes, and three gapped modes. Given the mean-field value in Eq. (10), (b†, b)(vc) with different composite index (vc) are decou- pled. The (vc) = (11) mode is gapless, which is the Goldstone mode due to the spontaneously broken U(1) symmetry, i.e. the separate charge conservation within each layer. Its dispersion is: (cid:114)(cid:104)2Q2 (cid:105)(cid:104)2Q2 (cid:105) , ω(11) = g±( (cid:126)Q) = gX ( (cid:126)Q, 0) + gX (0, (cid:126)Q) − gX (0, 0) ± gX ( (cid:126)Q, (cid:126)Q), + (2gH + g+)( (cid:126)Q)nI + g−( (cid:126)Q)nI 2M 2M (C2) where g−( (cid:126)Q) has a Q2 dependence at small (cid:126)Q. Therefore, the (11) mode has a linear dispersion in Q → 0 limit. The (vc) = (12), (13) and (14) modes are degenerate, with a gapless dispersion: +(cid:2)gx(0, (cid:126)Q) − gx(0, 0)(cid:3)nI, 2Q2 2M ω(12) = (C3) which has a quadratic Q dependence in Q → 0 limit. Similaryly, the (vc) = (21) mode is also gapless and quadratic at small (cid:126)Q: +(cid:2)gx( (cid:126)Q, 0) − gx(0, 0)(cid:3)nI. (C4) ω(21) = 2Q2 2M FIG. 8: mentum. d/a∗ assumed. gH ((cid:126)q), gX ((cid:126)q, 0) and gX ((cid:126)q, (cid:126)q) as a function of mo- B is 0.1 in (a) and 0.5 in (b). me = mh is Hartree or Fock channels in the Hubbard-Stratonovich scheme38. The HS approach is not appropriate here be- cause a proper description of the SIEXC requires that Hartree and Fock interactions to be treated on the same footing. Appendix B: Interacting boson model for excitons with zero center-of-mass momentum We consider another variational wave function with all excitons condense into zero center-of-mass momentum state. (cid:89) (cid:2)s1((cid:126)k)a Ψ0(cid:105) = (cid:2)χ((cid:126)k)a (cid:126)k † v1(cid:126)k † v1(cid:126)k + f ((cid:126)k) † b(v1c)a c(cid:126)k † + s2((cid:126)k)a v2(cid:126)k + f ((cid:126)k) † b(v2c)a c(cid:126)k (B1) (cid:88) (cid:88) c c (cid:3)× (cid:3)0(cid:105), where v1 = 1 and v2 = 2, representing two valence bands, and c represents different conduction bands. f ((cid:126)k) is the exciton wave function in Eq. (A13). b(vc) are com- plex parameters, which are independent of momentum (cid:126)k. In 0(cid:105), both valence and conduction bands are empty, ac(cid:126)k0(cid:105) = av(cid:126)k0(cid:105) = 0. Facotrs s1,2((cid:126)k) and χ((cid:126)k) are de- termined by normalization and orthogonality conditions. By normalization conditions, we have that (cid:115) 1 − f ((cid:126)k)2(cid:88) (cid:115) 1 − χ((cid:126)k)2 − f ((cid:126)k)2(cid:88) b(v1c)2, c s1((cid:126)k) = s2((cid:126)k) = (B2) b(v2c)2. c 01234567830301234567850510𝑔𝐻( 𝑞)𝑔𝑋( 𝑞,0)𝑔𝑋( 𝑞, 𝑞)𝑔𝐻( 𝑞)𝑔𝑋( 𝑞,0)𝑔𝑋( 𝑞, 𝑞)𝑞𝑎𝐵∗𝑞𝑎𝐵∗𝑔/(Ry∗𝑎𝐵∗2)𝑔/(Ry∗𝑎𝐵∗2)(a) 𝑑/𝑎𝐵∗=0.1(b)𝑑/𝑎𝐵∗=0.5The (vc) = (22), (23) and (24) modes are degenerate and gapped: ω(22) = 2Q2 2M − gx(0, 0)nI, (C5) which shows that phase-I is stable against small fluctua- tions provided that gx(0, 0) is negative. In phase-II, the mean-field values are given in Eq. (12) and all eight collective modes are gapless. The (vc) = (11) and (22) fluctuations are coupled, and give rise to two non-degenerate gappless modes: (cid:105) nII 2 2M (cid:114)(cid:104)2Q2 (cid:114)(cid:104)2Q2 (cid:114)(cid:104)2Q2 (cid:114)(cid:104)2Q2 2M 2M 2M ω(11),(22) = × ω(cid:48) (11),(22) = × + (4gH + g+)( (cid:126)Q) (cid:105) (cid:105) (cid:105) , , + g−( (cid:126)Q) + g+( (cid:126)Q) + g−( (cid:126)Q) nII 2 nII 2 nII 2 12 decoupled, and have degenerate collective modes: +(cid:2)gx(0, (cid:126)Q) − gx(0, 0)(cid:3) nII ω(13) = 2Q2 2M , (C7) 2 which has a quadratic dispersion at small Q. TABLE II: Classification of collective modes in phase-I and II. Ngapped is the number of gapped collective modes. N1 and N2 are respectively the number of gapless collective modes with linear and quadratic dispersion. NBSG is the number of the broken symmetry generators. Ngapped N1 N2 NBSG phase-I phase-II 3 0 1 4 4 4 9 12 (C6) In Table II, we list N1 and N2, respectively the num- ber of gapless collective modes with linear and quadratic dispersion, and NBSG, the number of the broken sym- metry generators for each phase. There is a relationship among these three numbers in both phase-I and II, both of which have a linear dispersion at small Q. and lead to two modes that are degenerate with ω(cid:48) The (vc) = (12) and (21) fluctuations are also coupled, (11),(22). The (vc) = (13), (14), (23) and (24) fluctuations are N1 + 2N2 = NBSG. (C8) This relationship is typical for broken symmetry states in system without Lorentz invariance53. Appendix D: explicit expressions for E(cid:126)k,(cid:126)p( (cid:126)Q) and Γ(cid:126)k,(cid:126)p( (cid:126)Q) Below are explicit expressions for E(cid:126)k,(cid:126)p( (cid:126)Q) and Γ(cid:126)k,(cid:126)p( (cid:126)Q), which appear in the energy variation δE(2) in Eq. (22). E(cid:126)k,(cid:126)p( (cid:126)Q) = δ(cid:126)k,(cid:126)p(ζ(cid:126)k+ (cid:126)Q − ζ(cid:126)k + E(cid:126)k + E(cid:126)k+ (cid:126)Q) (cid:2)V ( (cid:126)Q) − V ((cid:126)k − (cid:126)p)(cid:3)(u(cid:126)ku(cid:126)pv(cid:126)k+ (cid:126)Qv(cid:126)p+ (cid:126)Q + v(cid:126)kv(cid:126)pu(cid:126)k+ (cid:126)Qu(cid:126)p+ (cid:126)Q) 1 + A − 1 A − 1 A 1 A − 1 A 1 A + Γ(cid:126)k,(cid:126)p( (cid:126)Q) = U ( (cid:126)Q)(v(cid:126)ku(cid:126)pu(cid:126)k+ (cid:126)Qv(cid:126)p+ (cid:126)Q + u(cid:126)kv(cid:126)pv(cid:126)k+ (cid:126)Qu(cid:126)p+ (cid:126)Q) U ((cid:126)k − (cid:126)p)(u(cid:126)ku(cid:126)pu(cid:126)k+ (cid:126)Qu(cid:126)p+ (cid:126)Q + v(cid:126)kv(cid:126)pv(cid:126)k+ (cid:126)Qv(cid:126)p+ (cid:126)Q), (cid:2)V ( (cid:126)Q) − V ((cid:126)k + (cid:126)Q − (cid:126)p)(cid:3)(u(cid:126)ku(cid:126)pv(cid:126)k+ (cid:126)Qv(cid:126)p− (cid:126)Q + v(cid:126)kv(cid:126)pu(cid:126)k+ (cid:126)Qu(cid:126)p− (cid:126)Q) U ( (cid:126)Q)(v(cid:126)ku(cid:126)pu(cid:126)k+ (cid:126)Qv(cid:126)p− (cid:126)Q + u(cid:126)kv(cid:126)pv(cid:126)k+ (cid:126)Qu(cid:126)p− (cid:126)Q) U ((cid:126)k + (cid:126)Q − (cid:126)p)(v(cid:126)ku(cid:126)pv(cid:126)k+ (cid:126)Qu(cid:126)p− (cid:126)Q + u(cid:126)kv(cid:126)pu(cid:126)k+ (cid:126)Qv(cid:126)p− (cid:126)Q), (D1) where u(cid:126)k and v(cid:126)k are defined in Eq. (19), and V ( (cid:126)Q) and U ( (cid:126)Q) are respectively intralayer and interlayer Coulomb interactions. 1 A. K. Geim and I. V. Grigorieva, Nature 499, 419 (2013). 2 T. C. Berkelbach, M. S. Hybertsen, and D. R. Reichman, Phys. Rev. B 88, 045318 (2013). 3 D. Y. Qiu, F. H. da Jornada, and S. G. Louie, Phys. Rev. Lett. 111, 216805 (2013). 4 C.-J. Zhang, H.-N. Wang, W.-M. Chan, C. Manolatou, and F. Rana, Phys. Rev. B 89, 205436 (2014). 5 A. Chernikov, T. C. Berkelbach, H. M. Hill, A. Rigosi, Y. Li, O. B. Aslan, D. R. Reichman, M. S. Hybertsen, and T. F. Heinz, Phys. Rev. Lett. 113, 076802 (2014). 6 Z. Ye, T. Cao, K. OBrien, H. Zhu, X. Yin, Y. Wang, S. G. Louie, and X. Zhang, Nature 513, 214 (2014). 7 K. He, N. Kumar, L. Zhao, Z. Wang, K. F. Mak, H. Zhao, and J. Shan, Phys. Rev. Lett. 113, 026803 (2014). 8 F. Wu, F. Qu, and A. H. MacDonald, Phys. Rev. B 91, 075310 (2015). 9 H. Fang, C. Battaglia, C. Carraro, S. Nemsak, B. Ozdol, J. S. Kang, H. A. Bechtel, S. B. Desai, F. Kronast, A. A. Unal, G. Conti, C. Conlon, G. K. Palsson, M. C. Martin, A. M. Minor, C. S. Fadley, E. Yablonovitch, R. Maboudian, and A. Javey, PNAS 111, 6198 (2014). 10 X. Hong, J. Kim, S.-F. Shi, Y. Zhang, C. Jin, Y. Sun, S. Tongay, J. Wu, Y. Zhang, and F. Wang, Nat. Nan- otechnol. 9, 682 (2014). 11 R. Cheng, D. Li, H. Zhou, C. Wang, A. Yin, S. Jiang, Y. Liu, Y. Chen, Y. Huang, and X. Duan, Nano Lett. 14, 5590 (2014). 12 Y. Gong et al., Nat. Mater. 13, 1135 (2014). 13 Y. Yu, S. Hu, L. Su, L. Huang, Y. Liu, Z. Jin, A. A. Purezky, D. B. Geohegan, K. W. Kim, Y. Zhang, and L. Cao, Nano Lett. 15, 486 (2014). 14 P. Rivera, J. R. Schaibley, A. M. Jones, J. S. Ross, S. Wu, G. Aivazian, P. Klement, K. Seyler, G. Clark, N. J. Ghimire, J. Yan, D. G. Mandrus, W. Yao, and X. Xu, Nat. Commun. 6, 6242 (2015). 15 K. Lee, B. Fallahazad, J. Xue, D. C. Dillen, K. Kim, T. Taniguchi, K. Watanabe, and E. Tutuc, Science 345, 58 (2014). 16 K. Kim, S. Larentis, B. Fallahazad, K. Lee, J. Xue, D. C. Dillen, C. M. Corbet, and E. Tutuc, ACS Nano 9, 4527 (2015). 17 Y. E. Lozovik and V. I. Yudson, Sov. Phys. JETP 44, 389 (1976). 18 X. Zhu, P. B. Littlewood, M. S. Hybertsen, and T. M. Rice, Phys. Rev. Lett. 74, 1633 (1995). 19 Y. E. Lozovik and O. L. Berman, Journal of Experimental and Theoretical Physics Letters 64, 573 (1996). 20 J. Fern´andez-Rossier, C. Tejedor, L. Munoz, and L. Vina, Phys. Rev. B 54, 11582 (1996). 21 L. Vina, J. Phys. Condens. Matter 11, 5929 (1999). 22 M. Combescot, O. Betbeder-Matibet, and F. Dubin, Phys. Rep. 463, 215 (2008). 23 M. Combescot and D. W. Snoke, Phys. Rev. B 78, 144303 (2008). 24 A. A. High, J. R. Leonard, A. T. Hammack, M. M. Fogler, L. V. Butov, A. V. Kavokin, K. L. Campman, and A. C. Gossard, Nature(London) 483, 584 (2012). 25 J. P. Eisenstein and A. H. MacDonald, Nature 432, 691 (2004). 26 J.-J. Su and A. H. MacDonald, Nature Phys. 4, 799 (2008). 27 S. K. Banerjee, L. F. Register, E. Tutuc, D. Reddy, and 13 A. H. MacDonald, IEEE Electron Device Letters 30, 158 (2009). 28 D. Xiao, G.-B. Liu, W. Feng, X. Xu, and W. Yao, Phys. Rev. Lett. 108, 196802 (2012). 29 C. Gong, H. Zhang, W. Wang, L. Colombo, R. M. Wallace, and K. Cho, Appl. Phys. Lett. 103, 053513 (2013). 30 M.-H. Chiu, C. Zhang, H.-W. Shiu, C.-P. Chuu, C.-H. Chen, C.-Y. S. Chang, C.-H. Chen, M.-Y. Chou, C.-K. Shih, and L.-J. Li, Nat. Commun. 6, 7666 (2015). 31 Y. Cai, L. Zhang, Q. Zeng, L. Cheng, and Y. Xu, Solid State Commun. 141, 262 (2007). 32 A. Korm´anyos, G. Burkard, M. Gmitra, J. Fabian, V. Z´olyomi, N. D. Drummond, and V. Falko, 2D Mater. 2, 022001 (2015). 33 L. Liu, L. ´Swierkowski, and D. Neilson, Physica B 249- 251, 594 (1998). 34 S. De Palo, F. Rapisarda, and G. Senatore, Phys. Rev. Lett. 88, 206401 (2002). 35 V. V. Nikolaev and M. E. Portnoi, Superlattices Mi- crostruc. 43, 460 (2008). 36 K. Asano and T. Yoshioka, J. Phys. Soc. Jpn. 83, 084702 (2014). 37 M. M. Fogler, L. V. Butov, and K. S. Novoselov, Nat. Commun. 5, 4555 (2014). 38 J. W. Negele and H. Orland, Quantum Many-Particle Sys- tems (Addison-Wesley, Redwood City, California, 1988). 39 A. K. Fedorov, I. L. Kurbakov, Y. E. Shchadilova, and Y. E. Lozovik, Phys. Rev. A 90, 043616 (2014). 40 G. F. Giuliani and G. Vignale, Quantum theory of the elec- tron liquid (Cambridge University Press, 2005). 41 L. V. Keldysh and Y. V. Kopaev, Sov. Phys. Solid State 6, 2219 (1965). 42 C. Comte and P. Nozieres, J. Phys. (Paris) 43, 1069 (1982). 43 O. L. Berman, R. Y. Kezerashvili, and K. Ziegler, Phys. Rev. B 85, 035418 (2012). 44 J. Fern´andez-Rossier and C. Tejedor, Phys. Rev. Lett. 78, 4809 (1997). 45 S. B.-T. de Leon and B. Laikhtman, Phys. Rev. B 63, 125306 (2001). 46 M. Combescot, R. Combescot, M. Alloing, and F. Dubin, Phys. Rev. Lett. 114, 090401 (2015). 47 A. Filinov, N. V. Prokof'ev, and M. Bonitz, Phys. Rev. Lett. 105, 070401 (2010). 48 G. Moody, C. K. Dass, K. Hao, C.-H. Chen, L.-J. Li, A. Singh, K. Tran, G. Clark, X. Xu, G. Bergauser, E. Malic, A. Knorr, and X. Li, Nat. Commun. 6, 8315 (2015). 49 E. Hanamura, J. Phys. Soc. Jpn. 29, 50 (1970). 50 S. Okumura and T. Ogawa, Phys. Rev. B 65, 035105 (2001). 51 C. Schindler and R. Zimmermann, Phys. Rev. B 78, 045313 (2008). 52 M. Combescot, O. Betbeder-Matibet, and F. Dubin, Phys. Rep. 463, 215 (2008). 53 H. Watanabe and H. Murayama, Phys. Rev. Lett. 108, 251602 (2012). 

1005.0450

1

1005

Spectator Behavior in a Quantum Hall Antidot with Multiple Bound Modes

We theoretically study Aharonov-Bohm resonances in an antidot system with multiple bound modes in the integer quantum Hall regime, taking capacitive interactions between the modes into account. We find the spectator behavior that the resonances of some modes disappear and instead are replaced by those of other modes, due to internal charge relaxation between the modes. This behavior is a possible origin of the features of previous experimental data which remain unexplained, spectator behavior in an antidot molecule and resonances in a single antidot with three modes.

cond-mat.mes-hall

cond-mat

Spectator Behavior in a Quantum Hall Antidot with Multiple Bound Modes Department of Physics, Korea Advanced Institute of Science and Technology, Daejeon 305-701, Korea (Dated: November 17, 2018) W.-R. Lee and H.-S. Sim We theoretically study Aharonov-Bohm resonances in an antidot system with multiple bound modes in the integer quantum Hall regime, taking capacitive interactions between the modes into account. We find the spectator behavior that the resonances of some modes disappear and instead are replaced by those of other modes, due to internal charge relaxation between the modes. This behavior is a possible origin of the features of previous experimental data which remain unexplained, spectator behavior in an antidot molecule and resonances in a single antidot with three modes. PACS numbers: 73.43.-f, 73.23.Hk, 73.23.-b Electron-electron interactions play an important role in an antidot in the integer quantum Hall regime [1]. In an antidot with one or two bound modes (edge states), the number of which is determined by local filling factor νc around the antidot, the interactions cause interest- ing phenomena [2 -- 9], such as charging effects and h/2e Aharonov-Bohm (AB) effects. It is valuable to extend the phenomena to generic effects in antidots with mul- tiple modes. The extension is reminiscent of the stream of studies from a quantum dot to multiple dots [10], and useful for applying antidots to the fractional quantum Hall regime [11, 12] or to qubit implementation [13]. In Ref. Some works [14 -- 16] have been done in that direction, but require further studies. [14], an antidot molecule with νc = 4 was experimentally studied; see a simplified view with νc = 2 in Fig. 1(a). It has atomic modes X1, X2, and molecular modes Y. Under certain conditions, AB resonances with period ∆BY correspond- ing to the area enclosed by Y disappear in electron con- ductance GT through the system, while those to X1,2 were observed [Fig. 1(c)]. This finding disagrees with the noninteracting electron case, in which Y more clearly shows AB resonances than X1,2 since Y couples more strongly with extended edge channels. Such disappear- ance of AB resonances was called spectator behavior [14]. Its mechanism remains unclear despite of efforts [17]. Unexpected experimental results [15] were also found in a νc = 3 antidot with three modes [Fig. 1(b)]. The magnetic-field dependence of GT shows three peaks in one AB period ∆B, two of which have almost same peak height higher than the third [Fig. 1(d)]. And, the depen- dence of GT on the backgate voltage VBG applied to the antidot shows two alternating peak separations, i.e., the pairing of two neighboring peaks. Even more strange is that the peak pairing was not found for νc = 2 and 4. These features disagree with the noninteracting case, in which there appear three independent peaks with differ- ent height and separation within one period since each mode shows one peak per period and couples to extended edge channels differently from the others; for example, X1 couples to channels 1 ↑ and 2′ ↑, while X2 to 0 ↓ and 1′ ↓ [Fig. 1(b)]. The unexpected results may come from the interactions, however, they are different from the h/3e AB effect, a naive extension of the h/2e AB effect [2] of νc = 2, in which the three peaks have the same height. In this Letter, we theoretically study AB resonances in antidot systems with three modes in the integer quantum Hall regime, based on a capacitive interaction model. We predict the spectator behavior that the AB resonances of some modes disappear and instead are replaced by those of other modes, because of internal charge relaxation be- tween the modes. Which and how many modes show the spectator behavior depends on ratios of capacitances. Our finding provides unified understanding of the unex- pected results [14, 15] on the two different systems. Antidots with three modes. -- We consider two repre- sentative systems with three modes, a symmetric νc = 2 FIG. 1: Schematic views of (a-b) antidots and (c-d) relevant experimental data. (a) A symmetric antidot molecule with local filling factor νc = 2 and bulk filling νb ≥ 4. It has atomic modes X1, X2, and molecular mode Y. Each mode has Landau-level (0, 1, 2) and spin (↑, ↓) indexes. The solid and dashed lines represent edge states and electron tunneling, respectively. (b) A νc = 3 antidot with modes X1, X2, Y. (c) Fourier transformation of AB oscillations of conductance GT through an antidot molecule with νc = 4. From [14]. (d) Magnetic-field B and backgate-voltage VBG dependence of GT for an antidot with νc = 3. From Figs. 11 and 13 of [15]. molecule and a νc = 3 antidot [15] (Fig. 1); a νc = 4 molecule in Ref. [14] is spin-unresolved so that some of its features can be described by the νc = 2 molecule. Each mode tunnel-couples to extended edge channels with the same spin, and also to the other modes with the same spin. The two systems have one outermost mode Y and two inner modes X1 and X2. We treat X1 and X2 equally, since in the molecule they are symmetric, and in the νc = 3 antidot they have Zeeman gap (albeit exchange enhanced [18]) much smaller than Landau gap. We consider the regime of zero bias, zero temperature, and strong perpendicular magnetic field B0 ≫ ∆Bα. Here, ∆Bα is the AB period of mode α = X1, X2, Y. In the tunneling regime, we describe the total energy of the two systems by the same form (with system-dependent parameters) of the capacitive interaction model, E({δQα}) = X ξαmnαm + X Uαα′δQαδQα′ /e2, (1) αm αα′ where α, α′ = X1, X2, Y. We derive it by generalizing the case of a νc = 2 antidot [1, 7] in the same way as in mul- tiple dots [10]. It governs the ground-state transition as a function of B0 or VBG. By analyzing the transition, we predict the features (height and separation) of AB reso- nance peaks in GT . For the νc = 2 case, the model (1) successfully describes the charging effect, h/(2e) AB ef- fect and Kondo effect [7]; we here ignore Kondo effects. The first term of Eq. (1) comes from the energy ξαm and occupation nαm of single-electron orbital m of α. We will derive below that ξ satisfies ξαm = ξαm,0 + ∆ξαδB/∆Bα when the magnetic field varies from B0 by δB (≪ B0). Here ∆ξα is the single-particle level spac- ing of α, and ξαm = ξαm,0 at B0. This dependence of ξαm on δB leads to the fact that mode α shows one AB resonance per period ∆Bα in the noninteracting limit. The second term of Eq. (1) shows capacitive interac- tions Uαα′ ≡ e2(C −1)αα′ /2 between the excess charges δQα accumulated in mode α. δQα depends on δB as δQα = eNα − QG α + e δB/∆Bα. (2) The total charge eNα of α is compensated by gate charge QG α (∝ VBG) tuned by VBG. Nα(= Pm nαm) varies by an integer due to the discreteness of electron charge e < 0. We explain the dependence of ξαm and δQα on δB. As B0 increases by δB, each orbital αm spatially shifts toward the center of its antidot to keep enclosing a given number, saying m, of magnetic flux quanta. Then, its en- ergy changes by (∆ξα + 2 Pα′ Uα′αδQα′ /eNα)δB/∆Bα. The term ∆ξαδB/∆Bα, coming from antidot potential, results in the dependence of ξαm on δB. The other term, resulting from the interactions between the orbital and δQα′, causes the dependence on δB in Eq. (2). The de- pendence on δB captures the physics of antidots. We discuss the parameters of Eq. (1). For B0 ≫ ∆Bα, it is natural to apply the constant interaction model [10] 2 that ∆ξα and Cαα′ are constant over several AB periods, and that Cαα = Cg,α+Pα′6=α Cαα′ . Cg,α is the "gate" capacitance of α due to extended edge channels and VBG. X1 and X2 have the same values of ∆Bα, ∆ξα, Uαα, UαY, and Cg,α because of the symmetry. Charge accumulation and relaxation. -- As δB in- creases, δQα continuously accumulates with rate 1/∆Bα as in Eq. (2). The accumulated charges are relaxed with resonant tunneling, resulting in the transition of the ground-state configuration (NX1 , NX2, NY). There are two kinds of single-electron relaxation. External relaxation occurs between a mode (here, X1) and ex- tended edge channels (with Fermi level ǫF ), e.g., when E(δQX1 ± e, δQX2, δQY) = E(δQX1, δQX2 , δQY) ± ǫF . This causes resonance peaks in GT . By contrast, internal relaxation occurs between modes, through tunneling or cotunneling mediated by virtual states, e.g., when E(δQX1 ±e, δQX2, δQY ∓e) = E(δQX1 , δQX2, δQY). (3) It does not cause peaks in GT . It occurs only between Y and α ∈ {X1, X2} in our case with the symmetry between X1 and X2. In general, relaxations involving more than one electron can occur. Two-electron relaxation occurs in the molecule (see below), while not in the νc = 3 antidot. The ground-state evolution of the antidots and the re- sulting AB resonances are governed by the relaxations. We study them by analyzing a charge stability dia- gram [10]. In Fig. 2(a), it is drawn for a νc = 3 antidot in (δQX+ , δQY) plane, where δQX± ≡ δQX1 ± δQX2; this two-dimensional view is possible due to the symmetry of X1 and X2. Below, we first consider the strong interac- tion regime of Uαα ≫ ∆ξα, which is analogous to the case of metallic dots, and then discuss finite ∆ξα. The internal relaxation results in the spectator behav- ior. The evolution of {δQα} follows different types of se- quences of AB resonances, depending on how many times the spectator behavior appears per ∆BX1 . For example, in a νc = 3 antidot, there are three types I, II, III [Fig. 2]. In type I of X1-Y-X2, the evolution never passes the in- ternal relaxation, and AB resonances occur sequentially by X1, Y, X2, X1, Y, X2, · · · . In type II of X1-Y-Y (III of Y-Y-Y), the evolution passes the internal relaxation once (twice) per ∆BX1, and the AB resonances by X2 (X1 and X2) disappear and are replaced by those by Y. Here, X1 or X2 shows the spectator behavior. We discuss the general features of the spectator behav- ior. Which mode shows the behavior is governed by η ≡ = , (UYY − UX1Y)δB/∆BY (UX1X1 + UX1X2 − 2UX1Y)δB/∆BX1 Cg,Y∆BY Cg,X1 ∆BX1 (4) the ratio of energy gains between δQX+ (= δQX1 +δQX2) and δQY in the internal relaxation between them; see Eq. (3). In Fig. 2(a), η equals the ratio of slopes be- tween the dash-dot relaxation line and the evolution ar- row. When η < 1, the relaxation occurs from δQY to 3 finding type J ∈ {I, II, III} in the ensemble of sequences with different initial values of {δQα} at B0 [Fig. 2(c)]. In Fig. 2(b), we plot GT (δB). We obtain it in the sequential tunneling regime using the standard master equation method [16], which is enough for demonstrating the positions and relative heights of AB peaks; here, we assumed low temperature (≪ Uαα) and the backward- reflection regime, as in Ref. [15], that mode α couples to edge channel β with coupling strength γα−β as γY−1↑ > γX1−1↑ > · · · . For each type, we describe the features of GT . In type I, each of X1, X2, and Y shows one peak per period ∆BX1. The resulting three peaks in ∆BX1 have different height, because of different γα−β's. The peak by Y is the highest, since Y is the outermost mode. In type II, two peaks among the three within ∆BX1 come from Y, and have the same height higher than the third. The separation κ∆BX1 between two con- secutive peaks by Y depends on interactions as κ = UX1Y/(2UX1Y + UYY), regardless of the initial values of {δQα} at B0. In the strong inter-mode interaction limit of Cg,X1 /CX1Y → 0, κ → 1/3. The position of the other peak by X1 or X2 depends on the initial values of {δQα}. In type III, all the three peaks within ∆BX1 come from Y, showing the same peaks. The separation be- tween them is determined by interactions as κ∆BX1 and (1−2κ)∆BX1. In the strong inter-mode interaction limit, it becomes ∆BX1 /3, showing h/(3e) AB effects, and the tot/e2, where total energy in Eq. (1) becomes E ≃ U δQ2 δQtot = Pα δQα = 3eδB/∆BX1 + · · · . This form of E, mentioned in literatures [16], cannot describe type II. So far, we have restricted to ∆ξα = 0. In the case of finite level spacing ∆ξα, the first term of Eq. (1) is absorbed into the second so that E has the same form as that of ∆ξα = 0, but with replacement (i) α , where QG Uαα → Uαα + ∆ξα/2 and (ii) QG α is obtained by Pα′ (Uαα′ +δαα′∆ξα/2) QG α′ = Pα′ Uαα′ QG α′. The replacement does not modify the dependence of δQα on δB in Eq. (2), but weakens the spectator behavior [see replacement (i)]. For example, when ∆ξα is comparable to e2/CX1X1 , type II and III are suppressed by 25 % and 100 %, respectively, for the antidot studied in Fig. 2. α → QG The above findings indicate that the result of Ref. [15], two peaks with equal height in ∆B [the upper panel of Fig. 1(d)], may be explained by type II; we do not exclude the possibility of type I that two modes among the three accidently give the two peaks with almost equal height. On the other hand, replacement (ii) affects the evolu- tion of δQα as a function of VBG. For ∆ξα = 0, the evolu- tion follows a line of slope QG X1) ≃ 0.5 in the stabil- ity diagram. When ∆ξα is finite, the slope becomes s = QG X1) ≃ 0.5[1 + (Cg,X1 + 3CX1Y)∆ξX1 /e2]/[1 + (Cg,Y + 3CX1Y)∆ξY/e2]. s can be very small for ∆ξα ≃ e2/CX1Y, Cg,Y ≫ Cg,X1 , CX1Y. The lat- ter condition can be satisfied in a νc = 3 antidot since the spatial separation between Y (X1) and extended edge channels is determined by Zeeman (Landau) splitting. Y/(2 QG Y/(2QG (color online) (a) Charge stability diagram for a FIG. 2: νc = 3 antidot. It consists of two types (A) and (B) of hexagonal cells in (δQX+, δQY) plane. Each cell represents a ground-state configuration of (NX1 , NX2 , NY) = (L, M, N ). Cell boundaries are determined by charge relaxation condi- tions such as Eq. (3). At dashed blue (solid red) bound- aries, AB resonances occur via tunneling through Y (X1 or X2), and at dash-dot green boundaries internal charge re- laxations occur between Y and X1,2. As the magnetic field increases, (δQX+, δQY) evolve along a line (solid arrow) of slope ∆BX1 /(2∆BY), while δQX− is constant within a cell and differs by charge e between (A) and (B). Depending on initial values of δQα's at given field B0, the evolution shows one of three possible sequences of AB resonances, "X1-Y-X2" (type I), "X1-Y-Y" (type II), "Y-Y-Y" (type III). Parame- ters are chosen as ∆ξα = 0, ∆BX1 = ∆BY, Cg,Y = 8Cg,X1 , CX1Y = 2Cg,X1 , CX1X2 = 8CX1Y, and δQX− = 0 for cell (A) and −e for (B). (b) Sequence of resonance peaks in GT as a function of δB for each type shown in (a). The modes giving peaks are shown. Triangles represent internal charge relaxation. (c) Probability PJ (1/η) of finding the sequences of type J = I, II, III is drawn with the parameters of (a). δQX1 or δQX2; hence Y shows the spectator behavior. For η > 1, X1 and X2 show it. On the other hand, the inter-mode interaction strength (∝ Cα′α(6=α′)) also gov- erns the behavior. As the strength increases and as η more and more deviates from 1, more sequences (with different "initial" values of {δQα} at B0) show the be- havior (type II or III); the dash-dot line in Fig. 2(a) be- comes longer so that the evolution has more chance to pass the internal relaxation. When the strength vanishes or η = 1, the spectator behavior is suppressed and only type I appears. Note that noninteracting electrons show only type I. Antidot with νc = 3. -- We discuss the spectator be- havior in a νc = 3 antidot [Fig. 1(b)]. Its geometry indi- cates ∆BX1 ≃ ∆BY and CX1X2 > CX1Y. The spatial separation between the outermost mode Y (inner modes X1,2) and the extended channels 1 ↓ is governed by Zee- man splitting energy (Landau gap) so that Cg,Y/Cg,X1 is much larger than 1. These lead to η > 1 [Eq. (4)]. Hence, as mentioned above, X1 and X2 show the spectator be- havior that type II sequence of X1-Y-Y and III of Y-Y-Y appear instead of I of X1-Y-X2. As η (> 1) and CX1,2Y increase, type II and III appear more dominantly. For a νc = 3 antidot, we obtain the probability PJ (1/η) of 4 The unexpected results of Ref. [14] can be understood by type II, provided that the inner modes X1 and X2 are almost decoupled from edge channels (i.e., only Y shows AB peaks). In this case, the period of AB peaks in type II is 2∆BY instead of ∆BY due to the spectator behavior [Fig. 3(a)]. This agrees with Fig. 1(c), since 2∆BY ≃ ∆BX1 . On the contrary, all the other types of η < 1 and those of η > 1 cannot explain Fig. 1(c); for example, type I shows peaks with ∆BY or a mixture of ∆BY and ∆BX1 , depending on the coupling of X1,2 with edge channels. These indicate that the molecule of Ref. [14] is in the regime of type II of η < 1. Conclusion. -- Electron-electron interactions give rise to the spectator behavior of AB resonances in antidots with three modes. The spectator behavior is generic, i.e., expected to appear in other quantum Hall systems with multiple modes, such as antidots and quantum dots. And it is useful for detecting interactions between edge states. To experimentally test the spectator behavior and our explanation of the experimental data [14, 15], one can monitor the modes showing resonance signals by selective injection and detection of edge channels [19]. We thank C. J. B. Ford, V. J. Goldman, and M. Kataoka for discussion, and NRF (2009-0078437). [1] H.-S. Sim, M. Kataoka, and C. J. B. Ford, Phys. Rep. 456, 127 (2008). [2] C. J. B. Ford et al., Phys. Rev. B 49, 17456 (1994); M. Kataoka et al., Phys. Rev. B 62, R4817 (2000). [3] I. J. Maasilta and V. J. Goldman, Phys. Rev. B 57 R4273 (1998). [4] M. Kataoka et al., Phys. Rev. Lett. 83, 160 (1999). [5] I. Karakurt et al., Phys. Rev. Lett. 87 146801 (2001). [6] M. Kataoka, C. J. B. Ford, M. Y. Simmons, and D. A. Ritchie, Phys. Rev. Lett. 89, 226803 (2002). [7] H.-S. Sim et al., Phys. Rev. Lett. 91, 266801 (2003); N. Y. Hwang, S.-R. E. Yang, H.-S. Sim, and H. Yi, Phys. Rev. B 70, 085322 (2004). [8] S. Ihnatsenka and I. V. Zozoulenko, Phys. Rev. B 74 201303(R) (2006). [9] M. Kato et al., Phys. Rev. Lett. 102, 086802 (2009). [10] W. G. van der Wiel et al., Rev. Mod. Phys. 75, 1 (2002). [11] I. J. Maasilta and V. J. Goldman, Phys. Rev. Lett. 84 1776 (2000). [12] D. V. Averin and J. A. Nesteroff, Phys. Rev. Lett. 99 096801 (2007). [13] D. V. Averin and V. J. Goldman, Solid State Commun. 121 25 (2001). [14] C. Gould et al., Phys. Rev. Lett. 77, 5272 (1996). [15] V. J. Goldman, J. Liu, and A. Zaslavsky, Phys. Rev. B 77, 115328 (2008). [16] S. Ihnatsenka, I. V. Zozoulenko, and G. Kirczenow, Phys. Rev. B 80, 115303 (2009). [17] Y. Takagaki, Phys. Rev. B 55, R16021 (1997). [18] W. Xu et al., J. Phys.: Condens. Matter 7, 4419 (1995). [19] M. Kataoka, C. J. B. Ford, M. Y. Simmons, and D. A. Ritchie, Phys. Rev. B 68 153305 (2003). FIG. 3: (color online) (a) Selected types of sequences of AB resonances, as a function of δB for a νc = 2 molecule. Filled and empty triangles represent the single- and two-electron internal relaxations, respectively. (b) Probability PJ (η) of finding the sequences of type J = I, II, III. Here, I (II) means the types, e.g., Ia and Ib (IIa and IIb), having two (one) Y- resonances within ∆BX1 . Parameters are chosen as ∆ξα = 0, ∆BX1 = 2∆BY, CX1Y = 10Cg,X1 , and CX1X2 = 0.5CX1 Y. The evolution line with small slope s can pass only the solid boundaries in the stability diagram [Fig. 2], showing paired peaks by X1 and X2. Or, depending on the initial value of {Qα} at B0, it can pass only the dashed lines, showing paired peaks by Y. The paired peaks agree with the lower panel of Fig. 1(d). In the cases of νc = 2 and 4 with finite ∆ξα, the slope s has a similar form to νc = 3, but the peak pairing does not appear, because the sep- aration between the outermost mode and edge channels is governed by Landau splitting so that s has a similar value to the case of ∆ξα = 0. These indicate that the peak pairing in Ref. [15] is due to finite ∆ξα and the interaction between Y and edge channels. Molecule. -- We discuss the molecule in Fig. 1(a). Its geometry implies ∆BX1 & 2∆BY and CX1X2 < CX1Y. Since the circumference of Y is shorter than two times of that of X1, one has Cg,Y < 2Cg,X1 , provided that VBG affects Cg,Y and Cg,X1 more dominantly than edge chan- nels. Then, η < 1, and Y shows the spectator behavior. For an example case of CX1X2 = 0.5CX1Y, selected AB resonance sequences are shown as a function of δB in Fig. 3. The spectator behavior occurs such that type Ia of Y-X2-Y-X1 is replaced by IIa of Y-X2-X2-X1 (III of X1-X2-X2-X1) when the relaxation in Eq. (3) occurs once (twice) within ∆BX1 . In the molecule, in addi- tion to the one-electron relaxation, there occurs two- electron relaxation, E(δQX1 ± e, δQX2 ± e, δQY ∓ e) = E(δQX1 , δQX2, δQY) ± ǫF , which is a mixture of internal and external relaxations. This additional process results in more types such as IIb of Y-X2-X2-X1. Type IIb has the same sequence as IIa, and results from Ib of Y-X2- X1-Y due to two one-electron internal relaxations and one two-electron relaxation within ∆BX1 . We plot PJ (η) of type J in Fig. 3(b). As η decreases from 1, type II (e.g., IIa and IIb) becomes dominant. Note that when ∆ξα ≃ e2/Cαα, PII and PIII are reduced by less than 10 % for 0.6 . η < 1. Qualitatively same features appear in other parameter ranges. 

1901.05077

2

1901

Breakdown of the Wiedemann-Franz law in AB-stacked bilayer graphene

We present a simple theory of thermoelectric transport in bilayer graphene and report our results for the electrical resistivity, the thermal resistivity, the Seebeck coefficient, and the Wiedemann-Franz ratio as functions of doping density and temperature. In the absence of disorder, the thermal resistivity tends to zero as the charge neutrality point is approached; the electric resistivity jumps from zero to an intrinsic finite value, and the Seebeck coefficient diverges in the same limit. Even though these results are similar to those obtained for single-layer graphene, their derivation is considerably more delicate. The singularities are removed by the inclusion of a small amount of disorder, which leads to the appearance of a "window" of doping densities $0<n<n_c$ (with $n_c$ tending to zero in the zero-disorder limit) in which the Wiedemann-Franz law is severely violated.

cond-mat.mes-hall

cond-mat

a Breakdown of the Wiedemann-Franz law in AB-stacked bilayer graphene 1Department of Physics and Astronomy, University of Missouri, Columbia, Missouri 65211, USA 2Yale-NUS College, 16 College Ave West, 138527 Singapore Mohammad Zarenia1 and Giovanni Vignale1,2 School of Physics, University of Manchester, Oxford Road, Manchester M13 9PL, UK Thomas Benjamin Smith and Alessandro Principi We present a simple theory of thermoelectric transport in bilayer graphene and report our results for the electrical resistivity, the thermal resistivity, the Seebeck coefficient, and the Wiedemann- Franz ratio as functions of doping density and temperature. In the absence of disorder, the thermal resistivity tends to zero as the charge neutrality point is approached; the electric resistivity jumps from zero to an intrinsic finite value, and the Seebeck coefficient diverges in the same limit. Even though these results are similar to those obtained for single-layer graphene, their derivation is considerably more delicate. The singularities are removed by the inclusion of a small amount of disorder, which leads to the appearance of a "window" of doping densities 0 < n < nc (with nc tending to zero in the zero-disorder limit) in which the Wiedemann-Franz law is severely violated. PACS numbers: 65.80.Ck , 72.80.Vp , 72.20.Pa Introduction -- The electric and thermal transport properties of graphene-based devices are a topic of great interest. Even setting aside their great potential for real- world applications, these systems have already offered unprecedented opportunities to study new modalities of transport, in which hydrodynamic flow patterns, gov- erned by global conservation laws, supersede the con- ventional diffusive dynamics of individual carriers [1 -- 5]. The simultaneous presence of carriers of opposite po- larities -- electrons and holes -- whose density can be tuned by chemical doping, electrostatic gating, or sim- ply by changing the temperature, creates a rich scenario of transport behaviors [6 -- 9]. A particularly interesting one has recently been observed in a single layer of ultra- clean graphene near the charge neutrality point (CNP), where the chemical potential µ = 0. The system is a zero-gap semiconductor with linearly dispersing conduc- tion and valence bands and equal numbers of electrons and holes arising from thermal fluctuations at finite tem- perature. Because µ = 0 the thermal (entropy) current coincides with the energy current, and because of the lin- ear dispersion the latter coincides with the total momen- tum density, which is a constant of the motion as long as impurities, lattice vibrations and umklapp effects are negligible. Thus we have an interesting situation in which the thermal resistivity vanishes while the electric resistiv- ity remains finite because electrons and holes, moving in opposite directions under the action of an electric field, exert mutual friction on each other [10 -- 13]. The result is that the Wiedemann-Franz (WF) ratio between the elec- tric resistivity and the thermal resistivity is enhanced well above the standard value of L0 ≡ π2(kB/e)2/3 (the so-called Wiedemann-Franz law) -- an effect that is in- deed observed experimentally [14], but only in a narrow window of doping densities around the CNP -- a window whose width shrinks to zero as the system is made less and less disordered. It should also be noted that this be- havior is diametrically opposite to what one expects and observes in heavily doped graphene: in that case, only one polarity of carriers contributes to both electric and thermal transport and the electric resistivity plummets (barring electron-impurity and electron-phonon scatter- ing) while the thermal resistivity rises to a finite value: in this case, the WF ratio drops below L0 [15 -- 18]. Motivated by these interesting findings, in this paper we investigate thermoelectric transport in a more com- plex system, AB-stacked bilayer graphene (BLG), which is a zero-gap semiconductor with electron-hole symmetric parabolic bands touching at the Dirac point [19]. BLG is a quite distinct system from its single layer with the pos- sibility of having a gate-induced tunable band gap [20]. We ask in particular, whether a large violation of the WF law will still be present in the double-layer system near CNP. At first glance, a major qualitative difference exists between the two systems, because neither the energy cur- rent nor the particle current are conserved in the bilayer. There seems to be no reason why the thermal resistivity would vanish in a bilayer at CNP when momentum-non- conserving processes are negligible. Indeed, a naive cal- culation, based on the textbook theory of thermoelectric transport, would lead precisely to this conclusion: that the thermo-electric transport coefficients are free of sin- gularities and qualitatively similar to what was obtained in single-layer graphene after the inclusion of disorder. One of the main purposes of this paper is to show that the naive conclusion is, in fact, incorrect. After introduc- ing a more careful treatment of the Boltzmann equation, which includes a conserved mode in which electrons and holes travel in the same direction, we are able to show that the transport coefficients remain singular as long as total momentum is conserved: that is to say, we find that, just as in single-layer graphene [10], the electri- cal resistivity jumps from zero to a finite universal value (controlled by the strength of the Coulomb interaction) at the CNP, the thermal resistivity tends to zero, and the Seebeck coefficient diverges. This happens because the conserved mode -- electrons and holes traveling in the same direction -- strongly overlaps the energy cur- rent mode and provides a dissipation-free channel of en- ergy transport at CNP. The inclusion of disorder, even in the smallest amount, "cures" the singularities and cre- ates a region of "disorder-enabled hydrodynamics" in the immediate vicinity of the CNP. Thus, the second pur- pose of this paper is to determine the qualitative behav- ior of the thermo-electric transport coefficients of bilayer graphene in this regime. We find that the WF ratio, plotted as a function of doping density, follows a squared Lorentzian behavior, whose quarter-maximum occurs at doping density nc proportional to the strength of disor- der, such that the WF ratio increases with decreasing dis- order for n < nc and decreases with decreasing disorder for n > nc. At the same time, the Seebeck coefficient ex- hibits an interesting non-monotonic behavior, vanishish- ing at CNP and peaking in absolute value at n ∼ nc. This behavior, admittedly very similar to what has been predicted and observed in single-layer graphene, should be promptly comparable with the results of experimental measurements, as soon as they become available. Within the framework of quasi-classical transport the- ory [21], the state of the carriers is described by a non- equilibrium distribution function fk,γ, where k is the Bloch wave vector and γ = ±1 is the band index. The deviation from equilibrium is δfk,γ = fk,γ − f0k,γ, where f0k,γ is the equilibrium distribution function at chemical potential µ and temperature T . The quantities of inter- est are the electric current je and the thermal current jq, however, in order to homogenize the dimensions we will be working with the particle current jn = je/(−e) (−e is the charge of the electron) and the entropy current in units of the Boltzmann constant kB, js = βjq, where β = (kBT )−1. With this choice, the thermoelectric ma- trix [see below Eq. (2)] manifestly satisfies Onsager reci- procity. The currents are related to the non-equilibrium distribution function by [22] vk,γδfk,γ, jn =Xk,γ js =Xk,γ βǫk,γvk,γδfk,γ , (1) where ǫk,γ = γ[p(t/2)2 + (vk)2 − t/2] and vk,γ = 2γv2k/pt2 + (2vk)2 are, respectively, the energy and the velocity of band γ = ± at the wave vector k, while ǫk,γ ≡ ǫk,γ − µ. Here v = 108 cm/s is the Fermi velocity and t = 0.4 eV is the vertical interlayer hopping between the two layers [23, 24]. Our two-band approximation is justified up to T ∼ t/kB ∼ 4, 600 K. The currents are connected to the electric field E and to the temperature gradient ∇T by the ther- moelectric resistivity matrix, ρ, which we define as (a) 2-mode 20 15 10 5 0 0 0.2 0.6 0.4 n (1012 cm-2) 0.8 1 2 (b) 3-mode intrinsic resistivity with disorder clean 0.5 1 1.5 n (1012 cm-2) 20 15 10 5 0 0 Figure 1. A comparison between the dimensionless electric and thermal resistivities ¯ρel and ¯ρth, calculated with a (a) two- mode Ansatz and (b) three-mode Ansatz. Curves are plotted as functions of density (in units of 1012 cm−2 and for T = 100 K. Solid curves correspond to the clean limit, whereas dotted ones are the calculated in the presence of small amount of disorder (λ = 0.005). − (eE , kB∇T )T = ρ·(jn , js)T (here "T" stands for the vector transposition). The elements of ρ are expressed in terms of three transport coefficients: the reduced electric resistivity ¯ρel, i.e. the ordinary electric resistivity multi- plied by e2, the reduced thermal resistivity ¯ρth, i.e. the usual thermal resistivity multiplied by k2 BT , and the di- mensionless Seebeck coefficient ¯Q, i.e. the ordinary See- beck coefficient expressed in units of kB/e, in the follow- ing form (see Ref. 21) ρ =(cid:18) ¯ρel + ¯Q2 ¯ρth ¯Q¯ρth ¯ρth (cid:19) , ¯Q¯ρth (2) with detρ = ¯ρel ¯ρth. The (dimensionless) Wiedemann- Franz ratio is defined as W F ≡ ¯ρel/ ¯ρth = detρ/ ¯ρ2 th,. W F = π2/3 when the Wiedemann-Franz law is satisfied in its standard form, e.g., for a parabolic band electron gas in the presence of quenched short-range disorder. Two-mode Ansatz. The standard textbook calculation of thermoelectric coefficients [21] starts with the intro- duction of a 2-parameter Ansatz δfk,γ = f ′ 0k,γvk,γ · (pn + βǫk,γps), where vk,γ and βǫk,γvk,γ are the "modes" used to expand δfk,γ. The two parameters pn and ps cor- respond to shifts of the particle momentum associated with the jn and js, respectively. Finally, the factor f ′ 0k,γ, which denotes the derivative of the Fermi distribution with respect to energy, accounts for the fact that only electrons around the Fermi surface are mobile. We stress that the choice of the Ansatz is the delicate point of the entire calculation. (The rest of the section is completely general and valid independently of such choice.) The two-parameter Ansatz above is, for BLG, incomplete and leads to the wrong results for ¯ρel and ¯ρth. In the following section we will amend it by including a third parameter, corresponding to the current conserved by the collision integral (i.e. the momentum density). Using Eq. (1) we obtain (jn , js)T = D · (pn , ps)T, where D is the 2 × 2 matrix of "Drude weights" ρ = D−1 · I · D−1. For future purposes we define the conductivity matrix σ ≡ ρ−1 = D · I−1 · D. 3 Dij = 1 2Xk,γ f ′ 0kγui k,γ · uj k,γ , (3) where ui k,γ = (vk,γ, βǫk,γvk,γ)i and i, j = 1, 2. Dij quantifies the "overlap" between the modes ui and uj (it can in fact be interpreted as a scalar product in the mode space). To determine pn and ps, we substitute the Ansatz for δfk,γ into the Boltzmann equation for the steady state response in the presence of fields E and ∇T , − f ′ 0k,γvk,γ · [eE + βǫk,γkB∇T ] = Ik,γ . (4) i.e. Here Ik,γ is the collision integral, which depends on the details of the microscopic scattering mechanism. Eq. (4) is projected over the same set of modes which are used to expand δfk,γ, it is multiplied by one of the modes ui k,γ, integrated over k and summed over bands. In this way, the differential Eq. (4) is transformed into an algebraic one, and is easily solved for pn and ps. The key inputs are the moments of the collision inte- gral, which to linear order in pn and ps are given by k,γIk,γ = Iij (pn , ps)j. Hereafter summation of repeated latin indices is understood. Such equation de- fines the 2×2 matrix I, which we refer to as "collision ker- nel", and whose matrix elements are the Iij (i, j = 1, 2). In the supplementary online material [25] we make use of a standard approximation for the Coulomb collision integral (screened interaction plus Fermi golden rule) to find Pk,γ ui Iij = − β 4π Xq Z ∞ −∞ dω V (q)2(ℑΠ1 i ℑΠ1 j − ℑΠ0ℑΠ2 ij ) sinh2(βω/2) , (5) where the response functions Πα(q, ω) are defined as Πα = 4Xγ,γ ′Xk F γγ ′ k,k+q(f0k,γ − f0k+q,γ ′) ǫk,γ − ǫk+q,γ ′ + ω + i0+ M α k,k+q,γ,γ ′. k,γ − ui k,γ − ui k,γ − uj k,k+q,γ,γ ′ = x · (ui k+q,γ ′) · (uj k,k+q,γ,γ ′ = 1, M 1 k,k+q,γ,γ ′ = (ui (6) Here M 0 k+q,γ ′), and M 2 k+q,γ ′). Furthermore, V (q) = 2πe2/[κ(q + qTF)] is the screened Coulomb interaction, qTF = 4e2εF /(κ2v2) is the Thomas-Fermi screening wave vector (εF is the Fermi energy). The factor 4 accounts for the spin and val- ley degeneracy and κ is the dielectric constant of the substrate (within our calculations, we set κ = 4 as for an h-BN substrate). The form factors F γγ ′ k,k+q = [1 + γγ′ cos(2θk − 2θk+q)]/2, where θk is the angle formed by the k vector with the x-axis, come from the over- lap of the wave functions at wave vectors k and k + q. Simple algebraic manipulations lead to the final expres- sion for the two-mode thermoelectric resistivity matrix, remarkably well This procedure works for both parabolic-band electron gases and monolayer graphene in the clean limit. [10] In such limit, i.e. when the only col- lision mechanism is the electron-electron interaction, the thermoelectric transport is strongly influenced by the ex- act conservation of the momentum density. For example, when one of the two currents overlaps sufficiently with the momentum density [where the overlap is defined as in Eq. (3)], then it is also conserved. This in turn im- plies that either ¯ρel or ¯ρth vanishes. For parabolic-band electron gases and (massive or massless) Dirac systems, such result is readily obtained with a two-mode Ansatz. In such special cases, the overlap between the two cur- rents and momentum (which coincides, in fact, with one of them at all times) is automatically taken into account. It is then clear what fails when the two-mode Ansatz is applied to BLG: the overlap between the currents and the momentum density, the conserved mode of the collision integral, is not explicitly taken into account. Hence, all the coefficients of the matrix ρ are non-zero, and so is its determinant. Therefore, all thermoelectric transport coefficients are finite, see Fig. 1(a). Such result is wrong as we proceed to show in the next section. Three-mode Ansatz. To account for the conserved mode of the collision integral (the momentum density) we express the deviation of fk,γ from equilibrium as δfk,γ = f ′ 0k,γ(cid:2)v2k · pk/t + vk,γ · (pn + βǫk,γps)(cid:3) , (7) where pk is the shift associated with the momentum den- sity scaled with t/v2, in such a way that it has the same dimension as the other modes. A more general situation (away from the CNP) might require to include more modes, in order to achieve a bet- ter quantitative agreement with the true solution of the Boltzmann equation. However, as we will shown later, the Ansatz (7) is able to capture all the main qualitative features (vanishing of the thermal/charge conductivities, divergence of the Seebeck coefficient, etc.). In addition, there is no real locking imposed a priori between the electron and hole velocities. As such, the three modes of the anstaz describe different situations, in which parti- cles and holes may co-propagate, counter-propagate, or any combination in between. The coefficients of the sep- arate modes are allowed to vanish, describing situations in which one particular configuration is realised. We now define the three-component mode vector as uj k,γ = (v2k/t, vk,γ, βǫk,γvk,γ)j. Using Eq. (1) and (3), we get (jn, js)i = Dij(pk, pn, ps)j , where i = 1, 2 and j = 0, 1, 2,. Note that D is now a 2 × 3 matrix. Similarly, for the Coulomb collision kernel we find Ijj ′ =  0 η 0 0 I11 I12 0 I12 I22  jj ′ (8) 4 (a) Numerical WF ratio (scaled with π2/3) and (b) Seebeck coefficient as a function of density for different Figure 2. temperatures as labeled. The disorder strength is λ = 0.005. The full red dots indicate the critical densities nc which are obtained by setting ρD = ¯ρel,C. where j, j′ = 0, 1, 2. The Iij are defined in Eq. (A.12). The conductivity matrix is easily seen to be σ = D · I−1 · D. Inverting it, in the limit of η → 0, we obtain (cid:18) [D02]2 −D01D02 [D01]2 (cid:19) −D01D02 [D01]2 σ22 + [D02]2σ11 − 2D01D02[σ12]2 ρ = (9) where σii′ = limη→0 σii′ coincide with the conductivities obtained with the two-mode Ansatz. Note that, since detρ = ¯ρel ¯ρth = 0, either the electrical or thermal re- sistivity must necessarily be zero. At charge neutrality D01 = 0. Therefore, ρ11(n = 0) = σ−1 11 , while all the other components of ρ are zero. Thus, at the CNP, the thermal resistivity vanishes and the electrical resistivity is finite. Away from the CNP we find ¯ρth(n 6= 0) = ρ22, ¯ρel(n 6= 0) = 0, ¯Q = −D02/D01 . (10) Numerical results for ¯ρel and ¯ρth are shown in Fig. 1. Within the two-mode approximation [Fig. 1(a)] both the electrical and thermal resistivities are always finite. In contrast, with the three mode Ansatz [Fig. 1(b)], the electrical resistivity ¯ρel(n 6= 0) = 0 and exhibits a dis- continuity at the CNP in the absence of disorder (solid curves). Interestingly, both models yield the same intrin- sic resistivity at n = 0 [full red dots in Fig. 1(a) and (b)]. At this point the only carriers in the system are ther- mally excited electrons and holes, in the conduction and valence bands, respectively. The two types of carries drift in opposite directions under the action of an electric field. Because of the transfer of momentum (known as Coulomb drag [26]) between them, the resistivity becomes finite. Disorder-enabled hydrodynamics. An infinitesimal amount of disorder, which breaks the exact conservation of momentum, regularizes the singular results of the clean limit. We assume a momentum-non-conserving kernel ID (D for disorder) proportional to a dimensionless momen- tum relaxation rate λ, which we take to be ≪ 1. The pre- cise form of ID is not important for our purposes. How- ever, for the sake of illustration, we will later make use of a simple model of electrons and holes scattering against randomly distributed impurities of density nd with short- range potential V0δ(r): for this model λ = ndV 2 0 /(v)2, as detailed in the supplementary online material [25]. el ≃ ¯ρ−1 el,C + ρ−1 The resistivity matrix is now the sum of two term ρ = ρC + ρD, where ρC is given by Eq. (9), whereas ρD = D−1 · ID · D−1. The electrical and thermal resistivities now read [10] ¯ρ−1 D and ¯ρth = ¯ρth,C + ¯ρth,D, where ρD = ¯ρth,D(D02/D01)2. In Fig. 1(b) we plot such ¯ρel and ¯ρth (dotted curves) and compare them with the results of the previous section (solid curves). The effect of disorder on the thermal resistivity is just a small shift. Conversely, it regularizes the electric resistivity. From the results above we obtain the WF ratio W F =(cid:18) Γ (D01/D02)2 + Γ2(cid:19)2 , Γ2 = ¯ρth,D ¯ρel,C . (11) which is a square of a Lorentzian. This formula shows that, at the CNP, W F → 1/Γ2, i.e. it is greatly enhanced relative to its standard noninteracting value π2/3 and diverges as the strength of disorder tends to zero. In Figs. 2(a) and (b) we show numerical results for W F and the Seebeck coefficient ¯Q, respectively. We define a crossover density nc, below which the enhancement of the WF persists, from the condition ρD = ¯ρel,C [the dots in Fig. 2(a) indicate its position]. We call such regime (n < nc) "disorder-enabled hydrodynamics" [10]. In it, W F remains much larger than π2/3, in fact larger than disorder-regularized 1/Γ2 ≫ π2/3. (A more lenient crossover density could be defined [10] as that at which W F first drops below π2/3.) The coefficient [Figs. 2(b)] exhibits a large swing about the CNP and goes to zero at n = 0, as expected from particle-hole symmetry. The swing region, in which the derivative of ¯Q vs density reverses its sign, is yet another incarnation of the disorder-enabled hydrodynamic regime [10]. Its width is defined by the same condition n < nc. Seebeck In Summary, we have presented the theory of ther- moelectric transport in clean bilayer graphene. We have shown that the conventional semi-classical (Boltzmann) textbook approach to the calculation of thermoelectric coefficients [21], which works remarkably well for both parabolic-band electron gases and Dirac systems [10], fails for such system. This is attributed to the fact that neither the particle current nor the energy current are conserved quantities in bilayer graphene. The cor- rect results are found by explicitly including the current conserved by the electron-electron collision integral, i.e. the momentum density, in the "augmented" Ansatz for the non-equilibrium distribution function. Note that in previously studied examples the explicit addition of the conserved momentum mode was not required. This for- tunate situation occurred because the momentum mode was automatically subsumed under either the particle current or the energy current mode. Bilayer graphene is the first system studied in this context in which the two-mode Ansatz fails and the momentum mode must be introduced explicitly. We find that, at the charge neutrality point: (i) the thermal resistivity vanishes; (ii) the electric resistivity jumps from zero to a finite value; (iii) the Seebeck coeffi- cient diverges. These singularities are cured by the inclu- sion of a small amount of disorder. Breaking the exact momentum conservation, it opens a "window" of doping densities 0 < n < nc (with nc tending to zero in the zero-disorder limit) in which the Wiedemann-Franz law is largely violated. There, the WF ratio greatly exceeds the standard value and the Seebeck coefficient exhibits a non-monotonic behavior (as a function of doping den- sity). Such predictions can be tested in experiments in sufficiently clean samples of bilayer graphene. Acknowledgment. This work was supported by the U.S. Department of Energy (Office of Science) under grant No. DE-FG02-05ER46203. A.P. and T.B.S. acknowledge support from the Royal Society International Exchange grant IES\R3\170252. [1] B. N. Narozhny, I. V. Gornyi, A. D. Mirlin, and J. Schmalian, Annalen der Physik 529, 1700043 (2017). and M. Polini, [2] A. Principi, G. Vignale, M. Carrega, 5 Phys. Rev. B 93, 125410 (2016). [3] B. N. Narozhny, I. V. Gornyi, M. Titov, M. Schutt, and A. D. Mirlin, Phys. Rev. B 91, 035414 (2015). [4] U. Briskot, M. Schutt, I. V. Gornyi, M. Titov, B. N. Narozhny, and A. D. Mirlin, Phys. Rev. B 92, 115426 (2015). [5] A. Lucas and K. C. Fong, Journal of Physics: Condensed Matter 30, 053001 (2018). [6] D. A. Bandurin, I. Torre, R. K. Kumar, M. Ben Shalom, A. Tomadin, A. Principi, G. H. Auton, E. Khestanova, K. S. Novoselov, I. V. Grigorieva, L. A. Ponomarenko, A. K. Geim, and M. Polini, Science 351, 1055 (2016). [7] F. Ghahari, H.-Y. Xie, T. Taniguchi, K. Watanabe, M. S. Foster, and P. Kim, Phys. Rev. Lett. 116, 136802 (2016). [8] R. Krishna Kumar, D. A. Bandurin, F. M. D. Pel- legrino, Y. Cao, A. Principi, H. Guo, G. H. Auton, M. Ben Shalom, L. A. Ponomarenko, G. Falkovich, K. Watanabe, T. Taniguchi, I. V. Grigorieva, L. S. Levi- tov, M. Polini, and A. K. Geim, Nature Physics 13, 1182 (2017). [9] R. Gurzhi, Sov. Phys. JETP 44, 771 (1963). [10] M. Zarenia, A. Principi, and G. Vignale, arXiv:1811.08914 (2018). [11] L. Fritz, J. Schmalian, M. Muller, and S. Sachdev, Phys. Rev. B 78, 085416 (2008). [12] M. Muller, L. Fritz, and S. Sachdev, Phys. Rev. B 78, 115406 (2008). [13] D. Svintsov, V. Vyurkov, S. Yurchenko, T. Otsuji, and V. Ryzhii, Journal of Applied Physics 111, 083715 (2012). [14] J. Crossno, J. K. Shi, K. Wang, X. Liu, A. Harzheim, A. Lucas, S. Sachdev, P. Kim, T. Taniguchi, K. Watan- abe, T. A. Ohki, (2016), 10.1126/science.aad0343. and K. C. Fong, Science [15] A. Principi and G. Vignale, Phys. Rev. B 91, 205423 (2015). [16] A. Lucas and S. Das Sarma, Phys. Rev. B 97, 245128 (2018). [17] A. Lucas and S. A. Hartnoll, Phys. Rev. B 97, 045105 (2018). [18] H.-Y. Xie and M. S. Foster, Phys. Rev. B 93, 195103 (2016). [19] E. McCann and M. Koshino, Reports on Progress in Physics 76, 056503 (2013). [20] Y. Zhang, T.-T. Tang, C. Girit, Z. Hao, M. C. Martin, A. Zettl, M. F. Crommie, Y. R. Shen, and F. Wang, Nature 459, 820 EP (2009). [21] N. Ashcroft and N. Mermin, Solid State Physics (Cen- gage Learning, 2011). [22] D. Pines and P. Nozi`eres, The Theory of Quantum Liq- uids: Normal Fermi liquids (W.A. Benjamin, 1966). [23] E. McCann, D. S. Abergel, and V. I. Fal'ko, Solid State Communications 143, 110 (2007). [24] E. McCann and M. Koshino, Reports on Progress in Physics 76, 056503 (2013). [25] See Supplemental Material at [URL will be inserted by publisher] for more details. [26] M. Schutt, P. M. Ostrovsky, M. Titov, I. V. Gornyi, B. N. Narozhny, and A. D. Mirlin, Phys. Rev. Lett. 110, 026601 (2013). BREAKDOWN OF THE WIEDEMANN-FRANZ LAW IN AB-STACKED BILAYER GRAPHENE SUPPLEMENTAL MATERIAL In this supplemental material, we present more details on our calculations for the electron-electron Coulomb (Sec. A) as well as our short-range model for the disorder collision moments (Sec. B). 6 A. ELECTRON-ELECTRON COLLISION MOMENTS The electron-electron collision integral for the band γ and at wave vector k, Ik,γ is given by [21, 22], Ik,γ = −Pk′Pγ ′,η,η′Pq W (q)(cid:2)fk,γ(1 − fk−q,γ ′)fk′,η(1 − fk′+q,η′) − fk−q,γ ′(1 − fk,γ)fk′+q,η′(1 − fk′,η)(cid:3) × δ(ǫk,γ + ǫk′,η − ǫk−q,γ ′ − ǫk+q,η′) (A.1) where the momentum conservation appears naturally when doing the second quantization of Coulomb interaction in k-space, and the energy conservation stems from the Fermi golden rule. fk,γ = f0k,γ + δfk,γ is the non-equilibrium distribution function and W (q) = (2π/)V (q)2 defines the collision probability where V (q) is the statistic screened Coulomb interaction. Having the two-mode Ansatz δfk,γ = f ′ 0k,γvk,γ · (pn + βǫk,γps) (A.2) and using and we obtain δ(ǫk,γ + ǫk′,η − ǫk−q,γ ′ − ǫk+q,η′) = Z ∞ −∞ dωδ(ǫk′,η − ǫk′+q,η′ + ω)δ(ǫk,γ − ǫk−q,γ ′ − ω) (A.3) f0k,γ(1 − f0k±q,γ ′)δ(ǫk,γ − ǫk±q,γ ′ ± ω) = f0k,γ − f0k±q,γ ′ ∓2e±βω/2 sinh(βω/2) δ(ǫk,γ − ǫk±q,γ ′ ± ω), (A.4) F γγ ′ k,k−q(f0k,γ − f0k−q,γ ′)δ(ǫk,γ − ǫk−q,γ ′ − ω)F ηη′ k′,k′+q(f0k′,η − f0k′+q,η′ )δ(ǫk′,η − ǫk′+q,η′ + ω) × Ik,γ = 2πβ 4 Pγ ′,η,η′Pk′PqR ∞ −∞ dω V (q)2 sinh2(βω/2) × [(vk,γ − vk−q,γ ′ + vk′,η − vk′+q,η′) pn + β (vk,γ ǫk,γ − vk−q,γ ′ǫk−q,γ ′ + vk′,ηǫk′,η − vk′+q,η′ǫk′+q,η′)] ps, (A.5) where {γ, γ′, η, η′} = ±1 denote the band index and F γγ ′ functions at vector k and k ± q. Now the moments of the electron-electron collision integrals are calculated using k,k±q is the form factor comes from the overlap of the wave Using Eq. (A.5) we write, where the matrix elements Gij(q, ω, T ) are Pk,γ βǫk,γvk,γ Ik,γ ! = I ·(cid:18) pn Pk,γ vk,γ Ik,γ ps (cid:19) . 2 Xq Z ∞ V (q)2 πβ dω −∞ sinh2(βω/2)(cid:18) G11 G12 G21 G22 (cid:19) ·(cid:18) pn ps (cid:19) , k,γF γγ ′ k′+q,η′)F ηη′ k,k−q(f0k,γ − f0k−q,γ ′)δ(ǫk,γ − ǫk−q,γ ′ − ω)(cid:17) × k′,k′+q(f0k′,η − f0k′+q,η′)δ(ǫk′,η − ǫk′+q,η′ + ω)(cid:17) + k,k−q(f0k,γ − f0k−q,γ ′)δ(ǫk,γ − ǫk−q,γ ′ − ω)(cid:17) × k−q,γ ′)F γγ ′ k′,k′+q(f0k′,η − f0k′+q,η′)δ(ǫk′,η − ǫk′+q,η′ + ω)(cid:17) . Pk,γ βǫk,γvk,γ Ik,γ ! = Pk,γ vk,γ Ik,γ Gij =(cid:16)Pk,γ,γ ′ ui (cid:16)Pk′,η,η′(uj (cid:16)Pk,γ,γ ′ ui k′,η − uj k,γ(uj k,γ − uj (cid:16)Pk′,η,η′ F ηη′ (A.6) (A.7) (A.8) As defined in the main text, the two-component vector ui manipulations, one can show that the first and the third lines in Eq. (A.8) are equivalent to k,γ = (vk,γ, βǫk,γvk,γ)i (i = 1, 2) . By simple algebraic 7 and 1 Pk,γ,γ ′ ui 2Pk,γ,γ ′ F γγ ′ Pk,γ,γ ′ ui 2Pk,γ,γ ′(ui k,γ(uj k,γ − ui − 1 k,γF γγ ′ k,k−q(f 0 k,γ − ui k,γ − f 0 k+q,γ ′)(f 0 k−q,γ ′)δ(ǫk,γ − ǫk−q,γ ′ − ω) = k,γ − f 0 k+q,γ ′)δ(ǫk,γ − ǫk+q,γ ′ + ω) k,k+q(ui k,γ − uj k+q,γ ′)(uj k−q,γ ′)F γγ ′ k,γ − uj k,k−q(f 0 k+q,γ ′)F γγ ′ k,γ − f 0 k−q,γ ′)δ(ǫk,γ − ǫk−q,γ ′ − ω) = k,k+q(f 0 k,γ − f 0 k+q,γ ′)δ(ǫk,γ − ǫk+q,γ ′ + ω). (A.9) (A.10) (A.11) (A.12) Inserting Eqs. (A.9) and (A.10) into Eq. (A.8) leads to Gij = 1 2(cid:16)Pk,γ,γ ′(ui k′,η − uj (cid:16)Pk′,η,η′(uj k,γ − ui k′+q,η′)F ηη′ k,γ − uj k,k+q(f0k,γ − f0k+q,γ ′)δ(ǫk,γ − ǫk+q,γ ′ + ω)(cid:17) × k+q,γ ′)F γγ ′ k′,k′+q(f0k′,η − f0k′+q,η′)δ(ǫk′,η − ǫk′+q,η′ + ω)(cid:17) − k+q,γ ′)F γγ ′ k,k+q(f 0 k,γ − f 0 k,γ − ui k+q,γ ′)(uj k+q,γ ′)δ(ǫk,γ − ǫk+q,γ ′ + ω)(cid:17) × (cid:16)Pk′,η,η′ F ηη′ k′,k′+q(f0k′,η − f0k′+q,η′)δ(ǫk′,η − ǫk′+q,η′ + ω)(cid:17) . 1 2(cid:16)Pk,γ,γ ′(ui Using this equation, I becomes, I = − β 4π Xq Z ∞ −∞ dω V (q)2 sinh2(βω/2)(cid:18) (ℑΠ10)2 − ℑΠ00ℑΠ20 ℑΠ10ℑΠ01 − ℑΠ00ℑΠ11 (ℑΠ01)2 − ℑΠ00ℑΠ02 (cid:19) ℑΠ10ℑΠ01 − ℑΠ00ℑΠ11 where the δ-functions are written in terms of the imaginary part of the response functions Πnm(q, ω) (n, m = 0, 1, 2) Πnm = 4Xγ,γ ′Xk F γγ ′ k,k+q(vk,γ − vk+q,γ ′)n(ǫk,γvk,γ − ǫk+q,γ ′vk+q,γ ′)m(f0k,γ − f0k+q,γ ′) ǫk,γ − ǫk+q,γ ′ + ω + i0+ . (A.13) Equations (A.12) and (A.13) are respectively represented in a more compact forms of Eqs. (5) and (6) in the main text. B. DISORDER COLLISION MOMENTS The non-momentum-conserving disorder collision integral is given by, I Dk,γ = 2πg  Xγ ′ Xk′ Uk−k′2F γγ ′ k,k′(fk,γ − fk′,γ ′)δ(ǫk,γ − ǫk′,γ ′) (B.1) where the factor g = 4 accounts for the spin and valley degeneracy. Considering the short-range disorder characterized by an effective strength of Uk−k′ ≈ (ndV 2 0 )1/2 (nd is the disorder density and V0δ(r) is the disorder potential), the linearized collision integral becomes I Dk,γ = 2πndV 2 0  Xγ ′ Xk′ (cid:20)(cid:18)vk,γ − vk′,γ ′ ǫk′,γ ′ β(cid:18)vk,γǫk,γ ∂f0k,γ ∂ǫk,γ ∂f0k′,γ ′ − vk′,γ ′ ∂f0k,γ ∂ǫk,γ ∂ǫk′,γ ′ (cid:19) · pn + ∂ǫk′,γ ′ (cid:19) · ps(cid:21) δ(ǫk,γ − ǫk′,γ ′) ∂f0k′,γ ′ in which we set F γγ ′ δ(ǫk,γ − ǫk′,γ ′) = δ(k − k′)/vk,γ, we easily construct the moments of the disorder collision integral k,k′ = 1 for the sake of simplicity and without loss of qualitative behavior of disorder. Using (cid:18) Pk,γ vk,γI Dk,γ ps (cid:19) , Pk,γ βǫk,γvk,γ I Dk,γ (cid:19) = ID ·(cid:18) pn (B.2) where, ID = ndV 2 0 22 Xk,γ vk,γ ∂f0k,γ ∂ǫk,γ (cid:18) 1 βǫk,γ (βǫk,γ)2 (cid:19) . βǫk,γ 8 (B.3) Taking kF , v, and β−1, respectively as units of wave vector, velocity and energy, ID can be further simplified to, ID = λβ  (cid:17)3 4π (cid:16) εF Xγ Z ∞ 0 dk k2γvk,γ exp(ǫk,γ) [1 + exp(ǫk,γ)]2 (cid:18) 1 βǫk,γ (βǫk,γ)2 (cid:19) , βǫk,γ (B.4) where the dimensionless disorder strength λ = ndV 2 full energy bands of bilayer graphene. 0 /(v)2. We solve the integrals in Eq. (B.4) numerically for the 

1508.03060

1

1508

Three-terminal heat engine and refrigerator based on superlattices

We propose a three terminal heat engine based on semiconductor superlattices for energy harvesting. The periodicity of the superlattice structure creates an energy miniband, giving an energy window for allowed electron transport. We find that this device delivers a large power, nearly twice than the heat engine based on quantum wells, with a small reduction of efficiency. This engine also works as a refrigerator in a different regime of the system's parameters. The thermoelectric performance of the refrigerator is analyzed, including the cooling power and coefficient of performance in the optimized condition. We also calculate phonon heat current through the system, and explore the reduction of phonon heat current compared to the bulk material. The direct phonon heat current is negligible at low temperatures, but dominates over the electronic at room temperature and we discuss ways to reduce it.

cond-mat.mes-hall

cond-mat

a Three-terminal heat engine and refrigerator based on superlattices Yunjin Choi1 and Andrew N. Jordan1, 2, ∗ 1Department of Physics and Astronomy & Rochester Theory Center, University of Rochester, Rochester, New York 14627, USA 2Institute for Quantum Studies, Chapman University, 1 University Drive, Orange, CA 92866, USA (Dated: June 25, 2021) We propose a three terminal heat engine based on semiconductor superlattices for energy har- vesting. The periodicity of the superlattice structure creates an energy miniband, giving an energy window for allowed electron transport. We find that this device delivers a large power, nearly twice than the heat engine based on quantum wells, with a small reduction of efficiency. This engine also works as a refrigerator in a different regime of the system's parameters. The thermoelectric perfor- mance of the refrigerator is analyzed, including the cooling power and coefficient of performance in the optimized condition. We also calculate phonon heat current through the system, and explore the reduction of phonon heat current compared to the bulk material. The direct phonon heat current is negligible at low temperatures, but dominates over the electronic at room temperature and we discuss ways to reduce it. I. INTRODUCTION There has been increasing interest in developing high efficiency, high power thermoelectric devices, constructed from the bottom-up using nanoscale designs. The pri- mary applications driving interest in this area are energy- harvesting, the collection and conversion of waste heat to electrical power, produced from sources ranging from hand-held electronics to industrial sources of heat, and refrigeration, actively cooling a spatial region via elec- trons to evacuate heat out of an area. The use of nanoscale architecture instead of bulk materials is moti- vated by the low figure of merit - or poor thermoelectric conversion efficiency - of bulk materials, whereas con- duction through nanoscale electronics can reach Carnot efficiency. One way to produce high thermodynamic efficiency in the conversion of heat to power is the use of structures with sharp spectral features, such as quantum dots [1 -- 3]. The use of quantum dots in thermoelectric transport has been extensively researched in the past several years [4 -- 11]. See Ref. [12, 13] for recent reviews of these and related activities. In particular, if the dot is transporting electrons via resonant tunneling, the quantum dot acts as an energy filter - permitting the "tight-coupling" of heat and charge transport which can lead to Carnot effi- ciency. Other structures from mesoscopic physics, includ- ing the quantum point contact and electron cavity [14], quantum wells [15, 16], quantum Hall bar [17, 18], su- perconducting leads [19], and Coulomb blockaded quan- tum dots(s) [20 -- 24] have also been investigated for their multi-terminal thermoelectric properties. The late Prof. Markus Buttiker, for whom this special issue is in mem- ory of, was highly influential in the theoretical develop- ment of these ideas, as can be seen in the above list of references. ∗ jordan@pas.rochester.edu Several experiments have begun exploring this physics. Prance et al. [25] performed experiments on a cavity con- nected to resonant tunneling quantum dots acting as an electronic refrigerator, based on the proposal of Edwards et al. [26]. They demonstrated that applying bias to the system results in cooling a large 6µm2 cavity from 280mK to below 190mK. Very recently, Roche et al. [27] and F. Hartmann et al. [28] showed rectification of electrical current of the nano Amp scale and power pro- duction on a pico Watt scale from a capacitively coupled source of fluctuations. This was based on the theoretical proposal of Sothmann et al. [14]. While a nanoscale thermoelectric generator can power nanoscale devices, it is of great interest to find practical ways to scale up these nanoengines. One way is to simply add them in electrical series while being able to couple to a common source of heat. In commercial thermoelectric generators, this is usually done by alternating the semi- conductor type, of either p-type or n-type to be able to apply the heat difference in parallel because the heat and electrical transport are in opposite directions in a p-type semiconductor [29]. This permits the generated voltage to grow with the number of elements, while keeping the current fixed. Various other ways of scaling the devices have been proposed [30 -- 33]. In Jordan et al., a layered structure was proposed by alternating layers of semicon- ductor and self-assembled quantum dots, so as to create a large-scale device where heat and electrical transport are separated, while keeping the high thermodynamic ef- ficiency [9]. This is a parallel strategy of scaling, so the generated current grows with the number of dots, while the voltage difference is fixed. Sothmann et al. consid- ered a technically simpler method of creating quantum wells that permit resonant tunneling [15]. The physics there is somewhat different because energy may be dis- tributed into the transverse degrees of the electron mo- tion. Nevertheless, reasonable thermodynamic efficiency was found, with increased power production. One of the outstanding challenges to creating high- efficiency thermoelectric devices is phonon transport. Phonons give a way for the hot and cold side of the de- vice to exchange energy directly, without converting it to power via the electrons. Therefore, any possible way to reduce the phonon transport while still allowing elec- tron transport will aid in the overall thermodynamic effi- ciency. Interface-based devices, such as described above can help with this, because the interface helps to reflect the phonons [34 -- 41]. Ideally, there will be additional ma- terial layers that act as thermal insulators. The purpose of the present article is to build on these accomplishments, and make an analysis of a thermoelectric device based on semiconductor super- lattices. These structures are fabricated by making a periodic layered structure of alternating materials, such as GaAs/AlGaAs. The effect on the electronic transport is to form a series of mini-bands of allowed and forbid- den energies where conduction electrons can transport [42 -- 45]. This structure can be considered as a general- ization of the resonant tunneling quantum wells. The mini-band gives a top-hat profile of variable width for the energy-filtering. Such a top-hat profile has been ar- gued by Whitney to offer the highest efficiency for a given power extraction [46]. However, we note the transverse degrees of freedom make the system somewhat differ- ent. At a small band width, our system will be sim- ilar to a quantum well, but can be extended to allow a fixed width longitudinal energy window. We make a first-principles analysis of the heat and charge transport in a three-terminal geometry, where two terminals carry charge, and a third carries heat (see Fig. 1). The off- set of the miniband centers and their respective widths determine the power produced and efficiency of the heat conversion given fixed temperature differences. We next make an analysis of the coefficient of performance of this device for the purposes of refrigeration of the central re- gion. The final purpose of the present work is to also make a systematic calculation of the heat transport due to phonons. We make a detailed investigate the heat current through the device from phonon transport using a Kronig-Penney model, and consider different ways to stop it. The paper is organized as follows. In section II, we introduce a model of the superlattice heat engine in its dual roles: the energy harvester in section II A and the refrigerator in section II C. We discuss our results for the generated power and the efficiency of the engine in section II B, and show the cooling power and the coefficient of performance in section II C. The second part, section III focuses on phonon heat current generated by the heat engine and discusses the effect on the efficiency of the heat engine. We finish with our conclusions in section IV. 2 II. HEAT ENGINE BASED ON SUPERLATTICES A. Energy harvesting by electron transport We consider a setup shown schematically in Fig. 1. It consists of a center cavity connected to two electronic reservoirs r = L, R via a superlattice. The electronic reservoirs are characterized by the occupation of the states given by the Fermi function, fr(E) = [exp[(E − µr)/(kBTr)]+1]−1 with temperature Tr and chemical po- tential µr, and the underlying assumption is that inelastic scattering processes restore the local thermal equilibrium on a fast time scale. The center cavity is also assumed to be in thermal equilibrium with a heat bath of tem- perature Tc. We assume the structure is translationally symmetric within the x and y directions, perpendicular to the growth direction z. The superlattices are designed as periodic structures with lattice constant d (sum of the width of well and barrier thickness). The periodicity of the structure in the z direction implies that the eigen- states of the Hamiltonian can be written as Bloch states with the Bloch vector q ∈ [−π/d, π/d], so the simple so- lutions with the Kronig-Penney model resemble the stan- dard superlattices [43]. The corresponding eigenvalues of the Hamiltonian form a miniband. The allowed energies of the miniband can be written, Eν(q) = E ν 0 − 2βν cos qd, (1) where this is the result of a standard tight-binding calcu- lation with band indices ν. E ν 0 is a center of miniband ν and its width is 4βν. Our discussion is restricted to the electron transport through the lowest miniband, thus the miniband index ν is neglected from here on. To find the electric and heat currents through the su- perlattices expressed in terms of an integral over energy, we first find the density of states of the superlattice. For the given miniband, the energy is that of a two dimen- sional electron gas with the bottom of the band at E(q) in Eq. (1). Therefore the three dimensional density of states is given by [47] ν3D(E) = 1 2 dε νSL(ε) ν2D(E − ε) (cid:90) ∞ (cid:90) E −∞ = ν2D 2 −∞ dε νSL(ε). (2) The factor 1/2 is to avoid double counting the spin and ν2D = m/π2 is a two dimensional density of states per unit area. The density of states of the superlattice can be factorized into longitudinal and transverse parts, and the one dimensional superlattice density of states νSL is νSL(E) = 1 πβd (cid:113) Θ(E − E− z )Θ(−E + E+ z ) 1 − (−E+E0 )2 2β , (3) where we use a Heaviside step function Θ to show the range of the energy E with the maximum/miminum value (cid:104) (cid:104) dEz TrK1( Er z ) − TcK1( Ec z) , (6) dEzEz TrK1( Er z ) − TcK1( Ec z) (cid:105) 3 (cid:105) (cid:105)(cid:35) dEz r K2( Er T 2 z ) − T 2 c K2( Ec z) , (7) expressions of the electric and energy currents are Ir = Jr = eν2DAkB 2π ν2DAkB 2π z (cid:90) Er+ (cid:34)(cid:90) Er+ (cid:90) Er+ (cid:104) Er− Er− z z z z +k2 B Er− z FIG. 1. (Top) Schematic of the superlattice heat engine. A hot cavity at temperature Tc is coupled via superlattices to cold electronic reservoirs at temperature Tr. (Bottom) The periodic structure of the superlattices form the miniband cen- tered at E L/R with the width 4βL/R when we apply the bias voltage µR − µL = eV . The gray shading area shows the energy miniband where the electrons can transport between regions. The shadings in the source, cavity, and drain regions indicate thermal smearing. 0 z < E < E+ z = E0 ± 2β. Therefore, the electrons only in selected E± values of energy E− z will transport and gen- erate current. We assume the central cavity region is strongly coupled to the external source of energy, so the occupation is described as a Fermi function with local temperature Tc determined by the thermal reservoir. The electric and energy currents for simplified mini- band transport in z direction emitted by the reservoir r into an cavity c can be evaluated within a Landauer- Buttiker approach as Ir = dEzdE⊥[fr(E) − fc(E)], (4) z z (5) Er− Jr = dEzdE⊥ E [fr(E) − fc(E)], where A is the surface area of the superlattice, Er± is the maximum/minimum energy of the reservoir r, and we de- note E⊥ as the energy carried in the transverse degrees of freedom, and Ez as the energy carried in the longitu- dinal degree of freedom, so that E = E⊥ + Ez. Here, the square root in Eq. (3) cancels the velocity of the electron, vg, in the current Ir = (e/4)(cid:82) dEzvgν3D[fr(E) − fc(E)] grals K1(x) = (cid:82) ∞ K2(x) =(cid:82) ∞ arithm Li2(z) =(cid:80)∞ [48]. We see that the range of the integral comes from the density of states which gives a transmission function of flat box form, T r(Ez) = Θ(Ez − Er− z )Θ(−Ez + Er+ z ). To rewrite the above equations, we introduce the inte- 0 dt(1 + et−x)−1 = log(1 + ex) and 0 dtt(1+et−x)−1 = −Li2(−ex) with the dilog- zk k2 . Then the simplified analytic k=1 (cid:90) Er+ (cid:90) Er+ Er− z z z eν2DA 2π ν2DA 2π where Er z = (µr − Ez)/kBTr and Ec z = (µc − Ez)/kBTc. B. Results We now analyze the system by focusing on linear re- sponse and later turn to the nonlinear regime. To sim- plify the analysis of the system, we introduce the aver- age temperature T = (Tr + Tc)/2 and the temperature difference ∆T = Tc − Tr. For energy harvesting, the temperature difference is considered to be ∆T > 0. (If we consider refrigeration, the temperature difference is ∆T < 0.) We introduce the bias µR − µL = eV to the system by applying µR = eV /2 and µL = −eV /2. We also rewrite the width of each miniband, βL = β + α and βR = β − α where α, with α < β, determines the asym- metry between the left and right energy width, so the relative thickness of the left and right miniband width is determined by α. The chemical potential of the cavity µc as well as the temperature Tc are determined by imposing conservation of charge and energy, IL + IR = 0 and J + JL + JR = 0 where J is the heat current entering from the heat source. From these conservation laws, we can obtain the electric and heat currents through the system, as well as J and µc. 1. Linear response To linear order in the temperature difference ∆T and the bias voltage V , the net current flowing through the system, IL = −IR ≡ I, is given by I = GeV + GeSt∆T. (8) The electrical conductance Ge and thermopower (or See- beck coefficient) St of the system are Ge = − e2ν2DA 2π (cid:20) GL2 + GL3 GL1GR1 GL1 + GR1 , St = kB e − GR2 + GR3 GR1 GL1 (cid:21) (9) , (10) (cid:276) (cid:276) Cavity Source Drain Position Energy d x y z voltage, Pmax = Ge (St∆T )2 4 . 4 (14) In order to evaluate the efficiency η given by the ratio of output power to input heat current, we need to find the heat current J = −JL − JR injected from the heat bath, J = GeΠ V + (GeStΠ + Ht1 + Ht2)∆T , (15) which shows the Peltier effect. Here the coefficients Π, Ht1, and Ht2 are defined as (cid:21) (cid:20) (GR2 + GR3)2 GL1 , − GL2 + GL3 GR1 (cid:20) GR2 + GR3 kBT GR1 GL1 − GR1 GL1 + GR1 e ν2DA 2π k2 BT − (GL2 + GL3)2 GL1 − 4(GR2 + GR3)(GL2 + GL3) , Π = Ht1 = Ht2 = ν2DA 2π k2 +(GR4 + 2GR5 − GR6)] , BT [(GL4 + 2GL5 − GL6) together with the auxiliary functions z Er− (cid:90) Er+ (cid:90) Er+ (cid:90) Er+ Er− z z z z Er− z Gr4 = Gr5 = Gr6 = dEz (Ez/kBT )2 1 + eEz/kB T , dEz Ez kBT log(1 + e−Ez/kB T ), dEzLi2(−e−Ez/kB T ). (16) (cid:21) (17) (18) (19) (20) (21) Rewriting the heat current in Eq. (15) as J = ΠI + (Ht1 + Ht2)∆T , we see the meaning of the coefficients more clearly. The first part is simply proportional to the charge current, and Π shows the presence of heating from the Peltier contribution. The relation Π = −T St from Eq. (10) and Eq. (16) shows the coefficient Π can be considered as the back action counterpart to St, and the minus sign of the relation comes because J is the heat current injected from the heat bath into the cavity. This relation shows our system satisfies Onsager sym- metry resulting from the time reversibility, which relates the Seeback and Peltier coefficients [49]. The second part shows the heat generated by the temperature difference and the thermal conductance is Ht1 + Ht2. The first one, Ht1, contributes to the heat flow from the asymmetric superlattices and disappears when the left and right su- perlattices are symmetric. The heat current at maximum power takes the form (cid:20) GeStΠ 2 (cid:21) JmaxP = + Ht1 + Ht2 ∆T. (22) 2 2 (a), ( kB ∆T in units (b) Maximum power of FIG. 2. ν2DA(kB T ) )2 within linear response with respect to the centers of each miniband for a symmetric configuration, α = 0, when (a) is β = kBT and (b) is β = 5kBT . (c), (d) Ef- ficiency at maximum power normalized by Carnot efficiency ηc within linear response as a function of the centers of each miniband for symmetric setup. Panel (c)/(d) is a correspond- ing efficiency for the case of (a)/(b). All plots are obtained for T = 300K and ∆T = 1K. where we have introduced the auxiliary functions z Er− (cid:90) Er+ (cid:90) Er+ (cid:90) Er+ Er− z z z z Er− z Gr1 = Gr2 = Gr3 = dEz 1 1 + eEz/kB T dEz Ez/kBT 1 + eEz/kB T , , dEz log(1 + e−Ez/kB T ). (11) (12) (13) The electrical conductance shows that GL1(GR1) is pro- portional to the electrical conductance of the left(right) superlattice, so the net conductance Ge is simply the series combination of the two conductors. The three- terminal thermopower St is determined by the differ- ence between left and right two-terminal thermopower of each superlattice, and vanishes when they are identical [15, 18]. Therefore, depending on the magnitude of the left and right properties, St shows whether the system is analogous to a p-type or n-type semiconductor. When St is positive for (GL2 + GL3)/GL1 > (GR2 + GR3)/GR1, the system acts as if the mobile charge carrier is positive and behaves like a p-type semiconductor, and vice versa. The bias voltage V applied against heat driven charge current generates a finite output power P = IV . The power vanishes either when no bias voltage is applied or at the stopping voltage, Vstop = −St∆T . The output power takes its maximum value at half of the stopping (a) (b) (c) (d) 5 FIG. 4. The figure shows nonlinear response when the mini- band width is β = kBT . (a) the maximum output power as a function of temperature difference ∆T /T for the optimized parameters E L 0 = 2kBT , and α/β = −0.25. (b) shows the corresponding efficiency. The power is in the units of ν2DA(kB T )3 0 = −5kBT , E R . 2π appropriate value of the other center, and it comes with suppressed output power. Therefore, depending on what we want to optimize, the output power or efficiency, we can chose the position of band centers. We also show the maximum power for the different miniband widths from the top panels of Fig. 2. As the electrons transport only within the miniband, a wider miniband allows more electrons to transport and generate more power. However, as the miniband width goes too far above kBT , the power increase stops. The reason is that the energy window of the order kBTc/r will be a more effective energy guard for the carriers. Different from the power, efficiency is reduced as the miniband width increases, Eq. (3(c),(d)). As the width increases, the energy filtering by the superlattices is lesser efficient so the efficiency decreases. However, as the width increases continuously, the efficiency will saturate at some point with the same reason for the power saturation. These results show that the superlattice heat engine has more power output than the quantum well engine [15]. The maximal output power of the superlattice heat engine with units P0 = ν2DAkBT /(2(kB∆T /2)2) is Pmax ≈ 1.8P0 with the efficiency ηmaxP ≈ 0.05ηc (for the miniband width, β = kBT ), while the quantum well heat engine is Pmax ≈ P0 with the efficiency ηmaxP ≈ 0.07ηc. From Eq. (11-13), we can give a simple reason for this: when the miniband width is suppressed (longitudinal en- ergy window becomes one value and the electrons trans- port only with the certain longitudinal energy), our re- sults approach the quantum well case. Therefore, a large miniband width permits more electrons to transport in the longitudinal degree of freedom. While the superlat- tice engine is more powerful, it is less efficient an energy filter than the quantum well case: this is from the differ- ence between the miniband of the superlattice and the sub-band threshold of the quantum well. We now turn to an asymmetric system, α (cid:54)= 0. To find the optimized combination of the centers of mini- band and the asymmetric parameter α for the maximum output power, we consider the output power and the effi- (a), 2 2 ( kB ∆T in units (b) Maximum power of FIG. 3. ν2DA(kB T ) )2 within linear response as a function of the center of right miniband E R 0 and the asymmetric param- 0 = −4kBT , and (b) eter α/β, when (a) is β = kBT and E L is β = 5kBT and E L 0 = −20kBT . (c), (d) Efficiency at maxi- mum power normalized by Carnot efficiency ηc within linear response as a function of E R 0 and α/β. Panel (c)/(d) is a corresponding efficiency for the case of (a)/(b). All plots are obtained for T = 300K and ∆T = 1K. Therefore, the efficiency at maximum power is given by (cid:12)(cid:12)(cid:12)(cid:12) (cid:12)(cid:12)(cid:12)(cid:12) ηmaxP = ≈ GeS2 t 2GeStΠ + 4Ht1 + 4Ht2 GeS2 t T 2GeStΠ + 4Ht1 + 4Ht2 (cid:12)(cid:12)(cid:12)(cid:12) ∆T (cid:12)(cid:12)(cid:12)(cid:12) ηc, (23) where the second approximation comes from the Carnot efficiency for the small temperature difference, ηc = ∆T /Tc ≈ ∆T /T . Therefore, the combination of the coef- ficients, inside of the vertical bars in Eq. (23) determines the efficiency ratio to Carnot efficiency. 0 0 and E R Fig. 2 shows the maximum output power and the cor- responding efficiency on equal size bands (α = 0) as a function of the centers of minibands of the two superlat- tices, E L/R . Both the power and efficiency are symmetric with respect to an exchange of E L 0 . On the one hand, the maximum power arises when one of the two center value stays around twice of β and the other center is deep below the equilibrium chemical potential, below about −3β. The center around 2β is the position that makes the bottom of the miniband stay around the equi- librium chemical potential, and the other center which is below −3β makes the top of the miniband deep be- low the chemical potential, because the miniband width is 4β. On the other hand, the efficiency acts symmetri- cally depending on the position of band centers as well as the power. However, maximum efficiency comes when one of the center of miniband is bigger than 2β with an (a) (cid:9)(cid:67)(cid:10)(cid:1)(c) (cid:9)(cid:69)(cid:10)(cid:1)0.20.40.60.81.00.20.40.60.81.01.2\x{FFFF}a\x{FFFF}Pmax\x{FFFF}T\x{FFFF}T0.20.40.60.81.00.020.040.060.08\x{FFFF}b\x{FFFF}Η\x{FFFF}Ηc\x{FFFF}T\x{FFFF}Tciency for E L 0 = −4β. Fig. 3 shows plots of the power as a function of α and E R 0 . When β = kBT , the largest output power Pmax ≈ 1.85P0 (around a 2% increases from the symmetric case) arises for α ≈ −0.2β with E R 0 ≈ 2.6kBT . In this case, the efficiency goes to ηmaxP ≈ 0.06ηc (a 20% increase from the symmetric case). Meanwhile, the max- imum efficiency η ≈ 0.16ηc comes with α → −β and E R 0 → 5kBT for β = kBT . When α → −β, the miniband widths become βL → 0 and βR → 2β, and the output power is strongly suppressed. We can understand the reason for this last fact because when the width of the left miniband vanishes, transport from the left is cut off, giving no power, but very good energy filtering, which in- creases efficiency. To explain other features of the plots, we note that for the right superlattice, the center of mini- band stays slightly above the equilibrium chemical po- tential, and the width of miniband ∆ = 4βR is larger than the width of the distribution function ∼ kBT . Con- sequently, making the miniband wider has no effect on the electron transport because the number of electrons with energies around minimum/maximum of the energy window is exponentially small. Simultaneously, for the smaller miniband of the left barrier, centering it around the energy region where the left reservoir is occupied but the cavity not, gives very good current production. As the miniband width β increases, Fig. (3b), α moves to a positive value while the center E R stays around the value that maximizes the power for the symmetric case. Therefore, depending on the size of the miniband width relative to the width of the occupation function of the reservoir and the cavity, α can be fixed to give the opti- mized results. 0 2. Nonlinear response It is interesting next to consider the output power and the efficiency in the nonlinear regime, where qualitatively new physics can appear [50 -- 52]. We numerically calcu- lated the stopping voltage Vstop, the centers of miniband E L/R , and asymmetry parameter α in order to maximize 0 the output power. For the case β = kBT , we show the maximum output power and the efficiency in Fig. 4. This 0 = −5kBT and is obtained for the miniband centers E L 0 = 2kBT with the asymmetric parameter α = −0.25β. E R The output power increases quadratically in the temper- ature difference for the fixed average temperature T as in the linear case, Eq. (14), but it also depends on the average temperature, Pmax ∼ T ∆T 2. Therefore higher average temperature as well as the temperature difference gives bigger output power. In Table I, we compare the maximum output power and the corresponding efficiency for three systems: the quantum dot [9], quantum well [15], and superlattice based three terminal heat engines for a realistic device parameter mef f = 0.067me with the room temperature T = 300K and the temperature difference ∆T = 1K. We see that Pmax of the superlattice heat engine generates 6 Quantum Dots Quantum Wells Superlattices Pmax(W/cm2) ηmaxP (ηc) 0.1 0.2 0.18 0.07 0.3 0.06 TABLE I. We compare our result with other three terminal geometries: (power-optimized resonant width) quantum dot and quantum well heat engines for T = 300K and ∆T = 1K. a larger power about 0.3W/cm2 with a small reduction for the efficiency 0.06ηc. Therefore, the superlattice heat engine is the more powerful heat engine. C. Quantum refrigerator based on superlattices Now, we consider the same geometry but for a differ- ent purpose, a cooling system. If proper bias voltage eV is applied over the junction and the minibands of super- lattices are suitably arranged, a current flows from right to left as hot electrons tunnel through the miniband of the left superlattice from the cavity to left reservoir and cold electrons from the right reservoir tunnel through the right miniband to the cavity. This leads to decrease in the average energy of electrons in the central cavity, that is cooling utilizing the Peltier effect: in the Peltier effect, the junction is electrically biased and a produced heat current flow given an electric current. For the refrigerator, ∆T is negative in our notation, and the temperature of the center cavity cooled down to the amount of ∆T. The purpose of refrigeration is to achieve a large temperature reduction. The base temper- ature of the refrigerator is defined as the temperature for equilibrium where the evacuated heat current balances any external heat leaks. Since we are now considering no external heat leaks, such as electron-phonon coupling, the base temperature is the temperature for which J = 0 in Eq. (15) with the temperature reduction ∆T0 = − GeΠ GeStΠ + Ht1 + Ht2 V, (24) in the linear regime. The refrigeration works only when the heat current is emitted by the cavity into the reser- voirs, J ≥ 0, therefore the temperature reduction of the cavity should be in the range of ∆T0 < ∆T < 0. The applied bias voltage V and the absorbed heat from the cold cavity let the heat flow J rejected to the left and right reservoirs. Therefore, the cooling power is J, and an efficiency of the cooling is normally characterized by the coefficient of performance (COP), defined as the ratio of the cooling power to the total input power P = IV , φ = J P . (25) Similar to the efficiency of the energy harvester, COP is also bounded by the Carnot value, φ ≤ Tc Tr−Tc Moving beyond the linear regime is important to find the nonlinear behavior of the refrigerator, which deter- mines the lowest temperature it can reach. Fig. 5(a) = φc. 7 shown in the plot. When the temperature difference is anywhere between the zero and ∆Tmax, as an example the vertical line in the figure, the applied voltage should stay in between Vmin and Vmax to make the engine work as the refrigerator. Specially, when the applied voltage is around 2kBT , the refrigerator gives maximum cooling power for a given temperature difference, and we also can have the maxumum temperature reduction ∆T0 of the system. When ∆T → 0, the heat current increases linearly in panel (b). So, the maximum temperature re- duction for this bias voltage is ∆T0 = ∆Tmax (cid:39) 0.11T . For example, at room temperature, the maximum tem- perature reduction is ∆Tmax = 30K by applying the If the temperature reduction is bias around 50meV. ∆T0 = 0.05T = 15K, the corresponding cooling power is approximately 11kW/cm2. The COP at eV = 2kBT as a function of the temperature difference is also plotted in Fig. (5(b)). The maximum COP φ (cid:39) 0.015φc comes with when ∆T (cid:39) 0.06T . Therefore, the refrigerator can be operated at the optimal regime for the cooling power or COP by reasonably choosing the parameters. III. HEAT TRANSPORT BY PHONONS A. Reduced phonon heat current One important consideration in thermoelectric heat engine is the heat flow carried not only by the conduction electrons, but by the phonons as well. Here, we calcu- late this effect for our system to see how it will affect the efficiency in a more realistic modeling. The major heat flow from hot to cold reservoirs is car- ried by excitations such as phonons. For this three ter- minal heat engine, the phonon heat flow J ph is in parallel with electronic flow J e, and the total heat flow for a given generated power P is the sum of them, J = J e + J ph. Then we can rewrite the efficiency in the present of the phonons ηe+ph = P J e + J ph . (26) The phonon heat current in the superlattice differs from the heat current based on the bulk material prop- erties due to the new periodicity of the structure. The presence of interfaces can alter the phonon spectra re- sult from wave interference scattered at the interfaces. The formation of miniband gaps in superlattices due to phonon interference leads to a reduction of the phonon velocity vg(k) which gives a reduction of the thermal conductivity. To illustrate this effect in a simple model, we assume the complete separation of longitudinal and transverse vibrations for phonons with the wave vector to parallel to z axis. Also assuming perfect interfaces, the transverse momentum qx(y) is conserved. When we consider phonons propagating in the growth direction, this can be treated from a Kronig-Penney type approach for one dimensional atomic chain [34]. We suppose the 0 , E R FIG. 5. (a) The nonlinear cooling power is shown as a function of the temperature difference and the bias voltage when the other parameters, E L 0 , β, and α, are optimized to give the maximum cooling power. The cooling power is in the units of ν2DA(kB T )3 . Panel (b) shows the heat current and the COP as a function of the temperature difference when the bias is chosen to give the maximum cooling power, eV (cid:39) 2kBT . 2π shows a plot of the cooling power vs applied bias and tem- perature differences, when other variables, E R 0 = −5kBT , E L 0 = 2kBT , β = kBT and α = 0.3β, are optimized to give the largest cooling power for the positive applied bias V > 0. Here, α > 0 means that left miniband width is wider than the right miniband, and the reason can be understood as the energy harvesting that we discussed in section II B. Note that, the position of the left/right miniband of the refrigerator is opposite from the energy harvesting, so this makes the conduction electrons take energy from the cavity and continue to cool it. The black curve in panel (a) of Fig. (5) represents when the cooling power becomes zero, and the right hand side of the line (within the parabola) is for the positive cool- ing power which is the region that works for the refrig- eration. Therefore, we can see the optimal region for the temperature difference and the applied voltage, as (a) (b) 8 where ∆B((cid:126)q, Tr, Tc) = br((cid:126)q) − bc((cid:126)q) is the occu- pation difference between the reservoir and cavity with the Bose-Einstein distribution function br/c((cid:126)q) = (exp(ω((cid:126)q)/kBTr/c) − 1)−1 and T (qz) represents the transmission function which depends only on the longi- tudinal momentum qz. In our analysis, we assume the ballistic case with T (qz) (cid:39) 1. When we put the longi- tudinal group velocity Eq.(29) into the first equation of Eq. (30), we can factorize the heat current to the longi- tudinal (z) and transverse (x, y) directions J ph r = dqz 2π ωqz vg,1d(qz) ∆F (qz, Tr, Tc), (31) (cid:90) π/d 0 (cid:90) π/a −π/a with a modified occupation difference ∆F (qz, Tr, Tc) = dqxdqy (2π)2 ∆B((cid:126)q, Tr, Tc). (32) This function gives the effective phonon occupation dif- ference per unit area between the cavity and reservoir due to the transverse momentum. The transverse momen- tum of phonons contributes only through the distribution function, and the energy and velocity dependencies of the heat current are only from the longitudinal momentum qz. This analysis suggests that the heat current can be treated as a one dimensional calculation with an effective occupation function. We rewrite the new expression in terms of an integral over longitudinal frequency instead of the momentum space J ph r = = (cid:90) ω+ (cid:90) ω+ − ω i i i − ω i n1+n2(cid:88) n1+n2(cid:88) i=1 i=1 −(+) i dωqz vg,1d 2 D1D(ωqz )ωqz ∆F (ωqz , Tr, Tc) dωqz 2π ωqz ∆F (ωqz , Tr, Tc), (33) where ω is the minimum (maximum) frequency of the ith miniband of longitudinal momentum from Eq. (27). The conversion to frequency space introduces the phonon density of state of the superlattice, and we have D1D(ωqz ) = (vg,1dπ)−1. Therefore, the group velocity and the density of states cancel out and we have a simple equation of the phonon heat current in Eq. (33). When we consider for the electron transport of the su- perlattice, note that only the lowest miniband was con- sidered, because the energy gap of the first and second miniband is larger than the thermal energy kBT , for the typical materials of the superlattice. So, when the chem- ical potentials of the reservoir and the cavity stay around the first miniband, energy window kBT of the occupation will guide the transport only through the first miniband. Therefore, it is a reasonable approximation to ignore the higher minibands. In contrast, the Bose-Einstein distri- bution function in the frequency domain is broader than the maximum frequency of the acoustic dispersion, and so all of the acoustic phonons contribute to the heat trans- port. FIG. 6. The phonon dispersion curves from a one dimensional atomic chain for the case of mass ratio δ = 2.6 in the extended zone representation. (a) and (b) show n1 = n2 = 1 and n1 = n2 = 3, respectively. The dimensionless parameters are defined as q ≡ qa and ω ≡ ω(cid:112)M1/g. same monolayer spacings a and the magnitude of atomic constants g in between all the atoms. The layer one(two) has n1(n2) atoms with mass M1(M2 = δM1), and the thickness of sublattice one (two) is d1 = n1a (d2 = n2a) which gives the length of unit period d = d1 + d2. Then the characteristic equation is cos(qzd) = cos(k1d1) cos(k2d2) − 1 − cos(k1a) cos(k2a) sin(k1a) sin(k2a) sin(k1d1) sin(k2d2), (27) where k1(2) is the phonon wave vector of each layer 1(2) with cos(k1(2)a) = 1 − M1(2)ω2/2g, and qz is a longitu- dinal superlattice wave vector. This model gives us the dispersion of the longitudinally polarized phonons for the cross plane transport with zero transverse momentum. The dispersions for different periods are shown in Fig. 6 for n1 = n2 = (1, 3) with the mass ratio δ = 2.6. Increas- ing the superlattice periods (n1 and n2) gives more band folding and decreases the average group velocity. At nonzero transverse momentum, the longitudinally polarized phonon mode for the cross plane dynamics is described as [36] ω2 q = ω2 qz + Ω2 t 1 2 (2 − cos(qxa) − cos(qya)), (28) where ωqz is a solution from Eq. (27) and Ωt is a charac- teristic frequency of the material. Therefore, the disper- sion of non zero transverse momentum can be used to cal- culate the group velocity in the superlattices, (cid:126)vg = ∇(cid:126)q ωq. Since we are dealing with the heat transfer only to the cross plane direction, the group velocity that we need is only the z direction, ωqz ωq vg,1d, vg,z = (29) where we use the one dimensional group velocity vg,1d ≡ ∂ωqz /∂qz. Now, we can write the phonon heat current from the reservoir r into the cavity through the superlat- tice, J ph r = d3(cid:126)q (2π)3 ωq vg,z((cid:126)q) T (qz) ∆B((cid:126)q, Tr, Tc), (30) (cid:90) (a) (b) 9 FIG. 8. Efficiency at maximum power of GaAs/AlAs super- lattices for the value of ∆T = 1K, β = kBT , and n1 = n2 = 2, and E R/L are chosen to maximize power. The efficiencies by the electron heat current (a) and the total heat current (b) are shown as a function of average temperature. 0 (cid:113) t (q2 well as only consider the long wavelength modes, so the dispersion relation may be approximated as, ω ≈ y)/4. Computing the effective dis- qz + a2Ω2 ω2 x + q2 tribution by integrating over qx, qy, we find, F (ωqz) ≈ ωqzkBT 2πa2(Ωt/4)2 e−ωqz/kB T . (34) This allows us to approximate the heat current per unit area from the first mini-band as, J ph ≈ ≈ 4 (πaΩt)2 [(kBTL)4 − (kBTc)4] (πaΩt)2 (kBT )3∆T, 16 (35) (36) where the last limit is in the linear response limit. This is consistent with a Debye treatment of the phonon trans- port. This result shows that for low temperatures, the phonons freeze out, and the energy is predominately car- ried by the electrons. Moving on to the high temperature limit, we consider a numerical investigation. This limit is quite impor- tant for room temperature applications in mind. Fig. 7 shows the heat currents for GaAs/AlAs superlattices when ∆T = 1K, as an example. The phonon heat currents of superlattices and bulk material in panel (a) increase with temperature increasing, but saturate at high temperature. Moreover, as the superlattice atomic FIG. 7. The plots show electron and phonon heat currents of GaAs/AlAs superlattice for the case of the maximum output power with optimized parameters with ∆T = 1K. (a) shows the heat current by phonons J ph when the atomic layers are n1 = n2 = (1, 2), with the bulk AlAs case. (b) shows the magnitude comparison between J e and J ph. The total phonon heat current from the heat engine is also obtained from the heat conservation J ph + J ph L + J ph R = 0. However, different from the heat current by the electrons, the difference of the phonon heat current through the left/right reservoir is only determined by the temperature difference between the cavity and reservoirs which is contained only in the modified occupation differ- ence ∆F (ωqz , Tr, Tc). The temperatures of the left/right reservoir are the same in our case, therefore, the phonon heat current conservation leads to the total phonon heat current such as J ph = −2J ph R only if there is symmetric heat conductance. In order to calculate the effect of the phonons on the thermodynamic efficiency of this engine, we consider a low temperature regime, where an analytic investigation can be made, followed by a numerical investigation of the high temperature regime. In the low temperature case, we assume the tem- perature only excites low energy modes. Up to a small numerical factor, this permits us to approxi- mate the Bose-Einstein as a Boltzmann distribution, as (a) (b) 100200300400T\x{FFFF}K\x{FFFF}0.04920.04940.04960.04980.0500\x{FFFF}a\x{FFFF}Ηe\x{FFFF}Ηc50100150200250300T\x{FFFF}K\x{FFFF}0.0010.0020.0030.0040.005\x{FFFF}b\x{FFFF}Η\x{FFFF}Ηclayer increases, the phonon heat current decreases and saturates when the atomic layers are n1, n2 > 10 [38]. The practical superlattices usually have the atomic lay- ers n1, n2 > 10, so, the practical phonon heat current of the superlattices has about 40% reduction of the heat current from the bulk material around the room tem- perature. Panel (b) compares the electron and phonon heat currents for the optimized parameters to give the maximum output power. As the temperature increases, the electron heat current increases quadratically while the phonon heat current saturates, so the electron heat current reaches the phonon heat current at high temper- ature. However, at room temperature, the phonon heat current is still an order of magnitude higher than the electron heat current. We compare the efficiency by the electron heat current and the total heat current in Fig. 8. The phonon heat current dominates over the electronic one, J ph (cid:29) J e, when the temperature is around room temperature. In this situation, we can rewrite the total efficiency ηe+ph (cid:39) P/J ph. (37) The efficiency depends only on the phonon heat current and the generated power. Therefore, we have the maxi- mal efficiency when the power is maximal, and the power and efficiency reach maximal value together. Because the phonon heat current is an order of magnitude higher than the electron heat current, the total efficiency decreases about an order of magnitude. B. Optimized condition for minimum phonon heat current Practically, there could be a way to reduce the phonon heat current more than we showed based on the sim- ple theory and example. Based on the one dimensional atomic chain model, we have calculated the phonon heat current in the growth direction and estimated their con- tributions to the efficiency in the heat engine based on GaAs/AlAs superlattices. Our approach shows that the reduction of the phonon heat current comes from the re- duction of the group velocity near the folded Brillouin zone edges. This calculation also assumes the ballistic transport T (qz) (cid:39) 1. However, as the constituent lay- ers become thicker than the phonon mean free path, the phonon transmission function needs to be modified as T (qz) < 1. In this case, we also need to treat the phonons as particles and use the theory such as the Boltzmann transport equation [37]. Moreover, other mechanisms such as phonon spectra mismatch and scattering aris- ing from the roughness of the layer interfaces also play 10 an important role to understand the reduction of the ex- perimental results [38, 53 -- 56]. Some of high figure of merit thermoelectric materials show reduced lattice thermal conductivity in the super- lattice structures, for example Si/Ge or Bi2Te3/Sb2Te3. The reduction of phonon heat current can be maximized by the proper choice of superlattice period compared to the mean free path of the phonons: a theory predicts the thermal conductivity minimum as a function of layer spacing [39], and some works show the minimum thermal conductivity depending on the superlattice period and a ratio of the layer thickness for the materials [40, 41, 57]. Therefore, these materials, instead of our example, with optimized superlattice period will give more reduction of phonon heat current of the system. IV. CONCLUSIONS In this paper, we investigated a heat engine by thermo- electric effects in superlattice structures in three termi- nal geometry. First, our work considers the engine as a energy harvester, and shows the advantages of superlat- tice heat engine in the large output power compared to the other similar geometry due to the box shaped trans- mission function which comes from the electron energy miniband. Our theory predicts that the maximum power under optimized conditions can be larger than similar resonant tunneling devices, with comparable efficiency at maximum power. Second, a different regime of the system parameters makes the engine works as a refriger- ator, and we shows the optimized regime of the param- eters for the maximum cooling power and coefficient of performance. In addition, we analyzed the phonon heat current to find the total efficiency by the performance of electrons and phonons together. The reduction of phonon heat current in the superlattice compared to the corre- sponding bulk material offers higher total efficiency. The trade off between power and efficiency is overcome by the reduction of the phonon heat current which can be achieved either by operating at low temperatures, or by engineering the system to have low phonon conductivity, while keeping high electron conductivity. Easy fabrica- tion of these devices with advantages for both the power and efficiency show this heat engine is a promising device for next-generation thermoelectrics. ACKNOWLEDGMENTS We would like to thank Antonio Badolato for suggest- ing this line of research and for discussions. We thank Jian-Hua Jiang, and Paul Ampadu for discussions, and Bjorn Sothmann and Rafa S´anchez for helpful comments on the manuscript. This work is dedicated to the memory of Markus Buttiker, a mentor, colleague, and friend. 11 [1] L. D. Hicks and M. S. Dresselhaus, Phys. Rev. B 47, [28] F. Hartmann, P. Pfeffer, S. Hofling, M. Kamp, and L. 12727 (1993). [2] L. D. Hicks and M. S. Dresselhaus, Phys. Rev. B 47, 16631 (1993). Worschech, Phys. Rev. Lett. 114, 146805 (2015). [29] F. J. DiSalvo, Science 285, 703 (1999). [30] T. C. Harman, P. J. Taylor, M. P. Walsh, and B. E. [3] G. D. Mahan and J. O. Sofo, Proc. Natl Acad. Sci. USA LaForge, Science 297, 2229 (2002). 93, 7436 (1996). [31] L. Chena, J. Lia, F. Suna, and C. Wub, Appl. Energy [4] C. W. J. Beenakker and A. A. M. Staring, Phys. Rev. B 82, 300 (2005). 46, 9667 (1992). [5] T. E. Humphrey, R. Newbury, R. P. Taylor, and H. Linke, Phys. Rev. Lett. 89, 116801 (2002). [6] B. Kubala, J. Konig, and J. Pekola, Phys. Rev. Lett. 100, 066801 (2008). [32] J. Yu and H. Zhao, J. Power Sources 172, 428 (2007). [33] L. E. Bell, Science 321, 1457 (2008). [34] C. Colvard, T. A. Gant, M. V. Klein, R. Merlin, R. Fis- cher, H. Morkoc, and A. C. Gossard, Phys. Rev. B 31, 2080 (1985). [7] G. Billings, A. D. Stone, and Y. Alhassid, Phys. Rev. B [35] S. Tamura, D. C. Hurley, and J. P. Wolfe, Phys. Rev. B 81, 205303 (2010). 38, 1427 (1988). [8] N. Nakpathomkun, H. Q. Xu, and H. Linke, Phys. Rev. [36] P. Hyldgaard and G. D. Mahan, Phys. Rev. B 56, 10754 B 82, 235428 (2010). (1997). [9] A. N. Jordan, B. Sothmann, R. S´anchez, and M. Buttiker, Phys. Rev. B 87, 075312 (2013). [37] G. Chen, Phys. Rev. B 57, 14958 (1998). [38] S. Tamura, Y. Tanaka, and H. J. Maris, Phys. Rev. B [10] D. M. Kennes and V. Meden, Phys. Rev. B 87, 075130 60, 2627 (1999). (2013). [39] M. V. Simkin and G. D. Mahan, Phys. Rev. Lett. 84, [11] I. Weymann and J. Barna´s, Phys. Rev. B 88, 085313 927 (2000). (2013). [12] B. Sothmann, R. S´anchez, and A. N Jordan, Nanotech- nology 26, 032001 (2015). [40] R. Venkatasubramanian, Phys. Rev. B 61, 3091 (2000). [41] R. Venkatasubramanian, E. Siivola, T. Colpitts, and B. OQuinn, Nature 413, 597 (2001). [13] G. Benenti, G. Casati, T. Prosen, and K. Saito, [42] P. A. Lebwohl and R. Tsu, J. Appl. Phys. 41, 2664 arXiv:1311.4430. (1970). [14] B. Sothmann, R. S´anchez, A. N. Jordan, and M. [43] L. Esaki and L. L. Chang, Phys. Rev. Lett. 33, 495 Buttiker, Phys. Rev. B 85, 205301 (2012). (1974). [15] B. Sothmann, R. S´anchez, A. N. Jordan, and M. [44] D. L. Smith and C. Mailhiot, Rev. Mod. Phys. 62, 173 Buttiker, New J. Phys. 15, 095021 (2013). (1990). [16] A. Agarwal and B. Muralidharan, Appl. Phys. Lett. 105, 013104 (2014). [17] B. Sothmann, R. S´anchez, and A. N. Jordan, Europhys. Lett. 107, 47003 (2014). [45] A. Wacker, Phys. Reports 357, 1 (2002). [46] R. S. Whitney, Phys. Rev. Lett. 112, 130601 (2014). [47] J. H. Davies, The physics of low-dimensional semicon- ductors (Cambridge University Press, 2005). [18] R. S´anchez, B. Sothmann, A. N. Jordan, Phys. Rev. Lett. [48] The factor 1/4 comes from the conversion of the generic 114, 146801 (2015). k-space to energy space. [19] F. Mazza, S. Valentini, R. Bosisio, G. Benenti, V. Gio- vannetti, R. Fazio, F. Taddei, arXiv:1503.01601. [49] L. Onsager, Phys. Rev. 37, 405 (1931). [50] D. Sanchez and R. Lopez, Phys. Rev. Lett. 110, 026804 [20] R. S´anchez and M. Buttiker, Phys. Rev. B 83, 085428 (2013). (2011). [51] J. Meair and P. Jacquod, J. Phys.: Condens. Matter 25, [21] R. S´anchez, B. Sothmann, A. N. Jordan, and M. 082201 (2013). Buttiker, New J. Phys. 15, 125001 (2013). [22] J.-H. Jiang, J. Appl. Phys. 116, 194303 (2014). [23] F. Mazza, R. Bosisio, G. Benenti, V. Giovannetti, R. Fazio, and F. Taddei, New J. Phys. 16, 085001 (2014). [24] L. Henriet, A. N. Jordan, and K. Le Hur, arXiv:1504.02073 (2015). [25] J. R. Prance, C. G. Smith, J. P. Griffiths, S. J. Chorley, D. Anderson, G. A. C. Jones, I. Farrer, and D. A. Ritchie, Phys. Rev. Lett. 102, 146602 (2009). [26] H. L. Edwards, Q. Niu, G. A. Georgakis, and A. L. de Lozanne, Phys. Rev. B 52, 5714 (1995). [27] B. Roche, P. Roulleau, T. Jullien, Y. Jompol, I. Farrer, D.A. Ritchie, and D. C. Glattli, Nat. Commun, 6, 6738 (2015). [52] R. S. Whitney, Phys. Rev. B 88, 064302 (2013). [53] G. Chen, IEEE Trans. Compon. Packag. Technol. 29, 238 (2006). [54] D. G. Cahill, W. K. Ford, K. E. Goodson, G. D. Mahan, A. Majumdar, H. J. Maris, R. Merlin, and S. R. Phillpot, J. Appl. Phys. 93, 793 (2003). [55] M. J. Huang, W. Y. Chong, and T. M. Chang, J. Appl. Phys. 99, 114318 (2006). [56] S. Y. Ren and J. D. Dow, Phys. Rev. B 25, 3750 (1982). [57] C.-K. Liu, C-K. Yu, H.-C. Chien, S.-L. Kuo, C.-Y. Hsu, M.-J. Dai, G.-L. Luo, S.-C. Huang, and M.-J. Huang, J. Appl. Phys. 104, 114301 (2008). 

1509.08060

1

1509

Valley-dependent spin-orbit torques in two dimensional hexagonal crystals

We study spin-orbit torques in two dimensional hexagonal crystals such as graphene, silicene, germanene and stanene. The torque possesses two components, a field-like term due to inverse spin galvanic effect and an antidamping torque originating from Berry curvature in mixed spin-$k$ space. In the presence of staggered potential and exchange field, the valley degeneracy can be lifted and we obtain a valley-dependent Berry curvature, leading to a tunable antidamping torque by controlling the valley degree of freedom. The valley imbalance can be as high as 100\% by tuning the bias voltage or magnetization angle. These findings open new venues for the development of current-driven spin-orbit torques by structural design.

cond-mat.mes-hall

cond-mat

Valley-dependent spin-orbit torques in two dimensional hexagonal crystals Hang Li1, Xuhui Wang1,∗ and Aur´elien Manchon1† 1King Abdullah University of Science and Technology (KAUST), Physical Science and Engineering Division, Thuwal 23955-6900, Saudi Arabia (Dated: June 14, 2021) We study spin-orbit torques in two dimensional hexagonal crystals such as graphene, silicene, germanene and stanene. The torque possesses two components, a field-like term due to inverse spin galvanic effect and an antidamping torque originating from Berry curvature in mixed spin-k space. In the presence of staggered potential and exchange field, the valley degeneracy can be lifted and we obtain a valley-dependent Berry curvature, leading to a tunable antidamping torque by controlling the valley degree of freedom. The valley imbalance can be as high as 100% by tuning the bias voltage or magnetization angle. These findings open new venues for the development of current-driven spin-orbit torques by structural design. PACS numbers: 72.25.Dc,72.20.My,75.50.Pp I. INTRODUCTION Inverse spin galvanic effect (ISGE), referring to the electrical or optical generation of a nonequilibrium spin density in non-centrosymmetric materials, has attracted much attention over the years.1 -- 7 It originates from the momentum relaxation of carriers in an electrical field and their asymmetric redistribution in subbands that are spin-split by spin-orbit coupling.2 ISGE was first ob- served in bulk tellurium and soon generalized to low- dimensional structures such as GaAs quantum wells.3 -- 5 From an applied perspective, in ferromagnets lacking inversion symmetry ISGE enables the electrical control of the local magnetization through angular momentum transfer, a mechanism called spin-orbit torque (SOT).6,7 This effect has been scrutinized in dilute magnetic semi- conductors such as ferromagnetic bulk (Ga,Mn)As8 -- 11 and metallic multilayers comprising heavy metals and ferromagnets12 -- 17. These observations have been re- cently extended to bilayers involving topological insula- tors displaying extremely large SOT efficiencies18,19. We note that in metallic multilayers, spin Hall effect in the adjacent heavy metal also leads to a torque14 (see discus- sion in Ref. 20), which complicates the interpretation of the underlying physics. From a theoretical perspective, the torque stemming from ISGE on the magnetization M has the general form T = TDLM × (u × M) + TFLM × u, (1) where the first term is called the antidamping-like torque and the second term is referred to as the field-like torque21 -- 26. The antidamping-like torque is even in mag- netization direction and competes with the antidamping, while the field-like torque is odd in magnetization direc- tion and acts like a magnetic field. The vector u de- pends on the current direction j and the symmetries of the spin-orbit coupling. For instance, in a ferromagnetic two-dimensional electron gas (normal to z) with Rashba spin-orbit coupling, u = z × j.6 An interesting aspect of the formula given above is that the antidamping-like torque arises from the distortion of the wavefunction in- duced by the electric field, a mechanism closely related to the material(cid:48)s Berry curvature11,22,24,26. In parallel to the development of SOT in ferromag- netic structures, the study of spin-orbit coupled trans- port has also been extended to low-dimensional hexag- onal crystals such as graphene. Experimentally, a spin-splitting induced by Rashba spin-orbit coupling has been observed in graphene grown on heavy met- als or surface alloys.27 -- 29 Furthermore, a ferromagnetic insulator EuO was successfully deposited on graphene and spin-polarized states were detected.30 -- 32 The recent fabrication of low-dimensional hexagonal crystals with strong intrinsic spin-orbit coupling such as silicene33,34, germanene35 and possibly stanene36, has enriched the graphene physics. These materials offer a rich platform for the investigation of spin, orbital and valley-dependent phenomena37,38. In this paper, we theoretically investigate the na- ture of SOT in two-dimensional hexagonal IV group ele- ments crystals such as graphene, silicene, germanene and stanene. As a matter of fact, the wide tunability of their model band structure presents an appealing opportunity to study the impact of the band geometry (e.g. their Berry curvature) on nonequilibrium mechanisms. Using Kubo formula, we investigate the impact of the band structure on the different components of SOT. We find that intrinsic spin-orbit coupling affects the antidamping- like and field-like components differently. The former is sensitive to the presence of a staggered potential while the latter is not. We understand these results in terms of Berry curvature origin of the antidamping torque. The presence of both magnetization and staggered potential enables the emergence of a valley-dependent antidamp- ing torque, providing an additional degree of freedom to the system. 5 1 0 2 p e S 7 2 ] l l a h - s e m . t a m - d n o c [ 1 v 0 6 0 8 0 . 9 0 5 1 : v i X r a 2 However, in the presence of both staggered potential and ferromagnetic exchange field, the valley degeneracy is lifted since both the time-reversal and sublattice sym- metries are broken as shown in Fig. 1(d). As a result, SOT becomes valley dependent. Furthermore, as dis- cussed in the next section, the band structure distortion displayed in Figs. 1(b)-(d) affects the magnitude of the SOT components. We adopt a low-energy continuum model Hamiltonian which describes Dirac electrons near to the Fermi energy and captures the physics behind the formation of the valley-dependent SOT in the vicinity of K and K(cid:48) points. The total Hamiltonian at K or K(cid:48) valley in the basis of {ψA,↑, ψB,↓, ψB,↑, ψA,↓} reads44 Hsys = v(τ kx σx − ky σy) ⊗ I + (τ σx ⊗ sy − σy ⊗ sx) + τ λso σz ⊗ sz + JexI ⊗ M · s + U σz ⊗ I, (2) λR 2 √ where v = 3at/2 with t being a nearest-neighbor hop- ping parameter, τ = +1(−1) stands for the K or (K(cid:48)) valley, I is a 2× 2 unity matrix, a is the lattice constant and Jex is the ferromagnetic coupling constant. σ and s are Pauli matrices denoting the AB-sublattice and spin degrees of freedom, respectively. M is the magnetization direction. The first term includes the spin-independent kinetic energy of the particle, the second term denotes the Rashba coupling and the third one represents the intrin- sic spin-orbit coupling. The fourth term is the interaction between the spin of the carrier and the local moment of the ferromagnetic system. The last term is the staggered potential (induced, for instance, by an electrical field or a substrate45 -- 47), where U = 1 (−1) for A (B) site. To compute the current-induced effective magnetic field, we first evaluate the nonequilibrium spin density δS at K (K(cid:48)) valley using Kubo formula:11 δSK(K(cid:48)) = (cid:104)ψkbsψka(cid:105)(cid:104)ψkaE · v ψkb(cid:105) (cid:88) ka − GR k,a,b Re e 2πV × [GR kbGA kbGR ka], (3) ka = (GA where E is the electric field, v = 1 ∂H ∂k is the velocity op- ka)∗ = 1/(EF − Eka + iΓ). Γ = /2τ is erator, GR the disorder-induced energy spectral broadening due to the finite life time of the particle in the presence of im- purities and τ is the momentum scattering time. EF is the Fermi energy, Eka is the energy of electrons in band a. The eigenvector ψk,a(cid:105) in band a can be found by diagonalizing Eq. (2). Equation (2) contains both intra- band (a = b) and interband (a (cid:54)= b) contributions to the nonequilibrium spin density. Simpler expressions in the weak Γ limit can be found in Ref. 24. The former stems from the perturbation of the carrier distribution func- tion by the electric field and it is inversely proportional to Γ. The latter arises from the perturbation of the car- rier wave functions by the electric field. The interband contribution also depends on Γ but survives when Γ → 0. FIG. 1. (Color online) Schematics of the device based on graphene-like materials with field-like and antidamping-like SOT. (b)-(d) Energy dispersion of graphene-like materials with (b) U = 0.03 eV (c) M = 0.03 eV (d) U = 0.03 eV and M = 0.03 eV . The current flows from left to right. Mag- netization is assumed to be directed along the z-axis. II. MODEL AND METHOD A possible structure to realize valley-dependent SOT is a single-layered hexagonal lattice (such as graphene, silicene, germanene or stanene) sandwiched by a fer- romagnetic layer and a non-magnetic substrate [see Fig. 1(a)]. The ferromagnetic layer may be chosen as EuO,30 or YIG,39 which induces a weak exchange coupling on the spin-polarized carriers. The underly- ing non-magnetic substrate provides Rashba spin-orbit coupling.28,29,40,41 Note that in principle, a magnetic in- sulator could supply for both exchange field and Rashba spin-orbit coupling.19,42,43 The concept of valley-dependent SOT is illustrated in Fig. 1(a). In the absence of a magnet, the interaction between the substrate and graphene-like layer breaks the inversion symmetry and leads to a Rashba spin-orbit cou- pling. As a results, a transverse nonequilibrium spin den- sity builds up when a current is injected along the hor- izontal direction. Both Rashba and intrinsic spin-orbit coupling are valley dependent as shown in Eq. (2) and thus they can not break the valley degeneracy. In the presence of a magnet, a field-like spin density and an antidamping-like spin density are generated as shown in Fig. 1(a).13,14 The exchange field only breaks the time- reversal-symmetry while the sublattice symmetry (two- fold rotational symmetry in the plane) is preserved as shown in Fig. 1(c). The interaction between the sub- strate and graphene-like layer can also induce a staggered potential, which enlarges the band gap without affecting the valley degeneracy, as shown in Fig. 1(b). x y z Current M δSFL δSDL Magnet Substrate (b) (c) (d) (a) In order to evaluate the current-driven SOT in different materials, we define the electrical efficiency of the torque as6 3 η = 2JexδS σxxE where σij is conductivity tensor component defined48 σij = e2Re [(cid:104)ψkavi ψkb(cid:105)(cid:104)ψkbvj ψka(cid:105)] (cid:88) ka − GR k,a,b × [GR kbGA kbGR ka]. (4) (5) III. INVERSE SPIN GALVANIC EFFECT The characteristics of the SOT in two-dimensional hexagonal honeycomb lattices are expected to be differ- ent from the well studied case of bulk GaMnAs6,7,21 -- 24. Unlike the three-dimensional ferromagnetic GaMnAs in the weak limit (λR (cid:28) Jex, Jex ∼ 1eV and λR ∼ 0.1eV), the graphene-like materials often fall into the strong limit (λR (cid:29) Jex), leading to a nonzero interband contribution. The nontrivial Dirac kinetic term [first term in Eq. (2)] gives rise to nonlinear transitions of spin density when tuning the Fermi energy. Furthermore, the spin density is more sensitive to band topology tunable by intrinsic spin-orbit coupling or staggered potential. More impor- tantly, the Dirac kinetic term and spin-orbit coupling terms are valley-dependent. In order to better under- stand the valley-dependent SOT, we first examine spin torque with and without valley degeneracy in section III and IV respectively. A. Non-magnetic honeycomb lattice We first compute the spin density induced by ISGE in non-magnetic graphene. In this material, we choose the following parameters: Ef ∈ [0, 0.3] eV,49 λR ∈ [10, 130] meV,28,29 and Jex ∈ [5, 30] meV.50,51 For all the calculations shown in this section, the electrical field is assumed to be along the x-axis and the energy broaden- ing is Γ =0.01 eV. To understand the physical origin of the SOT and establish connections with previous works (such as Ref. 24), we parse the SOT into intraband and interband contributions. Figure 2 presents the intraband (a,c) and interband contributions (b,d) to the ISGE-driven spin density for various strengths of λR (a,b) and λso (c,d). In non- magnetic graphene, intraband contribution produces a spin density aligned toward the y-direction, which is ex- pected from the geometry of our system and consistent with the well known ISGE in two-dimensional electron gases6,52. There is also a non-ignorable interband con- tribution in the strong limit (λR (cid:29) Jex), smaller than the intraband contribution and opposite to it, in agree- ment with our previous analytical solutions in the case FIG. 2. (Color online) (a) Intraband and (b) interband con- tributions to spin density as a function of Fermi energy Ef for various Rashba spin-orbit coupling in the absence of intrinsic spin-orbit coupling λso. (c) Intraband and (d) interband spin density as a function of Fermi energy Ef for various intrinsic spin-orbit coupling at λR=0.03 eV. Inset (b): Band structure of graphene-like materials with λR=0.03 eV and λso=0 eV. Inset (c): Band structure with λR=0 eV and λso=0.03 eV In- set (d): Same as inset (b) but with λso=0.03 eV. The current is injected along the x axis. of Rashba two-dimensional electron gas24. When increas- ing the absolute value of Fermi energy, the spin density first experiences a sharp enhancement at small values of Ef and quickly saturates. This result is consistent with Ref. 53 and can be readily understood by considering the band structure in the inset of Fig. 2(b). When the Fermi energy lies in the energy gap of two spin-split sub- bands, only one spin species contributes to ISGE and the intraband spin density increases with the Fermi en- ergy. As the Fermi energy lies above the subband gap, the two subbands compensate each other and the spin density saturates. The peaks in Fig. 2(a) correspond to the minimum (E > 0) or maximum (E < 0) of the spin- up subband [see inset of Fig. 2(b)] which is of the order of λR. Another interesting feature is the spin density as a function of the Rashba spin-orbit coupling. The intra- band contribution increases linearly with λR [see Fig. 2(a)], while the interband contribution first increases and then decreases [see Fig. 2(b)]. The interband contribu- tion depends on the energy difference between the sub- bands, which itself is of the order of λR. Indeed, one can show that in the weak impurity limit, the interband contribution is proportional to 1/(Eka-Ekb).11,24,26 This results in the non-linear dependence as a function of λR observed in Fig. 2(b) as well as in Fig. 3(c). Rashba spin-orbit coupling is not the only spin-orbit coupling that affects the spin density. In graphene-like -0.10.00.10.2-0.4-0.20.00.20.4-0.02-0.010.000.010.020.03-0.10.00.10.2-0.4-0.20.00.20.4-0.02-0.010.000.010.020.03-0.20.00.2-0.12-0.060.000.060.12-0.20.00.2-0.12-0.060.000.060.12-0.20.00.2-0.12-0.060.000.060.12 sintraFL( eV-1nm-1)sintraFL( eV-1nm-1) R=0.01 R=0.03 R=0.05 R=0.07(a)SO= 0 eV (b)sinterFL( eV-1nm-1) Ef(eV)(c) so=0.00 so=0.01 so=0.03 so=0.05 R= 0.03 eV (d)sinterFL( eV-1nm-1) Ef(eV)E(eV)k(nm-1)E(eV)k(nm-1)E(eV)k(nm-1)systems Rashba spin-orbit coupling is always accompa- nied by an intrinsic spin-orbit coupling, ∼ τ λso σz ⊗ sz, which originates from the substrate or a low buckled structure.45,54In Figs. 2(c) and (d), we display the Fermi energy dependence of the intraband and interband con- tributions to spin density for various intrinsic spin-orbit coupling. As expected, the intrinsic spin-orbit coupling opens up a band gap and distorts the topology of the band structure as seen in the inset of Fig. 2(c) and (d). For a given K or K(cid:48) Valley (ignore τ ), this term plays the same role as the ferromagnetic exchange field along the z axis in unit cell when the two sublattices contribute to spin density equivalently (σz replaced by I ). When the two sublattices contribute to spin density inversely, this term acts as an anti-ferromagnetic exchange field and the symmetry of profiles of the spin density is broken and it shifts to the left. Furthermore, the asymmetry of the profiles of the spin density becomes more evident with the increase of λso. The energy at which the spin density is maximum equals λso + λR when Ef < 0. Note that the intrinsic spin-orbit coupling does not drive ISGE by itself, but it affects the ISGE-induced spin density driven by Rashba spin-orbit coupling through the modulation of the topology of the bands. B. Magnetic honeycomb lattice Let us now turn to the case of magnetic two- dimensional honeycomb lattices. To understand the role of spin-orbit coupling, we plot the intraband and inter- band spin density as a function of Rashba spin-orbit cou- pling for different intrinsic spin-orbit coupling in the pres- ence of magnetization in Fig. 3. Due to the presence of magnetism, the interband contribution also produces an antidamping component [see Figs. 3(c)], i.e. a spin den- sity contribution oriented towards ∼ M × y11,24,26 and with a magnitude comparable to the one of the field-like component [see Figs. 3(c)]. As seen in Figs. 3(a)-(c) the interband field-like and antidamping contributions first increase and then decrease. This can be understood as a competition between the spin density driven by Rashba spin-orbit coupling and the suppression of interband scat- tering due to the distance between the subbands that increases with λR. The intraband contribution decreases with the increas- ing intrinsic spin-orbit coupling while the interband con- tribution behaves the opposite way. By opening a band gap, the intrinsic spin-orbit coupling alters the band fill- ing, resulting in a reduced intraband contribution to spin density. An analytical solution of energy depending on intrinsic spin-orbit coupling can be found in Ref. 55. On the other hand, the intrinsic spin-orbit coupling reduces the splitting between the subbands for Ef > 0 [see in- set in Fig. 2(d)], which results in an enhancement of the interband contributions. This result is valuable to current-driven magnetic excitations since the antidamp- ing torque is responsible for magnetization switching and excitations13,14 (see also, for instance, Ref. 56). 4 FIG. 3. (Color online)(a) Intraband and (b)-(c) interband spin density as a function of Rashba spin-orbit coupling for different intrinsic spin-orbit coupling with Jex = 0.01 eV and Ef = 0.1 eV . The magnetization is directed along the z axis. IV. VALLEY-DEPENDENT SPIN-ORBIT TORQUE The valley degree of freedom can be used as a tool to enhance the functionality of two-dimensional hon- eycomb lattices.57 Recently, a valley-dependent anoma- lous quantum hall state has been predicted in silicene and silicene nanoribbons owing to the topological phase transition45,58. A charge-neutral Hall effect has been measured in graphene devices59,60. These suggest the emergence of valley Hall effect. It is thus natural to ex- pect a valley-modulated SOT in our settings. A. Staggered Potential The sublattice degeneracy can be removed by de- positing graphene-like materials on hexagonal boron- nitride59,61,62 or silicon carbide,47 or by applying an elec- trical field in a low buckled structure45. When the stag- gered potential and exchange field are present and the valley degeneracy is broken, the spin density becomes valley-dependent as shown in Figs. 4. In Figs. 4(a)-(c), we display the intraband and in- terband contributions to spin density as a function of 0.000.030.060.090.120.150.00-0.01-0.02-0.030.00-0.08-0.16-0.240.000.010.020.03 sinterDL( eV-1nm-1)(c)(a) so=0.00 so=0.01 so=0.03 so=0.05sintraFL( eV-1nm-1)(b)sinterFL( eV-1nm-1)R(eV)Fermi energy in the presence of staggered potential with and without the intrinsic spin-orbit coupling. The im- balance between the contribution of the two valleys to the spin density, i.e., valley polarization, defined as P = δSK−δSK(cid:48) δSK +δSK(cid:48) , is reported on Figs.4(d)-(f). The largest imbalance occurs mainly around the neutrality point Ef = 0. The valley imbalance of the antidamping- like component can reach 100% as shown in Figs.4(f), i.e., that for certain energies, this component is domi- nated by only one valley. When the intrinsic spin-orbit coupling is present, the magnitudes of the valley imbal- ance can be switched from -100% to 100% by simply tun- ing the Fermi energy. FIG. 4. (Color online) (a) Intraband and (b)-(c) interband spin density of two valleys as a function of Fermi energy for different intrinsic spin-orbit coupling with U = 0.01 eV and Jex = 0.01 eV . Valley polarization for intraband (d) and interband (e)-(f) components for different intrinsic spin-orbit coupling. 5 FIG. 5. (Color online) Intraband and interband spin density as a function of the magnetization direction with (solid lines) and without (dashed lines)staggered potential for the different valleys when U = 0.03 eV. Inset (b) Valley polarization of interband spin density for x-component. relaxation63 or in the intermediate regime (λR ∼ Jex) due to a "breathing" Fermi surface26. Similarly, in the case of magnetic honeycomb lattices, different components of the spin density display a clear deviation from the simple ∼ cos θ dependence of the fer- romagnetic Rashba gas (see dotted lines in Fig. 5). This is attributed to the "breathing" Fermi surface, i.e., the distortion of the Fermi surface, and the modification of the band filling as a function of the direction of the magnetization when the exchange is comparable to the Rashba parameters. In the absence of valley degeneracy, the angular depen- dence at K and K' points differ significantly from each other (red and blue lines in Fig. 5, respectively). As a consequence, by tuning the magnetization angle the val- ley imbalance varies strongly [from -100% to 100% for the x-component, as shown in inset of in Fig. 5(b)]. We also notice that additional structures are visible in the angular dependence of the field-like component, related to interband transitions [see Fig. 5(c)]. These features are unique to the case of honeycomb lattices and absent in standard two dimensional free electron gases. B. Angular dependence V. CONNECTION BETWEEN SPIN-ORBIT TORQUE AND BERRY CURVATURE A noticeable effect of lifting the valley degeneracy is its impact on the angular dependence of SOT compo- nents. Figure 5 displays the angular dependence of the different components of the spin density when the mag- netization is rotated in the (x,z) plane. In a ferromag- netic two-dimensional electrons gas with Rashba spin- orbit coupling, the spin density has the general form δS = δS(cid:107) cos θx + δSym × y − sin θδS(cid:107)z (e.g., see Ref. 24), where θ is the angle between the magnetization and z. More complex angular dependence may appear in the strong Rashba limit (λR (cid:29) Jex) due to D(cid:48)yakonov-Perel Berry's phase plays a crucial role in the transport properties of semiconductors especially for graphene-like materials. Due to the inequivalent contribution from two valleys, Berry curvature induces valley hall effect in graphene with broken inversion symmetry.59 Recently, the link between SOT and Berry curvature was estab- lished in bulk ferromagnetic GaMnAs.11 The intrinsic spin-orbit coupling distorts the Fermi surface and gives rise to the oscillations in torque magnitudes, as already observed in (Ga,Mn)As11. -0.20.00.2-0.02-0.010.000.01-0.10-0.050.000.050.10-100-50050100-0.050.000.050.10-100-50050-0.02-0.010.000.010.02-50050sinterDL( eV-1nm-1) K valley K valley K' valley K' valley pinterFLpintraFL(c)pinterDL so=0.0 so=0.01(f)(a)sintraFL( eV-1nm-1) so=0.0 so=0.01 so=0.0 so=0.01 (d)Ef (eV)(b)sinterFL( eV-1nm-1)Ef (eV) (e)0306090120150180-0.04-0.020.000.020.040.080.100.120.14-0.036-0.032-0.02804590135180-100-5005010003060901201501800.000-0.004-0.008-0.012 sinterz( eV-1nm-1)sintery( eV-1nm-1)(b)sinterx( eV-1nm-1) K' valley K valley without U(a)sintray( eV-1nm-1) (c)pinterxdegree(d)degree degree6 FL and δSinter at K and K(cid:48) valley are displayed in Fig. 6. We find that both the field-like intraband and interband contributions to the spin density, δSintra FL . They increase with the intrinsic spin-orbit coupling and are only weakly affected by the staggered potential [see Figs. 6 (a,b) and In contrast, the antidamping-like component of (d,e)]. the spin density, δSinter DL , displays a non-linear dependence as a function of the intrinsic spin-orbit coupling that is very different for the two valleys and highly sensitive to the staggered potential [see Figs. 6 (c) and (f)]. To understand this difference, we plot the contour of Berry curvature for different intrinsic spin-orbit coupling at K and K(cid:48) valleys in kx − ky plane in Fig. 7. A large Berry curvature mainly concentrates around the Dirac point and decays away from it, in agreement with pre- vious results44,64. For the K valley, Berry curvature de- creases with the increase of intrinsic spin-orbit coupling. Yet for the K(cid:48) valley, Berry curvature increases. This trend is in accordance with the variations of δSinter DL dis- played in Figs. 6 (c) and (f) and not in accordance with the variations of δSinter FL displayed in Figs. 6 (b) and (e). It illustrates the fact that while δSinter DL both originate from interband transitions, only the latter is re- lated to Berry curvature, i.e., the field-like SOT is purely due to ISGE instead of the superposition of Berry curva- ture and ISGE in ferromagnetic GaMnAs as pointed out by Kurebayashi et al.11 and δSinter FL FIG. 6. (Color online) Intraband and interband spin density as a function of intrinsic spin-orbit coupling for different stag- gered potential for the K valley (a)-(c) and K(cid:48) valley (d)-(f). The parameters are: Ef = −0.16 eV , and Jex = 0.01 eV and λR = 0.03 eV . VI. DISCUSSION To complete the present study, we computed the mag- nitude of antidamping-like and field-like components of the spin density and corresponding electrical efficiencies for various graphene-like honeycomb lattices, assuming λR = 0.1 eV and Jex = 0.03 eV. The results are re- ported in Table 1, showing that the largest SOT is ob- tained for stanene [∼ 100 × 1010 eV/(A · m)]. As a com- parison, the corresponding efficiencies of field-like SOT in (Ga,Mn)As24, two-dimensional Rashba systems6 and topological insulators65 are of the order of ∼ 1 × 1010, ∼ 10 × 1010 and ∼ 100 × 1010 eV/(A · m), respec- tively, in agreement with the orders of experimental results8,10,13,19. Therefore, for moderate Rashba and ex- change parameters, honeycomb lattices seem to display large field-like torques. Interestingly, the antidamping- like torque remains about one order of magnitude smaller than the field-like torque, as already observed in two- dimensional Rashba gases and (Ga,Mn)As24. Finally, we propose a device to detect the valley- dependent SOT. We consider a multi-terminal device as shown in Fig. 8. This is a typical device used to de- tect charge neutral-currents59,60. The device consists of a graphene-like material sandwiched between a magnetic insulator and a non-magnetic substrate such as a topolog- ical insulator40. The substrate67 can induce a staggered potential that breaks the valley degeneracy. The voltage is applied to the sidearms and the current flows from the FIG. 7. (Color online) Contour of valley-polarized Berry cur- vature distribution for different intrinsic spin-orbit coupling in kx−ky plane with U = 0.03 eV. (a) and (d) λso = 0.005 eV. (b) and (e) λso = 0.01 eV. (c) and (f) λso = 0.015 eV. Others parameters are the same as in Fig.6. In order to show the connection between the SOT and the band structure distortion, let us analyze the influence of intrinsic spin-orbit coupling on SOT in the presence of a staggered potential. The intraband and interband contributions to spin density as a function of intrinsic spin-orbit coupling for various staggered potential both 0.000.030.060.09-0.0120.0000.0120.000.030.060.090.120.0040.0080.0120.0160.060.090.120.150.180.060.090.120.150.18-0.015-0.010-0.0050.000-0.015-0.010-0.0050.000sinterDL( eV-1nm-1)(c)sinterDL( eV-1nm-1) (f) U = 0.01 eV U = 0.02 eV U = 0.03 eV(a)sintraFL( eV-1nm-1)sintraFL( eV-1nm-1)sinterFL( eV-1nm-1)(d) so(eV)(b)sinterFL( eV-1nm-1)so(eV)K' valley(e) K valley-0.20.00.2-20.00020.0040.0060.0080.00100.0110.0ky(nm-1)(a)so=0.005-0.20.00.2K' valley(d)ky(nm-1) K valley-0.20.00.2so=0.010(b)ky(nm-1)-0.20.00.2(e)ky(nm-1) -110.0-55.00055.00110.0-0.20.00.2-0.20.00.2so=0.015(c)ky(nm-1)kx(nm-1)-0.20.00.2-0.20.00.2(f)ky(nm-1)kx(nm-1)TABLE I. Efficiency of spin torque for various two dimensional hexagonal lattices 7 Carbon E(eV) 44 2.7 Silicene 1.04 Germanene 0.97 Stanene 0.76 66 66 66 a(A) 54 2.46 54 3.87 54 4.06 54 4.67 σxx(e2/) 23.3809 × 10−3 9.0068 × 10−3 8.4004 × 10−3 6.5818 × 10−3 0.0083 0.0137 sDL((eV nm)−1) sF L((eV nm)−1) ηDL(eV (Am) 2.13 × 1010 9.13 × 1010 10.07 × 1010 14.13 × 1010 0.1193 0.2019 0.0141 0.0155 −1) −1) ηF L(eV (Am) 30.6 × 1010 131.6 × 1010 144.2 × 1010 200.6 × 1010 0.1975 0.220 lower sidearm to the upper one. In the absence of mag- netization, a valley Hall effect may be detected in the two horizontal terminals.59 In the presence of magnetization, the torque exerted on the magnetization of the magnetic insulator deposited on top of the left or right terminal will be different. FIG. 8. Schematics of the realization of valley-dependent antidamping-like SOT: (a) Top view and (b) Side view. The current is injected into the vertical arm. The presence of both magnetization and staggered potential results in a nonequiv- alent spin density for the valleys. This leads to a different valleys on the horizontal sidearms. VII. CONCLUSION In summary, we have investigated the nature of SOTs in two dimensional hexagonal crystals and qualitatively recovered most of the results obtained on different sys- tems such as (Ga,Mn)As and two-dimensional Rashba gases24. We showed that the staggered potential and intrinsic spin-orbit coupling can strongly affect the mag- nitude of the torque components as well as their angular dependence. In the presence of staggered potential and exchange field, the valley degeneracy can be lifted and we obtain a valley-dependent antidamping SOT, while the field-like component remains mostly unaffected. This feature is understood in terms of Berry curvature and we show that the valley imbalance can be as high as 100% by tuning the bias voltage or magnetization angle. ACKNOWLEDGEMENT H.L. and A.M. were supported by the King Abdullah University of Science and Technology (KAUST). ∗ xuhuiwangnl@gmail.com † Aurelien.Manchon@kaust.edu.sa + V -V M M Graphene-like Substrate (a) (b) K+ K- Current 1 E.L. Ivchenko, G.E. Pikus, JETP Lett. 27, 604 (1978). 2 E. L. Ivchenko and S. Ganichev, in Spin Physics in Semi- conductors, edited by M. I. Dyakonov (Springer, New York, 2008). 3 L.E. Vorobev, E.L. Ivchenko, G.E. Pikus, I.I. Farbstein, V.A. Shalygin, A.V. Sturbin, JETP Lett. 29, 441 (1979). 4 A.Yu. Silov, P.A. Blajnov, J.H. Wolter, R. Hey, K.H. Ploog, N.S. Averkiev, Appl. Phys. Lett. 85, 5929 (2004). 5 S.D. Ganichev, S.N. Danilov, P. Schneider, V.V. Belkov, L.E. Golub, W. Wegscheider, D.Weiss,W. Prettl, J. Magn. Magn. Mater. 300, 127 (2006). 6 A. Manchon and S. Zhang, Phys. Rev. B 78, 212405 (2008); Phys. Rev. B 79, 094422 (2009). 7 I. Garate and A. H. MacDonald, Phys. Rev. B 80, 134403 (2009); A. Matos-Abiague and R. L. Rodriguez-Suarez, Phys. Rev. B 80, 094424 (2009). 8 A. Chernyshov, M. Overby, X. Liu, J. K. Furdyna, Y. Lyanda-Geller, and L. P. Rokhinson, Nat. Phys. 5, 656 (2009). 9 M. Endo, F. Matsukura, and H. Ohno, Appl. Phys. Lett. 97, 222501 (2010). 10 D. Fang, H. Kurebayashi, J. Wunderlich, K. V´yborn´y, L. P. Zarbo, R. P. Campion, A. Casiraghi, B. L. Gallagher, T. Jungwirth, and A. J. Ferguson, Nat. Nanotechnol. 6, 413 (2011). 11 H. Kurebayashi, J. Sinova, D. Fang, A. C. Irvine, J. Wun- derlich, V. Novak, R. P. Campion, B. L. Gallagher, E. K. Vehstedt, L. P. Zarbo, K. Vyborny, A. J. Ferguson, and T. Jungwirth, Nat. Nanotechnol. 9, 211 (2014). 12 K. Garello, I. M. Miron, C. O. Avci, F. Freimuth, Y. Mokrousov, S. Blugel, S. Auffret, O. Boulle, G. Gaudin, and P. Gambardella, Nat. Nanotechnol. 8, 587 (2013). 13 I. M. Miron et al. , Nat. Mater. 9, 230 (2010); U. H. Pi et al. , Appl. Phys. Lett. 97, 162507 (2010); T. Suzuki, S. Fukami, N. Ishiwata, M. Yamanouchi, S. Ikeda, N. Kasai, and H. Ohno, Appl. Phys. Lett. 98, 142505 (2011); I. M. Miron, et al. , Nature 476, 189 (2011). 14 L. Liu et al. , Phys. Rev. Lett. 109, 096602 (2012); L. Liu et al. , Science 336, 555 (2012). 15 J. Kim, J. Sinha, M. Hayashi, M. Yamanouchi, S. Fukami, T. Suzuki, S. Mitani and H. Ohno, Nat. Mater. 12, 240 (2013). 16 X. Fan, J. Wu, Y. Chen, M. J. Jerry, H. Zhang and J. Q. Xiao, Nat. Commun. 4, 1799 (2013). 17 M. Jamali, K. Narayanapillai, X. Qiu, L. M. Loong, A. Manchon, and H. Yang, Phys. Rev. Lett. 111, 246602 (2013). 18 A. R. Mellnik, J. S. Lee, A. Richardella, J. L. Grab, P. J. Mintun, M. H. Fischer, A. Vaezi, A. Manchon, E. A. Kim, N. Samarth, and D. C. Ralph, Nature 511, 449 (2014). 19 Y. Fan, P. Upadhyaya, X. Kou, M. Lang, S. Takei, Z. Wang, J. Tang, L. He, L. T. Chang, M. Montazeri, G. Yu, W. Jiang, T. Nie, R. N. Schwartz, Y. Tserkovnyak, and K. L. Wang, Nat. Mater. 13, 699 (2014). 20 A. Brataas, A. D. Kent and H. Ohno, Nat. Mater. 11, 372 (2012). 8 and A. Manchon, Phys. Rev. B 91, 134402 (2015). 25 X. Wang and A. Manchon, Phys. Rev. Lett. 108, 117201 (2012). 26 K. Lee, D. Go, A. Manchon, P. M. Haney, M. D. Stiles, H. Lee, and K. Lee, Phys. Rev. B 91, 144401(2015). 27 A. Varykhalov, J. Sanchez-Barriga, A. M. Shikin, C. Biswas, E. Vescovo, A. Rybkin, D. Marchenko, and O. Rader, Phys. Rev. Lett. 101, 157601 (2008). 28 D. Marchenko, A. Varykhalov, M. Scholz, G. Bihlmayer, E. Rashba, A. Rybkin, A. Shikin, and O. Rader, Nat. Com- mun. 3,1232 (2012). 29 P. Leicht, J. Tesch, S. Bouvron, F. Blumenschein, P. Er- ler, L. Gragnaniello, and M. Fonin, Phys. Rev. B 90, 241406(R) (2014). 30 A. G. Swartz, P. M. Odenthal, Y. Hao, R. S. Ruoff, R. K. Kawakami, ACS Nano 6, 10063 (2012). 31 J. Klinkhammer, and M. Schlipf, and F. Craes, and S. Runte, and T. Michely, and C. Busse, Phys. Rev. Lett. 112, 016803 (2014). 32 J. Klinkhammer, D. Forster, S. Schumacher, Hans P. Oepen, T. Michely, and C. Busse, Appl. Phys. Lett. 103, 131601 (2013). 33 A. Fleurence, R. Friedlein, T. Ozaki, H. Kawai, Y. Wang, and Y. Yamada-Takamura, Phys. Rev. Lett. 108, 245501 (2012). 34 P. Vogt, P. Padova, C. Quaresima, J. Avila, E. Frantzeskakis, M. Asensio, A. Resta, B. Ealet, and G. Lay, Phys. Rev. Lett. 108, 155501 (2012). 35 L. Li, S. Lu, J. Pan, Z. Qin, Y. Wang, Y. Wang, G.Y. Cao, S. Du, and H. Gao, Adv. Mater. 26, 4820 (2014). 36 Y. Xu, B. Yan, H. Zhang, J. Wang, G. Xu, P. Tang, W. Duan, and S. Zhang, Phys. Rev. Lett. 111, 136804 (2013). 37 W. Han, R. K. Kawakami, M. Gmitra and J. Fabian, Nat. Nanotechnol. 9, 794 (2014). 38 A. Manchon, H.C. Koo, J. Nitta, S.M. Frolov, R.A. Duine, Nat. Mater. 14, 871 (2015). 39 Z. Wang, C. Tang, R. Sachs, Y. Barlas, and J. Shi, arxiv 1412.1521(2014). 40 K. Jin and S. Jhi, Phys. Rev. B 87, 075442 (2013). 41 C. Chang et al., Science 340, 167 (2013). 42 Y. Zou and Y. Wang, Nanoscale 3, 2615 (2011). 43 Z. Qiao, W. Ren, H. Chen, L. Bellaiche, Z. Zhang, A.H. MacDonald, and Q. Niu, Phys. Rev. Lett. 112, 116404 (2014). 44 Z. Qiao, S. A. Yang, W. Feng, W. Tse, J. Ding, Y. Yao, J. Wang, and Q. Niu, Phys. Rev. B 82, 161414(R) (2010). 45 M. Ezawa, Phys. Rev. Lett. 109, 055502 (2012). 46 C. L. Kane and E. J. Mele, Phys. Rev. Lett. 95, 146802 (2005). G. Li, A. Luican, D. Abanin, L. Levitov, and E. Y. Andrei, Nat. Commun. 4, 1744 (2013). 47 S. Y. Zhou, G. H. Gweon, A. V. Fedorov, P. N. First, W. A. de Heer, D. H. Lee, F. Guinea, A. H. Castro Neto, and A. Lanzara, Nat. Mater. 6, 770 (2007). 48 Y. Murayama, Mesoscopic Systems: Fundamentals and Applications (Wiley-VCH, Weinheim, Germany, 2001). 49 D. Pesin and A. H. MacDonald, Nat. Mater. 11, 409 21 D. A. Pesin and A. H. MacDonald, Phys. Rev. B 86 014416 (2012). (2012). 50 H. Haugen, D. Huertas-Hernando, and A. Brataas, Phys. 22 E. van der Bijl and R. A. Duine, Phys. Rev. B 86, 094406 Rev. B 77, 115406 (2008). (2012). 23 H. Li, X. Wang, F. Dogan, and A. Manchon, Appl. Phys. Lett. 102, 192411 (2013). 24 H. Li, H. Gao, L. P. Zarbo, K.V. Vyborny, X. Wang, I. Garate, F. Dogan, A. Cejchan, J. Sinova, T. Jungwirth, 51 H. X. Yang, A. Hallal, D. Terrade, X. Waintal, S. Roche, and M. Chshiev, Phys. Rev. Lett. 110, 046603 (2013); 52 V. M. Edelstein, Solid State Commun. 73, 233 (1990). 53 A. Dyrdal, J. Barnas, and V. K. Dugaev, Phys. Rev. B 89, 075422 (2014). 9 54 C. Liu, H. Jiang, and Y. Yao, Phys. Rev. B 84, 195430 (2011). 55 Z. Qiao, H. Jiang, X. Li, Y. Yao and Q. Niu, Phys. Rev. B 85, 115439 (2012). 56 V. E. Demidov, S. Urazhdin, H. Ulrichs, V. Tiberkevich, A. Slavin, D. Baither, G. Schmitz and S. O. Demokritov, Nat. Mater. 11, 1028 (2012). 57 A. Rycerz, J. Tworzyd and C. W. J. Beenakker, Nat. Phys. 3, 172 (2007). 58 H. Pan, Z. Li, C. C. Liu, G. Zhu, Z. Qiao, and Y. Yao, Phys. Rev. Lett. 112, 106802 (2014). 59 R. V. Gorbachev, J.C.W. Song, G.L. Yu, A.V. Kretinin, F. Withers, Y. Cao, A. Mishchenko, I.V. Grigorieva, K.S. Novoselov, L.S. Levitov, and A.K. Geim, Science 346, 448 (2014). 60 D. A. Abanin, R. V. Gorbachev, K. S. Novoselov, A. K. Geim, and L. S. Levitov, Phys. Rev. Lett. 107, 096601 (2011). 61 H. Min, G. Borghi, M. Polini, and A. H. MacDonald, Phys.Rev.B 77, 041407(R) (2008); J. E. Drut and T. A. Lahde, Phys. Rev. B 79, 165425 (2009);Y. Araki, Phys. Rev. B 84, 113402 (2011). 62 B. Hunt, J.D. Sanchez-Yamagishi, A.F. Young, M. Yankowitz, B.J. LeRoy, K. Watanabe, T. Taniguchi, P. Moon, M. Koshino, P. Jarillo-Herrero, and R.C. Ashoori, Science 340, 1427 (2013). 63 C. Ortiz Pauyac, X. Wang, M. Chshiev, and A. Manchon, Appl. Phys. Lett. 102, 252403 (2013). 64 D. Xiao, W. Yao, and Q. Niu, Phys. Rev. Lett. 99, 236809 (2007); D. Xiao, M. Chang, and Q. Niu, Rev. Mod. Phys. 82, 1959 (2010). 65 A. Sakai and H. Kohno, Phys. Rev. B 89, 165307 (2014). 66 W. F. Tsai, C. Y. Huang, T. R. Chang, H. Lin, H. T. Jeng, and A. Bansil, Nat. Commun. 4, 1500 (2013). 67 G. Giovannetti, P. A. Khomyakov, G. Brocks, P. J. Kelly, and J. van den Brink, Phys. Rev. B 76, 073103 (2007); X. Zhong, Y. K. Yap, R. Pandey, and S. P. Karna, Phys. Rev. B 83, 193403 (2011);B. Sachs, T. O. Wehling, M. I. Katsnelson, and A. I. Lichtenstein, Phys. Rev. B 84, 195414(2011). 

0909.2465

2

0909

Quantum spin Hall effect and spin-charge separation in a kagome lattice

A two-dimensional kagome lattice is theoretically investigated within a simple tight-binding model, which includes the nearest neighbor hopping term and the intrinsic spin-orbit interaction between the next nearest neighbors. By using the topological winding properties of the spin-edge states on the complex-energy Riemann surface, the spin Hall conductance is obtained to be quantized as $-e/2\pi$ ($e/2\pi$) in insulating phases. This result keeps consistent with the numerical linear-response calculation and the \textbf{Z}$_{2}$ topological invariance analysis. When the sample boundaries are connected in twist, by which two defects with $\pi$ flux are introduced, we obtain the spin-charge separated solitons at 1/3 (or 2/3) filling.

cond-mat.mes-hall

cond-mat

Quantum spin Hall effect and spin-charge separation in a kagom´e lattice Zhigang Wang1 and Ping Zhang1, 2, ∗ 1LCP, Institute of Applied Physics and Computational Mathematics, P.O. Box 8009, Beijing 100088, People's Republic of China 2Center for Applied Physics and Technology, Peking University, Beijing 100871, People's Republic of China (Dated: October 30, 2018) Abstract A two-dimensional kagom´e lattice is theoretically investigated within a simple tight-binding model, which includes the nearest neighbor hopping term and the intrinsic spin-orbit interaction between the next nearest neighbors. By using the topological winding properties of the spin- edge states on the complex-energy Riemann surface, the spin Hall conductance is obtained to be quantized as −e/2π (e/2π) in insulating phases. This result keeps consistent with the numerical linear-response calculation and the Z2 topological invariance analysis. When the sample boundaries are connected in twist, by which two defects with π flux are introduced, we obtain the spin-charge separated solitons at 1/3 (or 2/3) filling. PACS numbers: 73.43.-f, 71.10.Pm, 72.25.Hg 1 1 0 2 g u A 2 ] l l a h - s e m . t a m - d n o c [ 2 v 5 6 4 2 . 9 0 9 0 : v i X r a ∗Corresponding author. Email address: zhang ping@iapcm.ac.cn 1 Over the last two decades the topological band insulators (TBIs) have been a subject of great interest in condensed matter field [1, 2]. Different from the normal band insulators, the TBIs have a prominent feature, which is the necessary presence of gapless edge states on the sample boundaries [3, 4]. An early TBI model was proposed by Haldane [5]. Therein it was shown that the gapless edge states result in a remarkable character of TBIs by showing quantum Hall effect in the absence of an external magnetic field. Besides the Haldane model, several other lattice models have also been proposed to be quantum Hall TBIs, which include the two-dimensional (2D) [6 -- 8] and three-dimensional (3D) [9] spin-chiral kagom´e lattices, and the 3D distorted fcc lattice [10, 11]. All the quantum Hall TBIs rely on the breaking of time-reversal symmetry (TRS). Recently, Kane and Mele generalized the spinless Haldane model to a spin one by adding an intrinsic spin-orbit interaction (SOI) [12, 13]. The TRS is conserved in the Kane-Mele model and the gapless spin edge states in this model result in quantum spin Hall effect. These TBIs like the Kane-Mele model keep TRS and are different from the quantum Hall TBIs. Thus call them the quantum spin Hall TBIs. At present the quantum spin Hall TBIs are receiving considerable attention. One impressive example is that only one year later after its theoretical prediction to be a quantum spin Hall TBI in 2006 [14], it was proved in experiment that HgTe is an actual one [15]. One special character of the quantum spin Hall TBIs was recently attributed to their spin-charge separated excitations [16, 17] in the presence of a π flux. This attribution is motivated by the recent advance in studying 2D fractionalized quasiparticles [18, 19], and is a straightforward result when, like what Kane and Mele [12] have dealt with Haldane's TBI, considering spins of the 2D edge soliton. The separate spinon, holon and chargeon obey Bose statistics, and the experimental measurement of these soliton excitations would provide an undoubted verification of the Z2 topological properties of the quantum spin Hall TBIs. At present, besides the necessity for further studies to gain more insights into the nature of spin-charge separation and its connection to the other topological phenomena, obviously, it is also important to identify and study various model systems that exhibit the phenomenon of spin-charge separation. Motivated by this observation, as well as by the recent attention on the layered metal oxides as possible candidates for the quantum spin Hall TBI [20], in this paper, we study the quantum spin Hall effect and construct spin-charge separated edge solitons in a 2D kagom´e lattice. Different from the previously studied kagom´e TBI 2 A B C FIG. 1: (Color online). Schematic picture of the 2D kagom´e lattice. The dashed lines represent the Wigner-Seitz unit cell, which contains three independent sites (A, B, C). [6 -- 9, 21, 22] whearein the presence of ferromagnetic spin chirality breaks TRS, here TRS persists and the quantum spin Hall effect occurs due to the intrinsic SOI. Experimentally, the physical candidates for realizing our studied system might be 5d transition metal oxides with layered pyrochlore structure [23 -- 25]. The argument is that in 5d transition metal oxides, both the SOI and the electron correlation become important with the same order of magnitude. As a consequence, at high temparture the correlation-induced magnetic order can be overcome by SOI and the nontrivial topological insulator phase is expected to occur [20, 26 -- 29]. The other alternative way to experimentally realize our studied system is by modulating the 2D electron with a periodic potential with kagom´e symmetry, as recently demonstrated for artificial graphene [30]. By using the bulk linear-response theory, as well as the topological winding numbers of the spin-edge states on the complex-energy Riemann surface, we obtain the spin Hall conductance (SHC) σs xy is −e/2π (e/2π) at 1/3 (2/3) filling. Then, we construct spin-charge separated edge solitons by introducing π fluxes with a method similar to that in Ref. [18]. The quantum statistics xy. The quantized value of σs of these solitons is also discussed. Consider the tight-binding model for independent electrons on a 2D kagom´e lattice (Fig. 1). The spin-independent part of the Hamiltonian is given by H0 = t X hijiσ c† iσcjσ, (1) where tij=t is the hopping amplitude between the nearest-neighbor link hiji and c† iσ (ciσ) is the creation (annihilation) operator of an electron with spin σ (up or down) on lattice site i. For simplicity, we choose t=−1 as the energy unit and the distance a between the nearest 3 sites as the length unit throughout this paper. The Hamiltonian (1) can be diagonalized in the momentum space as H0 = X k ψ+ kσH0(k)ψkσ, (2) where the electron field operator ψkσ=(cAkσ, cBkσ, cC kσ)T includes the three lattice sites (A, B, C) in the Wigner-Seitz unit cell shown in Fig. 1. H0(k) is a 3×3 spinless matrix given by H0(k) = −t 0   2 cos (k·a1) 2 cos (k·a3) 2 cos (k·a2) 0 0 2 cos (k·a1) 2 cos (k·a3) 2 cos (k·a2)   , (3) where a1=(−1/2,−√3/2), a2=(1, 0), and a3=(−1/2,√3/2) represent the displacements in a unit cell from A to B site, from B to C site, and from C to A site, respectively. In this notation, the first Brillouin zone (BZ) is a hexagon with the corners of K=± (2π/3) a1, ± (2π/3) a2, ± (2π/3) a3. The energy spectrum for spinless Hamiltonian H0(k) is characterized by one dispersion- 1k =2), which reflects the fact that the 2D kagom´e lattice is a line graph less flat band (ǫ(0) 2(3)k = −1∓√4bk − 3 with of the honeycomb structure [32], and two dispersive bands, ǫ(0) bk=P3 i=1 cos2 (k · ai). These two dispersive bands touch at the corners (K-points) of the BZ 2(3)k=(−1∓√3k−K), which implies a "particle- and exhibit Dirac-type energy spectra, ǫ(0) hole" symmetry with respect to the Fermi energy ǫF = − 1. The corresponding eigenstates of H0(k) are given by where the expressions of the components qik and the normalized factor Gn(k) for each band nkE = Gnk (q1k, q2k, q3k)T , u(0) (cid:12)(cid:12)(cid:12) (4) are given in Table I. TABLE I: The expressions for the coefficients in Eq. (4) with xi=k·ai. q1k q2k q3k G−2 nk 1 2 [ǫ(0)2 nk − 4 cos2 x2] ǫ(0) nk cos x1 + 2 cos x2 cos x3 ǫ(0) nk cos x3 + 2 cos x2 cos x1 2bkǫ(0)2 nk + [4bk − 3ǫ(0)2 nk ] cos2 x2 + 6(bk − 1)ǫ(0) nk 4 Then, we introduce the intrinsic SOI term, which, according to the symmetry of the kagom´e lattice, takes the form [12, 33] HSO = i 2λSO√3 X hhijiiσ1σ2 (cid:0)d1 ij × d2 ij(cid:1) · sσ1σ2c† iσ1cjσ2. (5) Here λSO represents the SOI strength, s is the vector of Pauli spin matrices, i and j are next-nearest neighbors, and d1 ij and d2 ij are the vectors along the two bonds that connect i to j. Taking the Fourier transform, we have HSO=X kσ ψ+ kσHSO(k)ψkσ with HSO(k) = ±2λSO   0 −i cos(k·b1) i cos(k·b3) −i cos(k·b2) 0 i cos(k·b1) −i cos(k·b3) i cos(k·b2) 0   , (6) where b1=a3 − a2, b2=a1 − a3, b3=a2 − a1, and the +(−) sign refers to spin up (down) electrons. Inclusion of the intrinsic SOI in the Hamiltonian makes the appearance of the eigenstates unki and eigenenergies ǫnk very tedious with exception at some high-symmetry k points. In- stead of writing their explicit forms, here we show in Fig. 2(a) (solid curves) the numerically calculated energy spectrum for the total Hamiltonian H=H0+HSO along the high-symmetry lines (Γ→K, K→M, and M→Γ) in the BZ. The SOI coefficient is chosen to be λSO=0.1t. For comparison we also plot in Fig. 2(a) (dashed curves) the energy spectrum in the absence of SOI (λSO=0). One can see that while the spin degeneracy is not lifted by the presence of the intrinsic SOI, nevertheless, the Dirac contacts of the lower and middle bands at the K point are removed and a gap of amplitude ∆=4√3λSO opens between these two bands. The amplitude of this gap turns out to be ∆=4√3λSO. Similarly, the original contact at the Γ point between the middle and upper (flat) bands is also lifted by the presence of SOI and a gap with amplitude δ is opened. However, this gap is an indirect one, i.e., the middle-band maximum and upper-band minimum are not at the same k-point. To see the behavior of the system in the insulating state, we have calculated the SHC 5 (a) 2 0 k -2 -4 (c) 2 ) / 4 e ( s 0 -2 (b) K- K M -4 -2 0 F M K+ 2 FIG. 2: (Color online). (a) Energy spectrum (solid curves) of the 2D kagom´e lattice along the high-symmetry lines in the BZ with intrinsic spin-orbit couplings λSO=0.1t. There are two band gaps appearing with gap width ∆ and δ. For comparison, we also draw the energy spectrum without intrinsic spin-orbit couplings (dotted lines). (b) The corresponding BZ of the 2D kagom´e lattice. (c) The SHC σs xy as a function of the Fermi energy ǫF . The shaded areas correspond to the bulk gaps. using the following Kubo formula [34] xy = −eℏ X σs Imhunk 1 n6=n′,k × [f (ǫnk) − f (ǫn′k)] (7) 2 {vx, sz}un′kihun′kvyunki (ǫnk − ǫn′k)2 + η2 , where v (k)=∂H(k)/ℏ∂k and H(k)=H0(k) + HSO(k). The calculated result at zero temper- ature is shown in Fig. 2(c) by varying the Fermi energy ǫF with the SOI coefficient λSO=0.1t. From Fig. 2(c), one can see that initially the SHC σs xy decreases as the filling factor of the (spin-degenerate) lower band increases, arriving at the minimum value −(e/2π) at ǫF =−1.35 (=−1−∆/2), a value corresponding to the top of the lower band. Then, as the Fermi en- ergy ǫF continues to vary in the first gap region (shaded area wherein −1.356ǫF 6 − 0.65), the SHC keeps this minimum value unchanged. As shown in the following discussion, this quantized SHC can be understood by the Z2-valued topological invariant associated with this quantum spin Hall phase [13]. When the Fermi energy increases to touch the bottom of 6 the middle band at ǫF =−0.65(=−1 + ∆/2), then the SHC suddenly switches up and rapidly increases when the Fermi energy goes through the middle two bands. When ǫF increases to be at the top of the middle two bands (ǫF =1.65), σs xy arrives at the maximum value e/2π. Then as the Fermi energy ǫF continues to vary in the second gap region (shaded area 1.656ǫF 62), the SHC σs xy keeps this maximum value unchanged. When the Fermi energy increases to touch the bottom of the upper band at ǫF =2, the SHC then suddenly switches down and rapidly decreases when the Fermi energy goes through the upper band. Finally the SHC σs xy decreases to disappear when the three spin-degenerate bulk bands are all fully occupied. Topologically, the quantum spin Hall insulating state at 1/3 or 2/3 filling can be seen by calculating a selective Z2-valued invariant ν [13], which is related to the parity eigenvalues ξ2m(Γi) of the 2m-th occupied energy band at the four time-reversal-invariant momenta Γi [35]. Very recently, Guo and Franz [33] have numerically calculated the eigenstate of HΓi and they found that three ξ's are positive and one is negative. Although which of the four ξ's is negative depends on the choice of the inversion center, the product Πiξ(Γi)=(−1)ν is independent of this choice and determines the nontrivial Z2 invariant ν=1. That confirms the kagom´e lattice system to be a quantum spin Hall TBI at 1/3 (or 2/3) filling. On the other hand, the topological aspect of the quantized spin Hall phase can be distin- guished by the difference between the winding numbers of the spin-up and spin-down edge states across the holes of the complex-energy Riemann surface, Is=I↑ − I↓ [31]. The SHC is then given by σs xy=Is (e/4π). Using this topological index Is we have investigated the quan- tum spin Hall effect in the Kane-Mele graphene model [31]. For the present kagom´e lattice model we can also study the quantum spin Hall effect in terms of this topological winding index. For this purpose, let us first numerically diagonalize the total Hamiltonian H using the strip geometry. For convenience and without loss of generality, we suppose the system has two edges in the y direction while keeping infinite in the x direction [see Fig. 3(a)]. The number of sites A (or B, C) in the y direction is chosen to be Ny=40. The calculated energy spectrum is drawn in Fig. 3(b). From this figure one can clearly see that there are spin edge states occurring in each energy gap. These gapless edge states in the truncated kagom´e lattice are topologically stable against random-potential perturbation, provided that the perturbation is small compared to the bulk gaps. We have numerically confirmed this fact. The Riemann surface of Bloch function is plotted in Fig. 3(c) for 1/3 filling. According to 7 FIG. 3: (Color online). (a) Sketch of the 2D kagom´e lattice strip with two edges along the y direction. The red, blue, and green dots are used to distinguish the independent sites A, B, and C, respectively. (b) Energy spectrum of the kagom´e lattice strip (Ny=40) with the Hamiltonian H=H0+HSO. The spin-orbit coupling strength is set as λSO=0.1t. The red and blue lines represent the edge states localized at the down and up edges of the system, respectively. And the circle and triangle label the up and down spins, respectively. (c) The Riemann surface of the Bloch function corresponding to 1/3 filling. The purple and orange curves correspond to spin-up and spin-down channels, respectively. Ref. [31], the winding number of spin-up (spin-down) edge state in the lower gap is I↑=−1 (I↓=1), which gives Is=−2 at 1/3 filling. That means the SHC in this phase is quantized as σs xy=−(e/2π). The Riemann surface of Bloch function at 2/3 filling are the same as that at 1/3 filling, except that the directions of the curves corresponding to different spin channels are inverse. So the winding number of spin-up (spin-down) edge state in the upper gap is I↑=1 (I↓=−1), which gives Is=2. The corresponding SHC at 2/3 filling is then quantized as σs xy=e/2π. These conclusions are consistent with those calculated by using the Kubo formula (7) [see Fig. 2(c)]. Note that although the non-trivial Z2 invariant ν=1 in the bulk analysis can confirm the quantum spin Hall TBI phase, it does not provide the information on the sign of the SHC. In contrast, our winding-number analysis can resolve this sign at different fillings (i.e., σs xy=∓(e/2π) at 1/3 and 2/3 filling, respectively). 8  0 0.02 0.1 0.2 ) 8 / e ( s 4 0 -4 -4 -2 0 F 2 FIG. 4: (Color online). The SHC as a function of the Fermi energy ǫF with different values of the Rashba coefficient. The black, red, blue, and pink lines correspond to λR=0, 0.02t, 0.1t, and 0.2t, respectively. The intrinsic SOI strength is set as λSO=0.1t. The topological properties of the system in the insulating state are stable even when the Rashba SOI is considered. To clearly see this fact, we numerically calculated the SHC σs xy with the Kubo formula (7) when the total Hamiltonian includes the Rashba SOI [36] HR = i λR ℏ X hijiσ1σ2 c† iσ1(s × dij)zcjσ2, (8) where λR is the Rashba coefficient and dij is a vector along the bond the electron traverses going from site j to i. The calculated SHC is drawn in Fig. 4 for different values of λR. Clearly, one can see that the amplitude of SHC in the insulating phase keeps e/2π unchanged when the Rashba SOI strength 06λR/t<0.1. When the Rashba SOI is sufficiently large (for example λR=0.2t), the amplitude of SHC will depart little from the quantized value. However, the topology of this insulating state keeps unchanged, unless the bulk gaps disappear. In the following let us consider the spin-charge separation in the 2D kagom´e lattice. If we reconnect the two edges of the kagom´e lattice shown in Fig. 3(a) with weaker bonds, two small gaps then reappear in the edge spectrum [see Figs. 5(a) and (b)]. Topological excitations (edge solitons) are created by reversing the sign of the reconnected bonds along the right half row [see Fig. 5(c)]. In this manner two defects with π flux are introduced in the present kagom´e lattice. As a result, four degenerate in-gap spin states localized around these two defects are formed in each bulk gap [see Figs. 5(d)]. The corresponding energies of these in-gap states are ǫ=−t and 1.7t, respectively. Note that in the square lattice version of the Kane-Mele model [16], the in-gap modes are precisely at zero energy, while in the 9 FIG. 5: (Color online). (a) Two edges of the 2D kagom´e lattice are reconnected with weaker bonds (orange dotted lines) and its corresponding energy spectrum (b). (c) The restored bonds have a sign reversal along the right half bonds (violet dotted lines) and its corresponding energy spectrum (d). In these figures, the system size is set as Ny=40 and the spin-orbit coupling λSO=0.1t. The bonds connected two edges are set as 0.25 times of other ones. present kagom´e lattice model the in-gap modes are no longer at zero energy. Here, we would like to point out that the zero level discussed in a large amount of previous studies is a result of particle-hole symmetry. Its role in leading to fermion number fractionalization was initially found by Jackiw and Rebbi [37], and then stressed in various insulating systems [16 -- 19, 38, 39]. In this case, a simple physical picture of fractional charge of e/2 around the 10 FIG. 6: (Color online). (a) The charge density of a chargeon state and (b) the spin density of a spinon on a 24 × 24 lattice with periodic boundary condition. The chargon at coordinate (6, 6) is f+(1/2)↑,+(1/2)↓, it has charge −e and spin Sz=0. The spinon at coordinate (18, 18) is f+(1/2)↑,−(1/2)↓, it has charge 0 and spin Sz=ℏ/2. defect is ready to obtain by the combined fact that (i) under particle-hole symmetry the relative charge density ρ on the soliton and the chemical potential µ satisfy ρ(µ)=−ρ(−µ), and (ii) when µ is in the bulk gap, the only difference between ρ(µ) and ρ(−µ) is the In the present 2D kagom´e lattice, filling of the zero modes localized at the two defects. on the other hand, the system has no particle-hole symmetry and thus no zero mode. In this case, although we cannot resort to the above simple physical picture, the presence of soliton-antisoliton doublet (quadruplet when spin included) in each bulk gap in Fig. 5(d) still guarantees the occurrence of fractionalized excitations. At 1/3 or 2/3 filling, occupation (unoccupation) of these in-gap modes leads to an excess (deficit) of 1/2 fermion number per spin and per defect [22]. Four different types of solitons with the following quantum numbers are obtained when these in-gap modes are occupied by different ways: the chargeon f+(1/2)↑,+(1/2)↓ (charge −e, Sz=0); the holon f−(1/2)↑,−(1/2)↓ (charge e, Sz=0); the two spinons f+(1/2)↑,−(1/2)↓ (charge 0, Sz= ℏ 2 ) and f−(1/2)↑,+(1/2)↓ (charge 0, Sz=− ℏ 2 ). Here the subscript + (−) represents that the in-gap mode is filled (empty) and ↑ (↓) labels the spin mode. For example, f+(1/2)↑,−(1/2)↓ means that the up-spin mode is filled and the down-spin mode is empty. For further illustration, the charge-density distribution for the chargeon f+(1/2)↑,+(1/2)↓ is calculated and shown in Fig. 6(a), while the spin-density distribution for the spinon f+(1/2)↑,−(1/2)↓ is plotted in 6(b). Here, we use a 24×24 lattice for calculation. 11 The quantum statistics of these spin-charge separated solitons can be readily seen by using anyon fusion argument [16, 39], which is based on the observation that the bound states of fractional excitation acquire non-trivial Berry phases on adiabatic exchange. Since the spin-up and spin-down bands in the present case decouple, so without loss of generality, let us consider a bound state of two identical spin-up solitons f(1/2)↑ (or f−(1/2)↑), which carries charge −e (e) and flux 2π∼0 thus is a fermion. According to the anyon fusion rule, which states that the exchange phase Θ of a particle formed by combining n identical anyons with exchange phase θ is Θ=n2θ, one easily obtains that the exchange phase between two identical solitons f(1/2)↑ (f−(1/2)↑) should be 1/4 that of fermions, i.e., θ(f(1/2)↑, f(1/2)↑) (or θ(f−(1/2)↑, f−(1/2)↑)=±π/4. Next let us consider a bound state of an f(1/2)↑ and f−(1/2)↑ soliton, which carries charge 0 and should be a boson. Then the exchange phase between solitons f(1/2)↑ and f−(1/2)↑ is given by θ(f(1/2)↑, f−(1/2)↑)=∓π/4. Since the spin-down band is the Hermitian conjugate of the spin-up band, one obtains θ(fα1↓, fα2↓)=−θ(fα1↑, fα2↑). Sub- stituting these results into the formula calculating the exchange phase between spin-charge separated solitons θ(fα1↑β1↓, fα2↑β2↓)=θ(fα1↑, fα2↑)+θ(fβ1↓, fβ2↓), one immediately concludes that the spinons, holon, and chargeon are all bosons. However, each spinon has nontrivial mutual exchange phase π with the chargeon and holon. Before ending this paper, we would like to point out that there are other ways for creating the spin-charge separated solitons in the kagom´e lattice. For example, by trimerizing the kagom´e lattice [33], like the 1D way proposed by Su and Schrieffer [40], the excitations of possessing fractional charge ±e/3 or ±2e/3 in two spacial dimensions can be realized. To make this point more clear, now we construct a trimerized pattern of the kagom´e lattice by introducing an appropriate bond distortion as shown in Fig. 7(a). The tight-binding Hamil- tonian describing this trimerized kagom´e lattice is generally written as H0 = Phiji tijc† i cj, where the hopping amplitude tij are different for different nearest-neighbor link hiji when the distortion is introduced. For simplicity and without loss of generality, we set in Fig. 7(a) the hopping term along the thick (thin) bonds as t + η (t − η), where η depicts the distortion amplitude, which is smaller than the undistorted hopping amplitude t. In this case, the unit cell now becomes larger and contains 9 sites [see the dashed lines in Fig. 7(a)]. Clearly, there are three different phases corresponding to the ground state (denoted by A, B, and C, respectively). The corresponding spectrum of this distorted kagom´e lattice (A, B, or C) with infinite size is drawn in Fig. 7(c). When the system is finite, i.e., it has two 12 boundaries along one direction (say, the x direction), there are edge states appearing in the bulk gaps [see the red thick lines in the spectrum drawn in Fig. 7(d)]. For comparison, we also plot in Fig. 7(b) the spectrum of the 2D perfect undistorted kagom´e lattice. In this case the 2D kagom´e lattice system has no bulk gaps. Two facts should be noted. One is that the bulk gaps appearing in Figs. 7(c) and 7(d) now result from the distortion. The other is that the in-gap edge states in Fig. 7(d) do not connect the neighboring two bulk bands. According to the winding properties of the edge states [4, 31], one knows that the insulating phase risen from the distortion is topologically trivial. That means the Hall conductance is zero when the Fermi energy lies in the bulk gaps. This point has been validated by the numerical calculation with the linear-response formula [34]. We can now study the structure of kinks connecting different phases. Similar to the case of one spatial dimension [40], we distinguish two classes of kinks: type I, which leads from A to B, B to C, or C to A as one moves from left to right; and type II, which leads from A to C, C to B, or B to A as one moves along the same direction. Fig. 7(e) plots the energy spectrum of the trimerized kagom´e lattice with kink I (or with kink II) by using 1296 sites. Every phase has 6 × 8 units and in each unit there are 9 independent sites. For numerical calculation, we set t=−1 and the distortion η=0.25t. From Fig. 7(e) one can clearly observe that there are excitation energies appearing in the bulk gaps. In this case, in the lower gap there are 9 excitation energies while in the higher gap there are 48 ones. These excitations are localized around the kinks, which is shown in Fig. 7(f). Every kink has fractional charge −e/3. That means we have realized the excitations possessing fractional charge −e/3 in the trimerized kagom´e lattice. When we numerically increase the system size along the y axis, the number of the excitation energies is found to increase in proportion to the system size. In the infinite limit, an excitation band eventually forms in the bulk gap. However, when we increase the size along the x axis, the number of the excitations keeps unchanged. When the spin freedom is considered, one can obtain the spin-charge separated excitations at 1/3 (or 2/3) filling. For example, in a small system with 6 × 6 units, there are 4 double-degenerate excitation states lying in the lower gap, one half is contributed from the lower bands and one half from the middle one. According to the different occupation ways, the spin-charge separated excitations with fractional quantum numbers, chargeon ±2e/3, as well as other excitations with fractional charge ±e/3 and spin ±ℏ/2, are obtained. In summary, we have theoretically studied quantum spin Hall effect and spin-charge 13 FIG. 7: (Color online). (a) Schematic pictures of the three degenerated ground states of a trimerized 2D distorted kagom´e lattice. The hopping amplitude along the thick (thin) lines is set as t + η (t − η), where t=−1 is set as energy unit and η=0.25t. The dashed lines represent the unit cell, which contains 9 lattice sites. The corresponding energy spectrums are shown in (c) (without boundary) and (d) (with boundary), respectively. The thick lines in (d) represent the edge states. The energy spectrum of the undistorted 2D kagom´e lattice (η=0) is drawn in (b). (e) Energy spectrum of the 2D kagom´e lattice with kink I. The gray shaded regions represent the bulk bands and the thick black lines represent the excitation energies. (f) The charge density of a fractional chargeon state on a trimerized kagome lattice composed by 36 units. The chargeon at x=2 line has charge −2e/3 and spin Sz=0. separation in a 2D kagom´e lattice. By using the topological winding numbers of the spin edge states on the complex-energy Riemann surface, we have obtained that the SHC is quantized as ±(e/2π) when the system is in the insulating phases, which is consistent with the Z2 topological invariant analysis and the numerical linear-response calculation. Furthermore, 14 we have constructed the spin-charge separated solitons in the kagom´e lattice by connecting the system's boundaries in twist. The quantum statistics of these solitons has also been discussed. This work was supported by NSFC under Grants No. 10604010, No. 10904005, and No. 60776063, and by the National Basic Research Program of China (973 Program) under Grant No. 2009CB929103. [1] D. J. Thouless, M. Kohmoto, M. P. Nightingale, and M. den Nijs, Phys. Rev. Lett. 49, 405 (1982). [2] X.-G. Wen, Quantum Field Theory of Many-Body Systems (Oxford University Press, New York, 2004). [3] B. I. Halperin, Phys. Rev. B 25, 2185 (1982). [4] Y. Hatsugai, Phys. Rev. Lett. 71, 3697 (1993). [5] F. D. M. Haldane, Phys. Rev. Lett. 61, 2015 (1988). [6] K. Ohgushi, S. Marakami, and N. Nagaosa, Phys. Rev. B 62, 6065(R) (2000). [7] M. Taillefumier, B. Canals, C. Lacroix, V. K. Dugaev, and P. Bruno, Phys. Rev. B 74, 085105 (2006). [8] Z. Wang and P. Zhang, Phys. Rev. B 76, 064406 (2007); Phys. Rev. B 77, 125119 (2008). [9] T. Tomizawa and H. Kontani, Phys. Rev. B 80, 100401(R) (2009). [10] R. Shindou and N. Nagaosa, Phys. Rev. Lett. 87, 116801 (2001). [11] Z. Wang, P. Zhang, and J. Shi, Phys. Rev. B 76, 094406 (2007). [12] C. L. Kane and E. J. Mele, Phys. Rev. Lett. 95, 226801 (2005). [13] C. L. Kane and E. J. Mele, Phys. Rev. Lett. 95, 146802 (2005). [14] B. A. Bernevig, T. L. Hughes, and S.-C. Zhang, Science 314, 1757 (2006). [15] M. Koenig, S. Wiedmann, Christoph Brune, A. Roth, H. Buhmann, L. W. Molenkamp, X.-L. Qi, and S.-C. Zhang, Science 318, 766 (2007). [16] Y. Ran, A. Vishwanath, and D.-H. Lee, Phys. Rev. Lett. 101, 086801 (2008). [17] X.-L. Qi and S.-C. Zhang, Phys. Rev. Lett. 101, 086802 (2008). [18] D.-H. Lee, G.-M. Zhang, and T. Xiang, Phys. Rev. Lett. 99, 196805 (2007). [19] C.-Y. Hou, C. Chamon, and C. Mudry, Phys. Rev. Lett. 98, 186809 (2007). 15 [20] A. Shitade, H. Katsura, J. Kunes, X.-L. Qi, S.-C. Zhang, and N. Nagaosa, Phys. Rev. Lett. 102, 256403 (2009). [21] S. Fujimoto, Phys. Rev. Lett. 103, 047203 (2009). [22] P. Zhang, Z. Wang, N. Hao, and W. Zhang (unpublished). [23] D. Mandrus et al., Phys. Rev. B 63, 195104 (2001). [24] D. J. Singh, P. Blaha, K. Schwarz, and J. O. Sofo, Phys. Rev. B 65, 155109 (2002). [25] K. Matsuhira et al., J. Phys. Soc. Jpn. 76, 043706 (2007). [26] S. Raghu, X.-L. Qi, C. Honerkkamp, and S.-C. Zhang, Phys. Rev. Lett. 100, 156401 (2008). [27] G. Chen and L. Balents, Phys. Rev. B 78, 094403 (2008). [28] B. J. Kim, et al., Science 323, 1329 (2009). [29] D. A. Pesin and L. Balents, arXiv:0907.2962. [30] M. Gibertini et al., arXiv:0904.4191. [31] Z. Wang, N. Hao, and P. Zhang, Phys. Rev. B 80, 115420 (2009). [32] A. Mielke, J. Phys. A 24, L73 (1991); 24, 3311 (1991); 25, 4335 (1992). [33] H.-M. Guo and M. Franz, Phys. Rev. B 80, 113102 (2009). [34] J. Sinova, D. Culcer, Q. Niu, N. A. Sinitsyn, T. Jungwirth, and A. H. MacDonald, Phys. Rev. Lett. 92, 126603 (2004). [35] L. Fu and C. L. Kane, Phys. Rev. B 76, 045302 (2007). [36] G. Liu, P. Zhang, Z. Wang, and S.-S. Li, Phys. Rev. B 79, 035323 (2009). [37] R. Jackiw and C. Rebbi, Phys. Rev. D 13, 3398 (1976). [38] W. P. Su, J. R. Schrieffer and A. J. Heeger, Phys. Rev. B 22, 2099 (1980). [39] C. Weaks, G. Rosenberg, B. Seradjeh and M. Franz, Nature Phys. 3, 796 (2007). [40] W. P. Su and J. R. Schrieffer, Phys. Rev. Lett. 46, 738 (1981). 16 

1205.4193

1

1205

Majorana state on the surface of a disordered 3D topological insulator

We study low-lying electron levels in an "antidot" capturing a coreless vortex on the surface of a three-dimensional topological insulator in the presence of disorder. The surface is covered with a superconductor film with a hole of size R larger than coherence length, which induces superconductivity via proximity effect. Spectrum of electron states inside the hole is sensitive to disorder, however, topological properties of the system give rise to a robust Majorana bound state at zero energy. We calculate the subgap density of states with both energy and spatial resolution using the supersymmetric sigma model method. Tunneling into the hole region is sensitive to the Majorana level and exhibits resonant Andreev reflection at zero energy.

cond-mat.mes-hall

cond-mat

Ma jorana state on the surface of a disordered 3D topological insulator P. A. Ioselevich,1, 2 P. M. Ostrovsky,3, 1 and M. V. Feigel’man1, 2 1L. D. Landau Institute for Theoretical Physics, Kosygin str. 2, Moscow, 119334 Russia 2Moscow Institute of Physics and Technology, Institutsky per. 9, Dolgoprudny, 141700 Russia 3Max Planck Institute for Solid State Research, Heisenbergstr. 1, 70569 Stuttgart, Germany We study low-lying electron levels in an “antidot” capturing a coreless vortex on the surface of a three-dimensional topological insulator in the presence of disorder. The surface is covered with a superconductor film with a hole of size R larger than coherence length, which induces supercon- ductivity via proximity effect. Spectrum of electron states inside the hole is sensitive to disorder, however, topological properties of the system give rise to a robust Ma jorana bound state at zero energy. We calculate the subgap density of states with both energy and spatial resolution using the supersymmetric sigma model method. Tunneling into the hole region is sensitive to the Ma jorana level and exhibits resonant Andreev reflection at zero energy. Topological insulators and superconductors are very peculiar materials with a gap in the bulk electron spec- trum and a low-lying branch of subgap excitations on their surface (see [1] for a review). This surface metallic state appears due to topological reasons and is robust with respect to any (sufficiently small) perturbations. In particular, topological properties prevent these surface states from Anderson localization. One common exam- ple of a topological insulator is a two-dimensional system in the integer quantum Hall effect regime. The bulk of such a system has a spectral gap between successive Lan- dau levels and is hence an insulator. At the same time quantized Hall conductance appears due to a fixed in- teger number of chiral propagating edge modes on the background of the bulk gap. This type of materials are referred to as Z topological insulators. Another type of topological insulator is realized in three-dimensional (3D) semiconductor crystals with suf- ficiently strong spin-orbit interaction (BiSb, BiTe, BiSe, strained HgTe etc). The spin-orbit interaction leads to inversion of the spectral gap. As a result subgap surface excitations appear with a dispersion of the massless Dirac type. The topological invariant in these materials has a Z2 nature. When the number of surface states is odd, one of them always remains gapless due to topological protection. A general classification of topological insulators was developed in Refs. [2, 3] based on the symmetry of the underlying Hamiltonian. The quantum Hall effect is an example of a topologically nontrivial state of the uni- tary symmetry (class A of the Altland-Zirnbauer clas- sification [4]) in 2D. Strong spin-orbit interaction leads to a topological state in the symplectic class (AII) in 3D. Another important example is the 1D topological superconductor of the class BD symmetry (superconduc- tor with both time-reversal and spin rotation symmetries broken). The topologically protected mode in this case is zero-dimensional and is known as the Ma jorana bound state (MBS). It appears, in particular, in the core of an Abrikosov vortex in the spinless p-wave superconductor [5]. A similar MBS appears [6] in a vortex in an ordinary Figure 1: (Color online) A sketch of the considered setup. Three-dimensional topological insulator is covered by a su- perconducting film with a hole of radius R. The gap in the density of states is induced on the covered surface forming an antidot and confining surface excitations inside the hole. External magnetic field induces an Abrikosov vortex inside the hole. The spectrum of energy eigenstates inside the hole acquires a Ma jorana zero level which can be accessed in the tunneling measurement. s-wave superconductor brought in contact with the sur- face of a 3D Z2 topological insulator, which corresponds to the symmetry class AII. The MBS appears as a descen- dant of the topologically protected massless Dirac state on the free surface of the topological insulator. There are two general methods to observe the Ma jo- rana level. One of them relates to the anomalous Joseph- son effect [7, 8]. Another, and a more direct, way involves tunneling into the region where the MBS is supposed to be localized [9]. The differential conductance in such a tunneling experiment yields the local density of states with spatial and energy resolution. The Ma jorana state in the vortex core manifests itself by resonant Andreev reflection at zero energy [10]. In this Letter we study local tunneling conductance for a setup depicted in Fig. 1. A superconducting (with a spectral gap ∆) film with a circular hole of the radius R is deposited on the surface of a 3D topological insu- lator, e.g., Bi2Te3 crystal. Perpendicular magnetic field applied to the system produces a vortex pinned to the hole. The radius R is supposed to be relatively large, R ≥ (l, ξ0 ), where l is the mean free path for topologi- cal insulator surface states and ξ0 is the superconducting coherence length of the film. The condition R ≥ ξ0 al- lows one to avoid [11] the abundance of low-lying Caroli- de Gennes-Matricon states whose presence in the core of the Abrikosov vortex complicates observation of the E = 0 Ma jorana state. Under these conditions disorder is relevant for the states localized in the hole region. Al- though the Ma jorana state is protected against disorder, i.e. its energy stays exactly zero, the spatial distribution of its wavefunction ψ0 (r), and thus the local tunneling conductance are sensitive to disorder. The aim of this Letter is to calculate the tunneling conductance in the “dirty-limit” with spatial and energy resolution, and to identify the effects of the Ma jorana zero-energy state in the presence of strong disorder. The problem is described by the following Bogolyubov- de Gennes (BdG) Hamiltonian in polar coordinates r, ϕ: ∆(r)eiϕ ∆(r)e−iϕ −H0 (cid:19) , H0 = v0 s · p + V (r) − µ. H = (cid:18) H0 (1) Here the Hamiltonian H0 describes the dynamics of sur- face excitations in a topological insulator without a su- perconducting layer. The Fermi velocity of surface elec- trons is denoted by v0 , s is the spin operator, and V (r) is the random disorder potential. The Fermi level is shifted from the Dirac point by µ. The vector potential term in H0 is neglected due to smallness of magnetic field. We assume the dirty limit with a disorder induced mean free path l ≪ ξ = pD/∆ and a relatively large hole with R ≫ ξ ; here D = v2 0 τ /2. These conditions allow us to use a step-like radial dependence of the order parameter: ∆(r) = (0, ∆, Inside the hole, the order parameter is zero and elec- tron dynamics is governed by H0 . This Hamiltonian possesses time-reversal symmetry of the symplectic type, H0 = syH ∗ 0 sy , and hence belongs to the symplectic sym- metry class AII. Proximity to the superconductor induces a gap in the electronic spectrum outside the hole. This gap effectively confines the low-lying excitations and im- poses boundary conditions for the Hamiltonian H0 at r = R. These boundary conditions break time-reversal symmetry due to the spatially rotating phase of the order parameter. At the same time the total BdG Hamiltonian H acquires a specific particle-hole symmetry: r < R, r > R. (2) H = −sy τyH ∗ sy τy . Here τy is the Pauli matrix acting in Nambu-Gor’kov (NG) space. Identity (3) implies a symmetry of the elec- tron spectrum. For each eigenstate ψ with energy E there is a conjugate eigenstate sy τy ψ∗ with energy −E . (3) 2 The particle-hole symmetry (3) defines the class BD. On the level of random matrices there are two distinct versions of this class with even and odd number of eigen- states referred to as D and B class, respectively. In class B, one unpaired eigenstate has exactly zero energy and is self-conjugate: ψ = sy τy ψ∗ . The BdG Hamiltonian counts every physical excitation twice due to the dou- bling in NG space. An unpaired level is thus “half ” of a true excitation — a Ma jorana state. We will calculate the density of states inside the hole with the help of the supersymmetric non-linear sigma model. Two-dimensional Dirac fermions with potential disorder are described by a very peculiar model of the class AII with Z2 topological term [12]. This topologi- cal term appears as a consequence of the chiral anomaly of Dirac fermions. We will consider the minimal model operating with the 8 × 8 supermatrix Q. Apart from Fermi-Bose superspace, this matrix operates in the space of retarded-advanced (RA) fields and in a specific time- reversal (TR) space introduced to take into account Cooperons and diffusons on equal footing. RA space is completely analogous to the NG space (see below); we denote Pauli matrices in this space by τ . Notation σ is used for Pauli matrices in TR space. A detailed deriva- tion of the sigma model can be found in [13]. In class AII, the matrix Q obeys the non-linear con- straint Q2 = 1 and the linear constraint 0 iσy (cid:19)FB C = τx (cid:18)σx 0 As a result, Q contains 8 commuting and 8 anticommut- ing (Grassmann) variables in total. Commuting degrees of freedom parameterize diagonal fermion-fermion (F) and boson-boson (B) blocks of Q. The general Q-matrix can be decomposed as 0 QB(cid:19) U, Q = U −1 (cid:18)QF 0 where the central part contains only commuting variables while Grassmann parameters define the unitary super- matrix U . We will use the following explicit form of the central part of Q in terms of eight angle parameters: Q = ¯Q ≡ CQT C T , (5) . (4) QF = [τz cos θf + σz sin θf (τx cos φf + τy sin φf )] × [σz cos kf + τz sin kf (σx cos χf + σy sin χf )], (6) QB = τz cos θb [σz cos kb + sin kb (σx cos χb + σy sin χb )] + sin θb (τx cos φb + τy sin φb ). (7) This representation fulfills all the constraints imposed by the symmetry class AII of the Hamiltonian H0 . The F and B sectors of the sigma model are compact and non- compact, respectively. This is achieved by demanding that the angles θb , kb are imaginary while all other angles in Eqs. (6), (7) are real. Below we will find that the saddle point describing the density of states in the hole occurs S [Q] = with real θb . This implies a proper shift of the integration contour for this angle. The sigma model of class AII is designed to study transport properties of a disordered system. This im- plies that the Q matrix operates, in particular, in the RA space allowing for averaging the product of retarded and advanced Green functions. We are interested in the density of states and hence it suffices to average just the single retarded function. At the same time, supercon- ducting boundary conditions implemented in the BdG Hamiltonian (1) require to introduce an additional dou- bling of fields in the Nambu-Gor’kov space. This can be achieved within the standard AII class sigma model with the 8 × 8 supermatrix Q while the role of NG space is taken over by the RA structure of Q (for detailed discus- sion of the transformation from RA to NG representation see [13, 14]). Thus we can incorporate the superconducting order parameter directly into the action of the sigma model, πν 8 Z d2 r Str hD(∇Q)2 + 4(iǫΛ − ∆)Qi + Sθ [Q], ∆ = ∆(r)(τx cos ϕ − τy sin ϕ). Λ = τz σz , (8) Here ǫ = E + iGt δ(r − r0 )/4πν is the sum of the energy E and the local dwell term describing the coupling to a tunnel tip with dimensionless conductance Gt ≪ 1 [13]. The action (8) involves the Z2 topological term Sθ [Q] which appears due to massless Dirac nature of underly- ing electrons as was discussed above. The topological term involves only the compact part of the sigma-model manifold, i.e., its F sector MF . In the general version of the sigma model, that is capable of averaging several retarded and advanced Green functions, the homotopy group is π2 (MF ) = Z2 . However, in the minimal model we are considering, MF has the structure of the product of two spheres S 2 × S 2 as seen from Eq. (6). In this case the homotopy group is richer, Z×Z. Two integer topolog- ical invariants can be introduced counting the degree of covering of the two spheres by the mapping Q from real space to MF . This allows us to write the topological term explicitly although in a non-invariant form. With the parameterization (6), it reads (cf. Ref. [12]): i 4 Z d2 rh sin θf (cid:0)∇θf × ∇φf (cid:1) + sin kf (cid:0)∇kf × ∇χf (cid:1)i. Let us now analyze the minima of the action (8) for the setup Fig. 1. Circular symmetry of the problem fixes the phase equal to the polar angle, φf ,b = ϕ. The other parameters depend only on the radial coordinate r. Both angles θ in F and B sectors obey the Usadel equation D (cid:20) ∂ 2 θ 2r2 (cid:21) ∂ θ sin 2θ ∂ r2 + ∂ r − + 2iE sin θ cos k + 2∆(r) cos θ = 0. Sθ [Q] = (10) (9) 1 r 3 Inside the hole ∆ = 0. At low energies we can also neglect the E term and the equation becomes independent of k . The step-like dependence of the order parameter, Eq. (2), imposes the boundary condition θ(R) = π/2. There are two possible solutions to the Usadel equation with this boundary condition: (11) (12) θ1 = 2 arctan(r/R), θ2 = π − 2 arctan(r/R). Saddle point equations also require kf = 0 and hence the angle χf drops from the matrix Q and from the action. The two remaining angles kb and χb are free and can take any constant values. The spatial profile of Q is thus fixed by the Usadel equation. The solutions θ1,2 represent two disconnected saddle points in the F sector while in the B sector only the saddle point θb = θ1 is reachable. If the integration contour for θb , which runs along the imaginary axis, is shifted to the point θ2 a divergence occurs in the kb in- tegral. Thus the B sector is reduced to a hyperboloid parameterized by ikb > 0 and 0 < χb < 2π , see Ref. [13]. This is exactly the structure of the sigma model of class BD as we anticipated from symmetry analysis. The dis- tinction between even (D) and odd (B) versions of this class is related to the disjoint character of the manifold due to the discrete degree of freedom in the F sector. Namely, in class D (B) the two parts of the manifold contribute to the partition function with the same (op- posite) sign. In our problem the odd symmetry class B occurs. To demonstrate it, we compare the value of the action (8) at the two minima in the F sector. These two solutions indeed contribute with the opposite sign since the corresponding values of the topological term (9) differ by exactly iπ . The density of states is given by the integral Re Z DQ Str[kΛQ(r)] e−S [Q] . At low energies, this integral is to be calculated over the saddle manifold. Apart from the two variables kb and χb and two disconnected points in the F sector this manifold involves two Grassmann variables in the matrix U , Eq. (5). In order not to spoil the saddle point, these vari- ables must be constant in space and fulfill the condition [U, ∆] = 0. We introduce them according to 2 (cid:18) (cid:19)FB(cid:21) . U = exp (cid:20) 1 0 iζ σx − iησy Within this parameterization, we can rewrite the integral (13) in terms of kb , χb , η , ζ . Thus the problem is reduced to the 0D sigma model of class B (the two disjoint points in the F sector contribute with opposite signs). Explicit calculation of the integral [13, 15] yields the local DOS in the factorized form ησx + ζ σy 0 ρ(E , r) = (14) ν 8 (13) ρ(r, E ) = ν n(r)f (E /ω0 ), (15) 2 R Ν Π

E
N 1.0 0.8 0.6 0.4 0.2 0.0 0 2 4 6 8 10 E
E Th Figure 2: (Color online) Global density of states as a func- tion of energy E . The solid blue oscillating curve represents the low-energy result (19), the dashed curve shows the high- energy asymptotics (20), and the red curve presents numerical solution interpolating between the two limits. n(r) = cos θ1 (r) = f (x) = γ π(x2 + γ 2) + 1 − R2 − r2 R2 + r2 , sin(2πx) 2πx (16) (17) . Here γ = Gtn(r0 )/2π ≪ 1 and the low-energy level spac- ing is given by 0 = 2ν Z d2 r cos θ1 (r) = 2π(log 4 − 1)νR2 . ω−1 The spatial profile of DOS, n(r), is fixed by the solution of the Usadel equation, while the energy dependence is char- acteristic for the B class and contains a narrow lorentzian peak at zero energy. This peak is the zero-energy MBS broadened due to the finite tunneling time. Note that the width γ is position-dependent. Integrating ρ(r, E ) over space yields the global DoS (18) N (E ) = f (E /ω0) 2ω0 . (19) In the extreme tunneling limit γ → 0, this function ac- quires the contribution δ(E )/2 and coincides with the result [15] up to a factor 2 due to BdG double counting. The Ma jorana state appears as a half of a fermionic level. Spatial and energy dependence of ρ(r, E ) factorize at energies much less than the Thouless energy ETh = D/R2 . At higher energies fluctuations of Q are not im- portant and DOS is given just by a single saddle point. Approximate solution of the Usadel equation (10) in the limit E ≫ ETh yields [13] √2(cid:1)pETh /E i . N (E ≫ ETh ) = πνR2 h1 − (cid:0)2 − Local DOS is close to the normal value ν everywhere except for a narrow vicinity of the hole boundary. (20) 4 (23) I = (22) (21) dI dV , T ≪ γω0 , The local density is measured in the tunneling experi- ment. The tunneling current is determined by eGt 2πν Z ρ(E , r0 )(cid:2)f (E − eV ) − f (E )(cid:3)dE with f (E ) being the equilibrium Fermi distribution func- tion. At low temperatures and voltages, (T , eV ) ≪ ω0 , we keep only the first lorentzian term in the energy de- pendence of the local DOS (17). The differential conduc- tance exhibits a peak e2γ 2 =  π(cid:2)γ 2 + (eV /ω0 )2 (cid:3) e2γω0 γω0 ≪ T ≪ ω0 . ,  4T cosh2 (eV /2T ) At very low temperatures the height of this peak is uni- versal and equals e2/π. This signifies resonant Andreev reflection at the Ma jorana state. This effect in the ab- sence of disorder was studied in Ref. [10]. At larger (and more realistic) temperatures, γω0 ≪ T ≤ ω0 , the height of the peak is parametrically small, dI /dV ∼ γω0/T . Noise power of the tunneling current in the same regime (T , eV ) ≪ ω0 is [13] e2γω0 γω0 ≪ T ≪ ω0 . , S (V , T , r0 ) = 2 The noise produced by the Ma jorana level is T - and V - independent as long as γω0 ≪ T . The results (22) and (23) apply at low temperature and voltage. When temperature and/or voltage are higher than the level spacing ω0 , non-zero-energy states also contribute to the tunneling current. Positions and widths of these states depend on the realization of disorder. For low temperature but high voltage, T < ω0 < eV , narrow resonances similar to Eq. (22) will occur due to non-zero levels [16]. Positions of these resonances strongly depend on disorder realization; their heights are smaller but close to e2/π and widths are of order γω0 . When tempera- ture exceeds ω0 , all the resonances get smeared and the normal average DOS ν is recovered. To conclude, we have studied the local density of states in the superconducting vortex on the surface of a topo- logical insulator in the superconducting antidot setup de- picted in Fig. 1. The spatial and energy dependence of the density of states factorize at low energies and the lat- ter is given by the 0D sigma model of symmetry class B. We have identified the zero-energy Ma jorana state occur- ring due to the topological properties of the system. This Ma jorana state exhibits itself via the resonant Andreev reflection at zero energy yielding the peak in differen- tial conductance with the universal amplitude e2/π and width proportional to the normal conductance Gt . We are grateful to B. Sacepe for stimulating discus- sions. This work was supported by the RFBR grant 10- 02-00554-a and by the German Ministry of Education and Research (BMBF). [1] X.-L. Qi, S.-C. Zhang, Rev. Mod. Phys. 83, 1057 (2011). [2] A. Kitaev, AIP Conf. Proc. 1134, 22 (2009). [3] A. P. Schnyder, et al., Phys. Rev. B 78, 195125 (2008). [4] A. Altland and M. Zirnbauer, Phys. Rev. B 55, 1142 (1997). [5] D. Ivanov, Phys. Rev. Lett. 86, 268 (2001). [6] L. Fu and C. L. Kane, Phys. Rev. Lett. 100, 096407 (2008). [7] L. Fu and C. L. Kane, Phys. Rev. B 79, 161408 (2009). [8] P. A. Ioselevich and M. V. Feigel’man, Phys. Rev. Lett. 106, 077003 (2011). 1 [9] V. Mourik et al., Science 1222360 (2012). [10] K. T. Law, P. A. Lee, and T. K. Ng, Phys. Rev. Lett. 103, 237001 (2009). [11] A. S. Mel’nikov, A. V. Samokhvalov, M. N. Zubarev, Phys. Rev. B 79, 134529 (2009). [12] P. M. Ostrovsky, I. V. Gornyi, and A. D. Mirlin, Phys. Rev. Lett. 105, 036803 (2010). [13] For details of sigma model derivation and calculation of the density of states see supplementary material. [14] V. Koziy and M. A. Skvortsov, JETP Lett. 94, 222 (2011). [15] D. A. Ivanov, J. Math. Phys. 43, 126 (2002). [16] P. A. Ioselevich and M. V. Feigel’man, to be published. Supplementary Information Derivation of the sigma-model Sigma model for hybrid structure In this section we derive the sigma model describing the dynamics of the topological insulator surface excitations in the presence of a random potential. We assume the hole geometry depicted in Fig. 1 of the main text. The derivation starts with the microscopic Bogolyubov – de Gennes Hamiltonian (1). The sigma model is aimed at averaging the single retarded Green function determining the density of states. We will also demonstrate the equivalence of this sigma model to the model of the symplectic symmetry class AII describing the transport properties of massless Dirac fermions. The latter model is based on the normal, rather than superconducting, Hamiltonian and is capable of calculating averaged products of retarded and advanced Green functions. It is known that the sigma model for Dirac fermions possesses the specific Z2 topological term. The equivalence of the two models will prove the appearance of the topological term in the superconducting case considered in the present paper. Let us start with the retarded Green function in the system governed by the Hamiltonian (1). We will use the supersymmetric integral representation S = −i Z d2 r Φ† (cid:0)E + i0 − H (cid:1)Φ. ′ ) = −i Z DΦ∗ DΦ str (cid:2)kΦ(r)Φ† (r GR ′ )(cid:3) e−S , E (r, r Here the vector fields Φ and Φ† contain 4 commuting and 4 Grassmann parameters each. Apart from the spin and Bogolyubov – de Gennes structure of the Hamiltonian, we also introduce the superstructure in order to get rid of the normalizing denominator and facilitate further disorder averaging. The pre-exponential factor contains the supermatrix k = {1, −1}FB and supertrace is defined as in Ref. [17]: str A = AFF − ABB . The superconducting symmetry of the Hamiltonian H , Eq. (3), gives rise to specific soft modes – Cooperons – of the type hGR E GR −E i. These modes are relevant for the average density of states since they are built out of retarded functions only. In order to include them into our effective theory, we transform the action by writing half of it in the form (24) and another half in the time-reversed (transposed) form: i 2 Z d2 rhΦ†(cid:0)E + i0 − H (cid:1)Φ − ΦT k(cid:0)E + i0 + sy τyH sy τy (cid:1)Φ∗ i 2 Z d2 r (cid:0)Φ+ τz , iΦT ksy τx (cid:1) h(E + i0)σz τz − H0 − τz ∆i (cid:18) Φ sy τy Φ∗(cid:19) . i = − Here we have introduced the matrix σz operating in the space of time-reversed blocks – TR space. The matrix k appeared due to anticommutation of Grassmann variables. We will denote the doubled vectors as √2 (cid:18) Φ sy τy Φ∗(cid:19) , 1 1 √2 (cid:0)Φ+ τz , iΦT ksy τx (cid:1) . They are no longer independent, as Φ and Φ∗ were, but rather obey the linear constraint 0 iσy (cid:19)FB C = −isy τx (cid:18)σx 0 ¯Ψ = (CΨ)T , S = − ¯Ψ = Ψ = (24) (25) (26) (27) . At this stage, we are ready to perform the disorder averaging. We adopt the standard Gaussian white noise disorder model characterized by the correlator 2 ′ ) δ(r − r πν τ . (28) ′ )i = hV (r)V (r The single-particle density of states is defined as ν = µ/2πv2 0 . Note that the right-hand side is twice bigger compared to the standard definition applied to normal metals. The reason for this extra factor 2 is the Dirac nature of electron spectrum and hence doubling of the phase space due to the two types of excitations (electrons and holes). Disorder averaging produces the quartic term in the action, S = Z d2 r (cid:26) ( ¯ΨΨ)2 2πν τ − i ¯Ψ h(E + i0)Λ + µ − v0 s · p − τz ∆i Ψ(cid:27) . Here we use the notation Λ = σz τz . The interaction term is further decoupled with the help of the Hubbard- Stratonovich transformation by introducing an auxiliary matrix field Q. The construction of the nonlinear sigma model implies that the field Q contains all relevant slow modes of the disordered system – diffusons and Cooperons. There are in total three ways to decouple the four-fermion interaction [17]. One of them involves a scalar field ∼ h ¯ΨΨi analogous to the random potential V . This leads only to an irrelevant renormalization of the chemical potential. Two other decoupling schemes introduce a matrix field with the structure hΨ ¯Ψi and hΨΨT i corresponding to diffusons and Cooperons, respectively. In order to include all relevant slow modes in the theory, we have to perform decoupling in both ways, which can be achieved with a single matrix Q. Details of this calculation can be found in Ref. [17]. The resulting action has the form: (29) S = 2τ (cid:21) Ψ. str Q2 − i Z d2 r ¯Ψ (cid:20)EΛ + µ − v0 s · p − τz ∆ + iQ πν 16τ The vectors Ψ and ¯Ψ are related by Eq. (27). This allows us to limit the matrix Q by the linear constraint Q = ¯Q ≡ CQT C T . This relation keeps only those parameters in Q that couple to the product Ψ ¯Ψ. Integrating out the field Ψ yields the action for the matrix Q: (30) 1 2 S = (31) πν 16τ 2τ (cid:21) . str ln (cid:20)EΛ + µ − v0 s · p − τz ∆ + iQ str Q2 − Derivation of the sigma model proceeds with the saddle-point analysis of the above action. In the dirty limit, the energy and ∆ terms in the argument of the logarithm are relatively small compared to Q/τ . With these small terms neglected, the action possesses the uniform saddle point Q = Λ. This saddle point corresponds to the self energy of the average electron Green function in the self-consistent Born approximation. Other saddle points can be achieved by rotations Q = T −1ΛT , if the matrix T commutes with the spin operator s (E and ∆ terms are still neglected). Rotations T define the target manifold of the non-linear sigma model. The effective action of the sigma model within this manifold is derived with the help of the gradient expansion of Eq. (31) allowing for slow spatial variation of Q and perturbative expansion to the linear order in E and ∆. Since the Q matrix is trivial in the spin space, we can safely reduce its size to 8 × 8 keeping only Nambu, TR, and FB structure. Then the self-conjugacy relation acquires the form of Eq. (4). The gradient expansion of Eq. (31) is a highly nontrivial procedure in view of the chiral anomaly of the Dirac fermions. The momentum integrals arising after the expansion of the logarithm are divergent in the ultraviolet limit and require a proper regularization. The result of the expansion is independent of a particular regularization scheme provided the gauge invariance is preserved. The anomaly affects only the imaginary part of the action and leads to the appearance of the topological term [12]. At the same time, the real part can be obtained in a straightforward way since all the arising momentum integrals are convergent. The result reads πν 8 Z d2 r str hD(∇Q)2 + 4i(EΛ − τz ∆)Qi . Here the diffusion coefficient is D = v2 0 τ /2 and the matrix Q is reduced to the 8 × 8 size. The action (8), used in the main text, differs from Eq. (32) only by a π/2 rotation of the superconducting phase and by an imaginary contribution to the energy term. The latter corresponds to a finite dwell time of the electron due to the coupling to the tunneling microscope tip as we elaborate below. In order to explain the emergence of the topological term in the imaginary part of the action, we will prove the equivalence of the sigma model, derived here for the hybrid structure of Fig. 1, to the symplectic sigma model for Dirac fermions derived in Ref. [12]. Re S = (32) Sigma model for Dirac fermions 3 The sigma model obtained by the gradient expansion of Eq. (31) belongs to the symplectic symmetry class AII. This is quite natural since the Hamiltonian possesses only time-reversal symmetry in the absence of the ∆ term. Let us demonstrate the equivalence between our sigma model (8) describing the density of states for the Hamiltonian H , Eq. (1), and the standard symplectic class sigma model describing transport via the surface states of the topological insulator with Hamiltonian H0 . In the latter case, the sigma model yields averaged products of retarded and advanced Green functions with the energy difference ω . This requires introducing retarded-advanced (RA) structure of the action, S0 = − S0 = −i Z d2 r Φ†τz (cid:2)(ω/2 + i0)τz − H0 (cid:3)Φ. Here the matrix τz operates in the RA space. Subscript 0 is used to distinguish this model from the sigma model derived in the previous section and used in the main text. The time-reversal symmetry H0 = syH T 0 sy is taken into account by further doubling of variables in the TR space, i 2 Z d2 rhΦ† (cid:0)ω/2 + i0 − τz H0 (cid:1)Φ − ΦT k(cid:0)ω/2 + i0 − τz syH0 sy (cid:1)Φ∗ i 2 Z d2 r (cid:0)Φ† τz , iΦT ksy τz (cid:1) (cid:2)(ω/2 + i0)Λ0 − H0 (cid:3) (cid:18) Φ isyΦ∗(cid:19) . i = − Note that the matrix Λ0 = τz multiplying the frequency term is different from its counterpart Λ from the previous section. The vectors Ψ and ¯Ψ as well as the charge conjugation matrix C0 are also defined differently, 0 iσy (cid:19)FB C0 = −isy τz (cid:18)σx √2 (cid:18) Φ isy Φ∗(cid:19) , 1 1 0 √2 (cid:0)Φ+ τz , iΦT ksy τz (cid:1) , Already at this stage we can prove the equivalence of the two models by selecting a proper unitary rotation of the field vector Ψ. This unitary rotation should obey the following properties: ¯Ψ = (C0Ψ)T = Ψ = . (35) (33) (34) U T C0U = C, U −1Λ0U = Λ. (36) These relations bring the charge conjugation matrix C0 and the matrix Λ0 to the representation used in the previous section for the sigma model of the hybrid system. There are many matrices U that fulfill identities (36). One possible choice is 0 iτy (cid:19)TR U = (cid:18)1 0 In the case of usual rather than Dirac fermions, an analogous equivalence between the sigma model for the density of states in a superconducting hybrid structure and the orthogonal class sigma model is discussed in Ref. [14]. (37) . Topological term So far we have discussed the derivation of the real part of the sigma-model action (8) and also proved the equivalence of this model to the symplectic class sigma model for Dirac fermions (up to the boundary conditions involving the term ∆). The latter model is known to possess the Z2 topological term as an imaginary part of its action, see Ref. [12]. This allows us to include the topological term in the action (8) and thus complete its derivation. The explicit form of the topological term is written in the main text, Eq. (9), in a noninvariant form, using the parameterization (5) – (7). In this section we will discuss an indirect but explicitly invariant representation of this topological term. The target manifold of the symplectic class sigma model is fixed by the representation Q = T −1ΛT with the constraint Q = ¯Q. If we neglect the Grassmann parameters and consider the central part of Q, see Eq. (5), these conditions yield QF ∈ O(4)/O(2) × O(2) and QB ∈ S p(2, 2)/S p(2) × S p(2). The topological term arises in the compact F sector of the model. An explicit expression for the topological term can be written within a construction similar to the Wess-Zumino-Witten term, cf. Ref. [18]. 4 (38) (39) Sθ [Q] = δSθ [Q] = t = 1, t = 0. We first extend the target manifold by relaxing the condition Q2 = 1. Let us introduce the matrix Q = ¯T ΛT . The only restriction on this matrix is Q = ¯Q. The F sector of the extended manifold is QF ∈ O(4). It has trivial second homotopy group, π2 [O(4)] = 0. This allows us to introduce the third, auxiliary, coordinate 0 ≤ t ≤ 1, such that the real 2D space corresponds to t = 1, and continuously extend the matrix field according to Q(r, t) = (Q(r), Λ, Such an extension assumes that the physical space has no boundary and can be viewed as a surface of a 3D ball with t being the radial coordinate. With this definition of Q, the Wess-Zumino-Witten term has the form 24π Z 1 iǫabc dt Z d2 r str hQ−1 (∇aQ)Q−1 (∇bQ)Q−1 (∇cQ)i. 0 Here the indices a, b, and c take three values corresponding to the three coordinates t, x, and y , and ǫabc is the unit antisymmetric tensor. The integrand in this expression (39) is a total derivative therefore the result of the integration depends only on the value of Q at the boundary of integration domain, i.e., on the physical field Q = Qt=1 . This can be checked by an explicit calculation of the variation of Sθ : 8π Z 1 iǫabc i dt Z d2 r∇a str hQ−1 δQQ−1(∇bQ)Q−1 (∇cQ)i = 8π Z d2 r str nQ−1δQ(cid:2)Q−1(∇xQ), Q−1(∇y Q)(cid:3)o. 0 (40) A topological term does not change under continuous variations of the field Q but takes different values in different (disconnected) topological sectors of the model. The general Wess-Zumino-Witten term has a nonzero variation and hence does not obey this property. However, the term (39) is constrained by Q2 = 1. Under this condition, the variation (40) is identically zero. Hence the Wess-Zumino-Witten term (39) indeed plays the role of a topological term. The value of the topological term is quantized, yielding the topological charge of the field configuration, only in a system without boundary. This is not the case for the geometry considered in the paper. The matrix Q is defined in the finite region inside the hole (see Fig. 1 of the main text) with the boundary conditions fixed by the ∆ term in the Hamiltonian (1). The value of the topological term is not completely fixed in this case. In particular, the expression (39) depends not only on physical values of Q but also on the way it is extended in the third dimension, Eq. (38). However, this uncertainty leads only to an uncontrolled imaginary constant in the action. This constant is the same in both topological sectors of the model and hence does not alter any observable quantities. An alternative derivation of the sigma-model action, including the topological term in the form Eq. (39) is possible within the non-abelian bosonization formalism. Let us for a moment assume, that the Hamiltonian has an extra chiral symmetry, H = −szH sz . This situation corresponds to the symmetry class DIII and can be realized, e.g., at the Dirac point in the presence of a random velocity disorder. The sigma model of the class DIII has an extended target space, corresponding to the manifold of Q introduced above. This sigma model possesses the Wess-Zumino-Witten term (39), see Ref. [19], and is the result of the non-abelian bosonization of the initial fermionic problem. Non-zero chemical potential drives the system away from the Dirac point and breaks the chiral symmetry. This reduces the model to the symplectic symmetry class and restricts the field by the condition Q2 = 1. The resulting action has the topological term in the form of the restricted Wess-Zumino-Witten term (39) as discussed above. Tunneling coupling term In this section we discuss the appearance of the imaginary contribution to the energy due to the presence of the tunneling tip. In order to include the coupling to an external metallic probe, we extend the Hamiltonian as t† HM (cid:19) . H = (cid:18)H t Diagonal blocks of this matrix are the Hamiltonian H of the topological insulator surface (1) and the Hamiltonian HM describing electron states in the metallic tip. Off-diagonal elements t and t† are tunneling amplitudes between the two subsystems. We assume these tunneling amplitudes are local in space and couple the states at point r0 on the sample surface to the states at the tip of the metallic probe. (41) Deriving the sigma model as described in previous sections we arrive at the expression similar to Eq. (31) but with an extended matrix structure: 5 1 2 S = πν 16τ str Q2 + πνM 16τM iQM /2τM (cid:19)(cid:21) . str ln (cid:20)EΛ − τz H + (cid:18)iQ/2τ 0 str Q2 M − 0 The quantities νM , τM , and QM refer to the metallic tip. In the absence of tunneling, t = 0, the two subsystems decouple and we obtain independent sigma models describing topological insulator and the tip. We will assume the density of states νM to be large, that will fix QM = Λ. Tunneling coupling is assumed to be weak. This allows us to expand the logarithm in powers of t and t† . The lowest non-vanishing contribution appears in the second order of the perturbation theory and yields the tunneling term in the sigma model, π2 1 str h(EΛ − τz H + iQ/2τ )−1τz t(EΛ − τz HM + iQM /2τM )−1 τz t† i = − ν νM str (cid:0)Qτz tQM τz t†(cid:1). 2 8 In the last expression we have used the saddle-point relation between Q and the Green function (EΛ − τz H + iQ/2τ )−1 in coincident points. We introduce normal dimensionless (in units e2/h) tunnel conductance of the junction Gt = π2 ν νM tr(tt† ) (note that ν and νM are total rather than spinless densities of states at the Fermi energy). Assuming QM = Λ, we rewrite the action as St = (42) (43) Gt St = − 8 Thus the tunneling term has the same structure as the energy term and produces an imaginary contribution to E , see Eq. (8) of the main text. str ΛQ. (44) Calculation of the density of states Low energies E ≪ ETh The local density of states on the surface of the topological insulator is given by Eq. (13) of the main text. The integral over matrix Q is to be calculated over the saddle manifold fixed by the solution of the Usadel equation (10). There are two distinct solutions θ1,2 , [see Eqs. (11), (12) of the main text] for the angle θF . In the F sector of the model, all other parameters are fixed and thus θ1,2 represent two disjoint parts of the integration manifold. At the same time, only the solution θb = θ1 is allowed in the B sector, and two other angles, kB and χB , are free forming a hyperboloid (the parameter kB must be imaginary in order to ensure convergence of the integral). To calculate the density of states from Eq. (13), we have to express Str(kΛQ(r)), S [Q] and DQ in terms of kb , χb , η , and ζ for the two disjoint parts of the manifold. However, a direct use of our parameterization leads to an uncertainty in the integral over the part corresponding to θf = θ2 . The integral over dη dζ is zero, while the integral over kb diverges. We resolve this uncertainty by a trick proposed in Ref. [15]. The two disjoint parts of the integration manifold can be mapped onto each other by the similarity transformation T = (cid:18)σx 0 0 1(cid:19)FB Q 7→ T −1QT , We use the parametrization (6), (7) in the domain with θf = θ1 and then apply the transformation (45) to cover the second topological sector with θf = θ2 . Owing to the relation (45), integration measure is the same in both sectors. It is given by the superdeterminant of the metric tensor in the space of Q matrices, see Ref. [17]. The integration measure reads (45) . tanh DQ = dκ dχb dη dζ , 1 κ 2π 2 with κ = ikb (kb is imaginary to ensure the convergence of the integral). The integration runs over the hyperboloid κ > 0, 0 < χb < 2π . The action in the two topological sectors θf = θ1,2 has the form iπ = −2iπ x (cid:16)sinh2 κ 2 − iηζ cosh2 κ 2 − iηζ cosh2 κ S1 = −4iπν Z d2 r ǫ cos θ1 (cid:16)sinh2 κ 2 (cid:17) − 2 (cid:17) − 2 S2 = −4iπν Z d2 r ǫ cos θ1 cosh2 κ = −2iπ x cosh2 κ iπ iπ 2 − + . 2 2 2 iπ 2 (47) (46) (48) , ν 8 (49) (50) tanh κ 2 cosh2 κ , 2 2 − iηζ cosh2 κ sinh2 κ 2 6 Here we have introduced the complex dimensionless energy parameter x = R d2 r ǫ cos θ1 = x + iγ , with the real part x = E /ω0 [see Eq. (18) in the main text] and the imaginary part γ = Gtn(r0 )/2π . The latter appears due to the coupling to the tunneling tip. The terms ±iπ/2 in Eqs. (47), (48) appear due to the topological term Sθ [Q], see Eq. (9). With the above expressions for DQ and the action we can calculate the total partition function of the theory, Z DQ e−S [Q] = Z ∞ dκ dη dζ exp (cid:20)2iπ x (cid:16)sinh2 κ 2 (cid:21) = 1. 2 − iηζ cosh2 κ iπ 2 (cid:17) + 0 This is exactly what one expects for the partition function of a supersymmetric theory. Note that the topologically nontrivial sector does not contribute to this result since the action S2 contains no Grassmann variables. In order to calculate the density of states, we need the pre-exponential factor from Eq. (13). In our parameterization, it has the following form in the two sectors: Str(kΛQ) = ν cos θ1 ×   dκ dη dζ( cosh2 κ 2 (cid:21) exp (cid:20)2iπ x (cid:16)sinh2 κ iπ 2 − iηζ cosh2 κ 2 (cid:17) + 2 2 (cid:21) ) 2 (cid:17) exp (cid:20)2iπ x cosh2 κ + (cid:16)sinh2 κ 2 − iηζ cosh2 κ iπ 2 − 2 " − 2iπ x cosh2 κ 2 (cid:17) # 2 (cid:17) + exp (cid:16)2iπ x cosh2 κ exp (cid:16)2iπ x sinh2 κ κ κ sinh dκ cosh 2 2 1 + e2iπ x = ν cos θ1 (cid:18)1 − Re 2iπ x (cid:19) = ν cos θ1 (cid:20)1 + 2πx (cid:21) . γ sin(2πx) π(x2 + γ 2) − In the last expression we have used the condition γ ≪ 1. The result (51) coincides with Eqs. (15) – (17) of the main text. In the tunneling limit γ → 0 the lorentzian term yields a delta function. This implies that the Ma jorana state is robust in a closed system and disorder does not smear it. Note, that for an alternative model of the symmetry class D, when the topological term is absent, the density of states does not contain the Ma jorana delta-peak. Two topological sectors contribute with the same sign yielding [15] 2πx (cid:21) . ρD = ν cos θ1 (cid:20)1 + sin(2πx) ρ = ν cos θ1 Re Z ∞ 0 = ν cos θ1 Re Z ∞ 0 tanh κ 2 θf = θ1 , , θf = θ2 . (51) (52) High energies ETh ≪ E ≪ ∆ In the previous section, we have calculated the density of states at low energies. The result (51) exhibits a delta- peak at zero energy and oscillations which decay at the characteristic energy scale ω0 . These oscillations appear due to repulsion between low-lying levels with energies E and −E . The scale ω0 is the global mean level spacing inside the hole. At higher energies E ≫ ω0 , the effect of level repulsion can be neglected and the density of states is given by the mean-field expression ρ(r, E ) = ν Re cos θ(r, E ), (53) where θ is a solution of the Usadel equation (10) with k = 0. The result (53) can be obtained from the sigma model identity (13) by integrating over Q around the unique global minimum of the action given by the Usadel equation. Supersymmetry ensures that the integration with respect to small fluctuations around the minimum yields unity and hence Eq. (53) follows. The Usadel equation is a complicated non-linear equation that cannot be solved analytically for arbitrary E . At relatively low energies ω0 ≪ E ≪ ETh , the solution can be found perturbatively in E . While linear correction in E to θ is purely imaginary and does not change the density of states, the second order calculation up to terms ∼ E 2 is 7 Ν
Ρ 1.2 1.0 0.8 0.6 0.4 0.2 0.0 0.0 E
E Th 0 2 .5 5 10 20 0.2 0.4 0.6 0.8 1.0 r
R Figure 3: Spatial dependence of the density of states for different values of E . At lowest energies, the result (16) of the main text is reproduced, while at energies well above ETh the density of states is depleted only close to the hole edge within the strip ∼ RpET h /E in agreement with Eq. (55). Remarkably, the density of states is first increased above its normal value before dropping to zero at r = R. necessary to obtain the observable result. The spatial profile of this small energy correction does not factorize as in the main term, see Eq. (15) in the main text. Second order perturbation theory amounts to solving linear differential equations, obtained by linearizing Eq. (10), and leads to very complicated functions. We will not discuss any further details of this approach here. As the energy is increased further and exceeds the Thouless energy, ETh ≪ E ≪ ∆, the Usadel equation admits another approximate solution. The angle θ is small everywhere except a narrow vicinity of the hole boundary. In this vicinity we can neglect the curvature of the superconductor edge and write an approximate one-dimensional Usadel equation D (54) (55) ∂ 2 θ ∂ r2 + 2iE sin θ = 0. With the boundary conditions θ(R) = π/2 and θ(r ≪ R) → 0, solution to this equation reads θ = 4 arctan "(√2 − 1) exp "r−2i R − 1(cid:17)## . E ETh (cid:16) r The density of states is given by Eq. (53) with the above result for θ. We see that θ decreases from π/2 down to exponentially small values in a narrow strip of the width ∼ RpETh /E near the hole boundary. The density of states takes its normal value ν everywhere inside the hole except for this narrow strip where it is depleted down to zero. Spatial integration yields global density of states N (E ≫ ETh ) = πνR2 "1 − (2 − This is the result (20) of the main text. We have also solved the Usadel equation numerically for the whole range of energies below ∆. The spatial profile of the density of states is shown in Fig. 3 and the energy dependence of global DOS is depicted in Fig. 2 in the main text. It perfectly matches both the low-energy limit (up to delta peak and oscillatory term) of Eq. (15) and high-energy asymptotics (56). E # . √2)r ETh (56) Tunneling current and noise Tunneling current and noise can be found using the Landauer formalism instead of the sigma model. Electrons incident from the metallic probe are either reflected normally or experience Andreev reflection. Amplitudes of these processes can be found from the microscopic Hamiltonian of the system and used to calculate the current and higher cumulants, such as noise. The current I (V ) found this way coincides with the result derived from the sigma model and presented in the main text, Eq. (21). The noise, Eq. (23), was found by means of the Landauer approach since the sigma-model consideration of this problem is rather cumbersome. A detailed derivation of the Landauer formalism for our problem and the calculation of current and noise will be presented elsewhere [16]. 8 [17] K. B. Efetov, Supersymmetry in Disorder and Chaos (Cambridge University Press, New York, 1997). [18] E. J. Konig, P. M. Ostrovsky, I. V. Protopopov, A. D. Mirlin, arXiv:1201.6288 . [19] E. Witten, Commun. Math. Phys. 92, 455 (1984). 

0910.4867

2

0910

Spin heat accumulation and its relaxation in spin valves

We study the concept of spin heat accumulation in excited spin valves, more precisely the effective electron temperature that may become spin dependent, both in linear response and far from equilibrium. A temperature or voltage gradient create non-equilibrium energy distributions of the two spin ensembles in the normal metal spacer, which approach Fermi-Dirac functions through energy relaxation mediated by electron-electron and electron-phonon coupling. Both mechanisms also exchange energy between the spin subsystems. This inter-spin energy exchange may strongly affect thermoelectric properties spin valves, leading, e.g., to violations of the Wiedemann-Franz law.

cond-mat.mes-hall

cond-mat

Spin heat accumulation and its relaxation in spin valves T. T. Heikkila,1, ∗ Moosa Hatami,2 and Gerrit E. W. Bauer2 1Low Temperature Laboratory, Aalto University, P.O. Box 15100, FI-00076 AALTO, Finland 2Kavli Institute of NanoScience, Delft University of Technology, 2628 CJ Delft, The Netherlands (Dated: June 2, 2018) We study the concept of spin heat accumulation in excited spin valves, more precisely the effec- tive electron temperature that may become spin dependent, both in linear response and far from equilibrium. A temperature or voltage gradient create non-equilibrium energy distributions of the two spin ensembles in the normal metal spacer, which approach Fermi-Dirac functions through en- ergy relaxation mediated by electron-electron and electron-phonon coupling. Both mechanisms also exchange energy between the spin subsystems. This inter-spin energy exchange may strongly affect thermoelectric properties spin valves, leading, e.g., to violations of the Wiedemann-Franz law. The electric conductance through ferromagnetnormal metalferromagnet spin valves is a function of the mag- netic configuration.1 It reflects the spin accumulation, i.e., the spin (index σ) dependent chemical potential µσ of the normal-metal island. The latter parameterizes the spin dependence of the energy distribution functions fσ(E), whose description also requires spin-dependent temperatures Tσ.3,4 As shown below, these should in gen- eral be interpreted as effective parameters. In this Rapid Communication we describe the pro- cesses affecting the Tσ and through them the thermoelec- tric response in spin valves, which we find to be a sensitive probe for the non-equilibrium state in the non-magnetic spacer. Whereas the spin accumulation relaxes only by scattering processes that break spin rotation invariance such as spin-orbit interaction and magnetic disorder, the spin heat accumulation Ts = T↑ − T↓ is sensitive also to electron-phonon (e-ph) and electron-electron (e-e) inter- actions. Spin-flip scattering in Al, Ag, Cu, or carbon is weak and hardly temperature dependent; the typical spin-flip scattering time τsf is of the order 100 ps,2 which can be much longer than the dwell times in magnetoelec- tronic structures. The inter-spin energy exchange rate due to inelastic effects is strongly temperature depen- dent and above cryogenic temperatures typically dom- inates the direct spin-flip scattering in dissipating the spin heat accumulation. The spin heat accumulation in normal metal spacers should not be confused with the spin (wave) temperature of ferromagnets.5 In a spin valve (Fig. 1), a nonmagnetic island is cou- pled to two ferromagnetic reservoirs with parallel (P) or antiparallel (AP) magnetic alignments. The chemi- cal potential of the left (right) reservoir is µL(R) and the temperature is TL(R). The conductances GL/Rσ and See- beck coefficients SL/Rσ of the contacts between the island and the reservoirs depend on spin σ ∈ {↑,↓}. Biasing the spin valve with either a voltage ∆V = (µR − µL)/e or a temperature difference ∆T = TR − TL gives rise to a spin-dependent energy distribution function fσ(E) of the electrons on the island. As shown below, in the linear response regime this can be described exactly by spin-dependent chemical potentials and temperatures, such that fσ(E) = f0(E; µσ, Tσ), where f0(E; µ, T ) = mL T L IL,s R,sI spin flip ms m-s Ts spin flip e-e, e-ph T-s . . Q L,s Q R,s Tph mR RT FIG. 1: (Color online): Schematic spin-valve biased with a voltage and/or temperature difference. Spin-flip and inelas- tic electron-electron and electron-phonon scattering in the normal metal spacer lead to inter-spin energy exchange and change the thermoelectric characteristics. I and Q stand for the charge and heat currents flowing into the island. Tph is the temperature of the phonon bath. {exp[(E − µ)/(kBT )] + 1}−1 is the Fermi-Dirac distribu- tion function. µσ and Tσ are determined by conservation of charge, spin and energy (see Eqs. (2)). The response matrix of the spin valve Q(cid:19) =(cid:18) G GS (cid:18) I −∆T(cid:19) T GS K (cid:19)(cid:18) ∆V (1) relates the charge and heat currents I and Q to the biases ∆V and ∆T, respectively. Below, we derive expressions for the heat conductance K and thermopower S, in the presence of inter-spin energy exchange and for different magnetic configurations. Ii,σ + Gsf (µσ − µ−σ)/e = 0 Qi,σ + K ↑↓(Tσ − T−σ) + Ke−ph(Tσ − Tph) = 0. The steady state potentials and temperatures can be determined from Kirchhoff's laws for charge and energy for each spin.4 For small e∆V /kB, ∆T ≪ T↑, T↓, Xi=L,R Xi=L,R Here Ii,σ = Giσ(µσ−µi)/e+GiσSiσ(Tσ−Ti) is the charge current for spin σ through contact i, Qi,σ = L0GiσT (Tσ− Ti) + GiσSiσT (µσ − µi)/e is the corresponding heat cur- rent, Giσ and Siσ are the associated charge conductances (2) PSfrag replacements B/(3e2) is the and Seebeck coefficients, and L0 = π2k2 Lorenz number. Spin decay is described by the (inter- )spin conductance Gsf = e2νF Ω/τsf for an island with volume Ω, density of states at the Fermi level νF and spin-flip relaxation time τsf . The term Ke−ph describes the interaction with the phonons at temperature Tph. Inter-spin energy exchange is governed by the spin heat conductance K ↑↓ = L0Gsf T + K ↑↓ e−e, where the first term originates from the spin-flip scattering and the second is due to e-e interactions. We are allowed to discard the spatial dependence of the distribution functions when the diffusion time τD = L2/D in the island with length L and diffusion constant D is shorter than both τsf and the spin thermalization time τst = L0e2νF T Ω/(Ke−ph + 2K ↑↓). The in general lengthy solutions of Eqs. (2) are con- siderably simplified for left-right symmetric conductances and Seebeck coefficients, parameterized by G0 = G↑+G↓, P = (G↑ − G↓)/G0, S0 = (G↑S↑ + G↓S↓)/G0 and P ′ = (G↑S↑ − G↓S↓)/(G0S0) for both junctions. In the antiparallel case the signs of P and P ′ in one of the junc- tions are inverted. In the parallel configuration the heat conductance becomes KP = L0GP T + 2Ke−phr(1 − P 2γ) 1 − P 2γ + Ke−ph/(L0G0T ) (3) and in the antiparallel configuration it is 2Ke−phr . (4) 1 + Ke−ph/(L0G0T ) KAP = L0GP T (1 − P 2γ) + The factor r = (Tph − TL)/(TR − TL) − 1/2 parame- terizes the phonon temperature on the island: If the phonons are poorly coupled to the substrate, as for ex- ample in perpendicular spin valves or in suspended struc- tures, Tph = (T↑ + T↓)/2. For the P configuration this yields r = 0, whereas for the AP configuration we get In the op- r = −K↑↓P/[2(Ke−ph + K↑↓ + L0G0T )]. posite limit r = ±1/2, viz. Tph is fixed to the bath temperature of the left or right reservoir. The coeffi- cient γ = [1 + (Ke−ph + 2K ↑↓)/(L0G0T )]−1 describes inter-spin energy exchange. Factoring out the temper- ature dependence of Ke−ph ∝ T 4 and K ↑↓ e−e ∝ T ν+1 (see the discussion below) yields γ = [1 + (T /Tch,ph)3 + (T /Tch,e−e)ν + 2Gsf /G0]−1, where the characteristic temperatures are Tch,e−ph = [(L0G0T 4)/Ke−ph]1/3, Tch,e−e = [(L0G0T ν+1)/(2K ↑↓ e−e)]1/ν for electron-phonon and electron-electron couplings, respectively. The expo- nent ν depends on the dimensionality (nd) of the sample. We are here mainly interested in 3d samples (ν = 3/2) in which all sample dimensions exceed the thermal coher- ence length ξT =pD/(2πkBT ). SP = S0 and in the antiparallel one6 In the parallel configuration the thermopower satisfies = SAP SP 1 − P P ′ + 2Gsf /G0 + γP (P − P ′ − 2P ′Gsf /G0) (5) The temperature dependence of K and S is plot- ted in Fig. 2 for Ke−ph ≫ Ke−e,L0Gsf T . For T ≪ 1 − P 2 + 2Gsf /G0 . ) T 0 G L ( / K 2.5 2 1.5 1 0.5 0 0 1 2 3 T /Tcr,e−ph 4 2 5 4 3 2 1 0 5 P S / P A S FIG. 2: (Color online): Temperature dependence of the heat conductance K (solid lines, left axis) and thermopower S (dashed line, right axis) of a structurally left-right symmetric spin valve with P = 0.9, P ′ = 0.5 and when the electron- phonon relaxation dominates the inter-spin energy exchange. The lines are plots of Eqs. (3) -- (5) and the symbols have been calculated from the full nonequilibrium distribution function, Eqs. (10) and (11). The results have been calculated for P configuration with r = 0 (circles) and r = 1/2 (squares) and AP configuration with r = 0 (stars) and r = 1/2 (triangles). min(Tch,e−e, Tch,e−ph) ≡ Tch, the device operates as a spin heat valve in which the heat current can be con- trolled by the magnetization configuration. Contrary to the charge conductance, however, the magnetoheat conductance (KP − KAP)/KP vanishes for T ≫ Tch or γ → 0. Thus the presence of inelastic scattering leads to a violation of the Wiedemann-Franz law K = L0GT for T & Tch. The magnetothermopower (SP − SAP)/SP persists provided P 6= P ′.4 The measured heat conduc- tance and thermopower as a function of temperature and magnetic configuration may hence yield unprecedented information on the energy relaxation in normal metals. We now address the characteristic temperatures Tch,e−ph and Tch,e−e. The former can be obtained directly from the Debye form for the heat conductance between electrons and acoustic phonons,7 Ke−ph = 5 2 ΣΩT 4, valid for T ≪ TDebye. Here Σ is the e-ph coupling constant8 and the factor 1/2 takes into account spin degeneracy. The characteristic temperature for electron-phonon cou- pling thus reads Tch,e−ph =(cid:18) πk2 15ΣΩ(cid:19)1/3(cid:18) G0h e2 (cid:19)1/3 B . (6) For T & TDebye, the electron -- acoustic phonon scat- tering and thereby inter-spin energy exchange saturates. Optical phonons start to contribute in this temperature regime, but are disregarded here. The e-e scattering collision integrals with spin- dependent distribution functions contain three terms Ie−e,σ(ǫ) = I σσ (a)(ǫ) + I σ,−σ (b) (ǫ) + I σ,−σ (c) (ǫ), presented by the diagrams in Fig. 3. Processes (b) and (c) induce inter-spin energy ex- change, which can be described in terms of a heat current (a) σ σ (b) σ ¯σ (c) (cid:13) ¯σ σ ǫ + ¯hω ¯hω ǫ′ − ¯hω ǫ + ¯hω ¯hω ǫ′ − ¯hω ǫ + ¯hω (cid:13) ¯hω ǫ′ − ¯hω (cid:13) ǫ σ ǫ′ σ ǫ σ ǫ′ ¯σ ǫ σ ǫ′ ¯σ FIG. 3: Electron-electron scattering vertices. (a) Equal-spin scattering, which equilibrates the electrons but does not ther- malize the spins. (b) Spin conserving scattering and (c) spin exchange scattering, which do thermalize the spins. flowing between two spin ensembles,8 Q↑↓ e−e = νF ΩZ dǫǫ(I ↑↓ (b) + I ↑↓ (c)). (7) The direct spin current due to e-e interaction vanishes in the absence of spin-orbit scattering, R dǫ(I ↑↓ (b) + I ↑↓ (c)) = 0. In 3d, to lowest order in spin particle and heat accumulation, (µ↑ − µ↓) / (µ↑ + µ↓) ≪ 1 and Ts/(T↑ + T↓) ≪ 1, we arrive at Q↑↓ (c))(T↑ − T↓), where e−e ≈ (K ↑↓ (b) + K ↑↓ (a) + I ↑↓ K ↑↓ (b) = 105ζ(7/2)k7/2 B T 5/2 32[2πET (1 + F )]3/2 , K ↑↓ (c) = F C (F + 2) [π2ζ(3/2) + 35 16 ζ(7/2)]k7/2 2[2πET (F + 1)]3/2 B T 5/2 (8a) . (8b) Here C = −1 + (F + 1)3/2, F > −1 is the spin triplet Fermi liquid parameter (F = −1 corresponds to the Stoner instability), ET = D/Ω2/3 is the Thouless energy proportional to the inverse time it takes to diffuse over a length Ω1/3 and ζ(x) is the Zeta function. Summing the two contributions from Eqs. (8) yields the characteristic temperature e2 (cid:19)2/3 ET Tch,e−e = 8π(cid:18) G0h 48F Cπ2ζ(3/2) + 105[6 + F (3 + C)]ζ(7/2)(cid:27)2/3 (cid:26) (F + 1)× π(F + 2) kB (9) . In 1d and 2d structures the spin-flip contribution (c) has an infrared divergence9,10 that needs to be regular- ized. As a result, the inter-spin energy exchange due to e-e scattering becomes stronger and the corresponding Tch,e−e lower. This is especially relevant at low temper- atures and small structures since ξT may exceed 100 nm at T ≈ 1 K. We intend to analyze the resulting inter-spin energy exchange in reduced dimensions in the future. In order to assess the relevance of our results for re- alistic samples we consider a disordered island of a spin valve coupled to the reservoirs via tunnel contacts. For example, with F = −0.3 we get D 0.001 m2 /s(cid:20) 0.1 (µm)3 Ω Tch,e−e ≈ 0.9 K× Te−ph ≈ 1 K×(cid:20) 109 W m−3 K−5 Σ Ω 0.1 (µm)3 G0 0.01 S(cid:21)2/3 0.01 S(cid:21) G0 1/3 3 Making the sample smaller and conductance larger in- creases both characteristic temperatures, but the increase for Tch,e−ph is slower. For Ω = 0.001 (µm)3 and G0 = 1 S we get Tch,e−ph = 22 K whereas Tcr,e−e = 400 K. We may therefore conclude that in spin valves with metallic contacts and 3d spacers the inter-spin energy exchange due to e-e interaction can be neglected. The spin ther- malization rate with F = −0.3 is τst ≈(cid:20) 1 1 25ns(cid:18) T (cid:19)3/2 1 K(cid:19)3/2(cid:18) 0.001 m 2/s 109 W m−3 K−5(cid:19)(cid:21) × 20ns(cid:18) T 1 K(cid:19)3(cid:18) 1047 J−1 m−3 νF D + Σ 1 . The first term comes from e-e scattering and the sec- ond from e-ph scattering. This rate exceeds the spin-flip scattering rate ∼ 10 GHz at temperatures above ∼ 10 K. Above we assume that the electron energy distribution function is well represented by Fermi-Dirac distributions with spin-dependent chemical potentials and tempera- tures. This is not true in general, since fσ(ǫ) has the nonequilibrium form8,11 fσ(ǫ) = GLσfL + GRσfR + νF e2ΩIcoll[fσ, f−σ] GLσ + GRσ , (10) where fL/R = f0(ǫ; µL/R, T ) are the distribution func- tions for the reservoirs and Icoll describes all inelastic scattering events. The charge (n = 0) and heat (n = 1) currents through contact i then become Ii Qi =Xσ Z dǫ(ǫ − µi)n Giσ e1+n (ǫ)(fσ(ǫ) − fi(ǫ)). (11) Thermoelectric effects can be included by adding a weak energy dependence to the conductances, Giσ(ǫ) ≈ G0 iσ[1 + ciσ(ǫ− µi)], and expanding to linear order in ci,σ. Identifying Siσ = eL0ciσT , we recover Eqs. (4) and (5) in the regime e∆V /kB, ∆T ≪ TL, TR ≈ T even in the absence of collisions (i.e., γ = 1). For ciσ = 0 and to linear order in the applied bias, the nonequilibrium dis- tribution (10) is identical to the quasiequilibrium one. Under these conditions, the collision integrals can be cal- culated by replacing the full distribution functions by the quasiequilibrium ones. Numerical solutions of the kinetic equations (see Fig. 2) indicate that in linear response col- lisions and finite ciσ's do not change this conclusion. Beyond linear response spin-dependent temperatures can strictly speaking be invoked only in the presence of strong inelastic scattering such that T↑ ≈ T↓. Nev- ertheless we can define effective electron temperatures that satisfy the standard relation with the thermal en- ergy density in the Sommerfeld expansion:12 Tσ = √6 πkBsZ ∞ −∞ [fσ(ǫ) − 1 + θ(ǫ − µσ)]ǫdǫ. (12) . Proceeding with Fermi-Dirac distributions with effective spin-dependent temperatures and chemical potentials, µσ PSfrag replacements 3 PSfrag replacements eV /kT Tσ/T 1.02 1.01 TE/kT 2.5 / fσ(E) σ T eV /kT Tσ/T 2 1 −1 0 eV /kT ) E ( σ f eV /kT Tσ/T 1 1 0.5 Spin ↓ Spin ↑ 0 −10 0 E/kT 10 T / σ T PSfrag replacements eV /kT Tσ/T E/kT fσ(E) 1.5 1 −20 −10 0 eV /kT 10 20 FIG. 4: (Color online): Spin-dependent effective tempera- ture vs. voltage in an asymmetric spin valve with P = 0.9, P ′ = 0.5 and GR = 0.1GL. The lines are calculated from Eqs. (2) and (13) and the symbols from Eq. (12) for nu- merical solutions of the kinetic equations. The upper curves are for majority, the lower for minority spins, and different strengths of e-e scattering with F = 0: no scattering (solid line and circles), weak scattering with ET = 0.05kB T and GL = 100e2/h (dashed line and squares), and strong scatter- ing with ET = 0.001kB T and GL = 100e2/h (dash-dotted line and stars). Here T denotes the temperature of the reservoirs. Left inset: behavior at low bias with thermoelectric effects cL = 10cR = 0.02/(kT ). Right inset: distribution function at e∆V = 10ET with different strengths of e-e scattering. and Tσ can be obtained from Eqs. (2) by replacing the expression for the charge and heat currents through con- tact i with their nonlinear counterparts, ciσ Gi,σ Iσ = e nµσ − µi + Qi,σ = Gi,σ(cid:2)L0(T 2 σ − T 2 + Giσciσ(µσ − µi)(cid:2)L0(T 2 2 (cid:2)L0e2(T 2 i )/2 − (µ2 σ + T 2 σ − T 2 σ − µ2 i )/2 − (µ2 i ) − (µσ − µi)2(cid:3)o i )/(2e2)(cid:3) σ − µ2 i )/(6e2)(cid:3) . (13) These equations are obtained by a direct integration of Eq. (11) using Fermi-Dirac functions fi(ǫ) and fσ(ǫ). We also have to replace the linear-response forms of the spin 4 mixing terms in Eqs. (2) by their forms far from equilib- rium. For example, for e-e scattering with F = 0 we use Qσ¯σ = 15ζ(7/2)k7/2 −σ )/[16(2πET )3/2]. B (T 7/2 σ − T 7/2 In the absence of collisions and for weak thermoelec- tric effects it can be proven by direct integration that the effective temperatures defined by Eq. (12) agree with those which follow from heat conservation. In Fig. 4 we present a complete numerical solution of the kinetic equa- tions along with the results from the quasiequilibrium heat balance equations from which we conclude that the two approaches for calculating Tσ agree also in the pres- ence of inter-spin energy exchange. Spin heat accumulation cannot be directly measured by two-terminal transport experiments in linear sys- tems. In order to prove the presence of a sizable Ts far from equilibrium it should be probed by spin-selective thermometry, such as a generalization of the tunnel- spectroscopy in Ref. 11, by measuring the shot noise of the spin valve, or through electron spin resonance. In conclusion, we have shown that inter-spin energy exchange in a spin valve affects the temperature and magnetic configuration dependence of its thermoelectric properties. The different thermalization mechanisms can be quantified by characteristic temperatures, Eqs. (6) and (9), above which interaction effects become impor- tant. We introduce the concept of spin heat accumu- lation via the spin-dependent effective electron tempera- tures Tσ in Fermi-Dirac distribution functions, which can be used to describe transport properties beyond the lin- ear response regime. We demarcate the regime in which spin valves can be employed to control heat currents. Other types of operations can be envisaged as well, such as spin-selective cooling of the electrons (see the left inset of Fig. 4). We thank P. Virtanen for discussions. This work was supported by the Academy of Finland, the Finnish Cul- tural Foundation, and NanoNed, a nanotechnology pro- gramme of the Dutch Ministry of Economic Affairs. TTH acknowledges the hospitality of Delft University of Tech- nology, where this work was initiated. ∗ Electronic address: Tero.Heikkila@tkk.fi 1 A. Fert, Rev. Mod. Phys. 80, 1517 (2008); P. Grunberg, ibid 80, 1531 (2008). 2 F. J. Jedema et al., Nature 416, 713 (2002); N. Tombros, ibid. 448, 571 (2007); T. Kimura and Y. Otani, Phys. Rev. Lett. 99, 196604 (2007); J. Bass and W.P. Pratt, J. Con- dens. Matter 19, 183201 (2007). 3 F. Giazotto, F. Taddei, P. D'Amico, R. Fazio, and F. Bel- tram, Phys. Rev. B 76, 184518 (2007). 4 M. Hatami, G.E.W. Bauer, Q. Zhang, and P.J. Kelly, Phys. Rev. Lett. 99, 066603 (2007); Phys. Rev. B 79, 174426 (2009). 5 E. Beaurepaire, J.-C. Merle, A. Daunois, and J.-Y. Bigot, Phys. Rev. Lett. 76, 4250 (1996). from the Onsager-Kelvin relation Π = T GS. The Peltier coefficient Π is measured without a temperature gradient and therefore it cannot depend on r. 7 F.C. Wellstood, C. Urbina, and J. Clarke, Phys. Rev. B 49, 5942 (1994). 8 F. Giazotto, T.T. Heikkila, A. Luukanen, A.M. Savin, and J.P. Pekola, Rev. Mod. Phys. 78, 217 (2006). 9 O.V. Dimitrova and V.E. Kravtsov, JETP Lett. 86, 670 (2007). 10 N.M. Chtchelkatchev and I.S. Burmistrov, Phys. Rev. Lett. 100, 206804 (2008). 11 H. Pothier, S. Gu´eron, N.O. Birge, D. Esteve, and M.H. Devoret, Phys. Rev. Lett. 79, 3490 (1997). 12 N.W. Ashcroft and N.D. Mermin, Solid State Physics, 6 The fact that S is independent of Tph can be understood (Saunders College Publishing, Philadelphia, 1976). 

1212.2407

1

1212

A Cartesian Quasi-classical Model to Nonequilibrium Quantum Transport: The Anderson Impurity Model

We apply the recently proposed quasi-classical approach for a second quantized many-electron Hamiltonian in Cartesian coordinates [J. Chem. Phys. 137, 154107 (2012)] to correlated nonequilibrium quantum transport. The approach provides accurate results for the resonant level model for a wide range of temperatures, bias and gate voltages, correcting for the flaws of our recently proposed mapping using action-angle variables. When electron-electron interactions are included, higher order schemes are required to map the two-electron integrals, leading to semi-quantitative results for the Anderson impurity model. In particular, we show that the current mapping is capable of capturing qualitatively the Coulomb blockade effect.

cond-mat.mes-hall

cond-mat

A Cartesian Quasi-classical Model to Nonequilibrium Quantum Transport: The Anderson Impurity Model Bin Li,(a) Tal J. Levy,(b) David W.H. Swenson,(c) Eran Rabani,(b) and William H. Miller(a) (a) Department of Chemistry and Kenneth S. Pitzer Center for Theoretical Chemistry, University of California, and Chemical Sciences Division, Lawrence Berkeley National Laboratory, Berkeley, California 94720, USA (b) School of Chemistry, The Sackler Faculty of Exact Sciences, Tel Aviv University, Tel Aviv 69978, Israel (c) van 't Hoff Institute for Molecular Sciences, Universiteit van Amsterdam, PO Box 94157, 1090 GD Amsterdam, The Nether- lands Abstract: We apply the recently proposed quasi-classical approach for a second quantized many-electron Hamiltonian in Cartesian coor- dinates [J. Chem. Phys. 137, 154107 (2012)] to correlated nonequilibrium quantum transport. The approach provides accurate results for the resonant level model for a wide range of temperatures, bias and gate voltages, correcting for the flaws of our recently proposed map- ping using action-angle variables. When electron-electron interactions are included, higher order schemes are required to map the two- electron integrals, leading to semi-quantitative results for the Anderson impurity model. In particular, we show that the current mapping is capable of capturing qualitatively the Coulomb blockade effect. I. INTRODUCTION The growing interest in the properties of molecular electronic devices has raised fundamental and concep- tual issues regarding the physics of nanometer scale systems.1-3 Theory faces several important challenges: (a) understanding the coupling of individual mole- cules to macroscopic electrodes under nonequilibrium conditions,4,5 (b) the characterization of the tempera- ture dependence of conductance as well as the role of molecular vibrations, environmental disorder and dis- sipation are of crucial importance,6 (c) electron- electron correlations, that are usually neglected in macroscopic electrodes, by the application of an ef- fective band-structure description of the leads, are very important within the molecule or quantum dot and thus strongly affect the conduction.7,8 Finally, (d) characterizing transport behavior in systems driven by weak and strong electromagnetic fields and the optical properties of such junctions is another difficult issue which must be overcome by theory.9 Generally, transport through nanoscale interacting systems is a many-body out of equilibrium problem that cannot be solved exactly, but for a few simple cases.10-13 Several theoretical methods have been pro- posed to address the transport for such systems, and the different approaches can be loosely classified as numerically exact,14-21 perturbative,9,22,23 and phenom- enological.24-32 While successful, these approaches suffer from several limitations. In particular, the more advanced techniques cannot be extended to cases of more complex environments, an extension required to adequately describe transport through molecules. Parallel to these developments, another paradigm has been formulated based on a semiclassical description 1 of the dynamics of a many-body quantum system.33-35 Recently, Swenson et al.13,36 have adopted a semiclassical mapping approach based on the early work of Miller and White,37 constructing a classical Hamiltonian corresponding to the general second- quantized Hamiltonian operator for a many-electron system in which all the creation and annihilation op- erators for the spin-orbitals were substituted by a set of classical action-angle variables. This scheme was employed to calculate the current of the resonant level model13 and for an inelastic tunneling Holstein mod- el,36 ignoring in both cases electron-electron interac- tions. The mapping approach provides semi- quantitative results for a range of system parameters, i.e. different source-drain voltages, temperatures and electron-phonon couplings. However, it falls short in providing accurate and even qualitative results in sev- eral different limits. For example, it fails to capture the effects of a gate voltage in the noninteracting case,13 or the well-known inelastic tunneling peaks when the phonon bath is characterized by a sharp spectral function.36 Moreover, it does not fully capture the Coulomb blockade38 when electron-electron inter- actions are introduced (see further discussion below). In this paper we present a different mapping proce- dure for the many-electron second quantized Hamil- tonian which is isomorphic to quaternions, based on the approach described in Ref. 41. The occupation numbers and single-electron coupling operators are mapped to the cross product of coordinate and their conjugate momentum vectors, both represented in Cartesian coordinates. This kind of mapping can natu- rally keep all the anti-commutative relationship be- tween the fermionic creation and annihilation opera- tors in the classical Hamiltonian. With the Cartesian expression for the Hamiltonian, the semiclassical ini- tial value representation method34 can also be used to calculate the real time dynamics. For nonequilibrium dynamics, e.g. the noninteracting resonant level mod- el,39 the approach provides accurate results even when a finite gate voltage is applied, correcting for the flaws of our previous action-angle mapping.13 Moreo- ver, it qualitatively captures the Coulomb blockade observed for large electron-electron interactions as applied in the Anderson impurity model.40 The paper is arranged as follows. In section II we de- scribe the Cartesian model for the second quantized many-electron Hamiltonian, the quasi-classical ap- proximation for the occupation numbers, and provide an expression for the current. Results for the classical calculation for the resonant level model and the An- derson impurity model for the different system pa- rameters (i.e. the source-drain voltages, gate voltages, temperatures and one site repulsion) are described in section III. Section IV provides a summary and con- clusions. II. MODEL AND MAPPING PROCEDURE A. Mapping to Cartesian Hamiltonian for the many-electron system We describe the mapping procedure for the Anderson impurity model. The extension to more sophisticated models is straightforward following the principles described below. We note in passing that the limit U → corresponds to the resonant level model. The 0 second quantized many-electron Hamiltonian for the Anderson impurity model is given by:40 ∑ † † † d d Ud d d d H + ε σ σ σ ↑ ↑ ↓ ↓ ↓ =↑ σ ∑ ∑ k L R k L R , , ∈ ∈ =↑↓ =↑↓ σ σ † t c d k k σ σ † c c k k σ σ h.c. (1) ε k + + + = †dσ ( dσ) are the creation (an- In the above equation, ,σ =↑ ↓ nihilation) operators of an electron with spin † kc σ ( kc σ ) are the creation (annihila- and similarly, tion) operators of an electron in mode k of the leads. σε is the one-body energy of the impurity, U is the onsite two-body repulsion, kε is the energy associat- ed with mode k of the left (L) or right (R) noninteracting lead, and kt is the coupling between the quantum dot and mode k . 2 Following the recent work of Li and Miller,41 the anti- commutative relationship between the creation and annihilation operators, isomorphic to quaternions, can be mapped to the cross product of the 2-dimension coordinate ( r ) and conjugate momentum ( p ): 1 = + 2 r k σ + × 1 − 2 p k σ i j k k σ σ 1 = + 2 † c c k k σ σ 1 2 and  x p k σ y k , σ − y p k σ x , k σ (2) + = ( ) × x , + x y , ↓ x , ↓ − − y k , σ ) + ) + H = , k σ  x p σ y p σ y p ↑ y p ↓ , ↑ x p k σ † c d k σ σ r k σ y p − x , σ σ ) ( x p ↓ y p k σ 1 − i j j i + k k σ σ σ σ 2 p = × k σ y p − k x , σ σ † d c k σ σ p r + σ σ x p k y , σ σ (3) kj σ are the basic elements of qua- in which ki σ and ternions. In this notation, the Anderson impurity Hamiltonian takes the form: ( ∑ x p ε y , σ σ σ =↑↓ σ ( U x p y ↑ ( ∑ ε k k L R , ∈ =↑↓ σ ∑ k L R , ∈ =↑↓ σ (4) 2 from In the above equation, we have subtracted a 1 the classical mapping of the occupation number oper- † c c n = in Eq. (2) to include the Langer ator k k k σ σ σ correction. In what follows, we will mainly be interested in the calculation of the time dependent current. In the above notation, the current from the left/right lead to the dot is given by: x p y k , σ σ y p k x , σ σ − ) y p σ x p σ , ↑ − y k , σ ) . ( t k + x k , σ − y k , σ + x k , σ − I L R ( ) = − d dt = t k ∑ k L R ( ∈ =↑↓ σ ) ) ∑ k L R ( ∈ =↑↓ σ y p σ x p k σ y k , σ − y p k σ x k , σ + x p σ x k , σ − x p x k , σ σ − y p k y , σ σ y k , σ (5) and the total current is I = ( I L − I R ) / 2 . B. Special treatment for the two-body terms As will become clear below, the above mapping com- bined with a quasi-classical description of the dynam- ics fails to capture the well-known Coulomb blockade effect at low temperatures and high values of U . This is a result of the fact that within the quasi-classical approximation the occupation numbers assume a con- tinuum value and thus, when the dot is occupied by a "fraction" of an electron with spin up it does not block completely the path of an electron with spin down. One way to overcome this faulty is to increase the strength of the interactions between electrons of dif- ferent spins for fractional occupation by taking ad- vantage of the anti-commutation relation of the crea- tion and annihilation operators of fermions ) = and the relation ( λ † † † † , d d d d d d d d 1 + = σ σ σ σ σ σ σ σ where λ is an arbitrary positive integer. Thus, we ↓ in Eq. (1) with propose to replace the term † † Ud d d d ↓ ↑ ↑ ( ) ) ( λ λ † † . Quantum mechanically, this is d d 2 1 d U d ↓ ↓ ↑ ↑ an exact identity that has no effect on the transport properties of the system or on its thermodynamic state. For the mapped variables, we rewrite the Hamil- tonian as: ∑ σ =↑↓ ( U x p y ↑ ( ∑ ε k k L R , ∈ =↑↓ σ ∑ k L R , ∈ =↑↓ σ y p x , σ σ ) ( λ x p 1 ↓ y p k σ y p , ↑ ↑ x p k σ ( x p ε y , σ σ σ x p k y , σ σ y p k x , σ σ y , ↓ ) + y p σ x p σ y p ↓ x , ↑ − (6) ) λ 2 ) , H y k , σ y k , σ x k , σ x k , σ ) ( = − + − − − + − + t ↓ k x , where 1λ and 2λ are positive integer. Here, since the mapping is not exact, the results will depend on the choice of 1λ and 2λ . However, we find that the cur- rent converges as we increase 1λ and 2λ to about 20, which is the choice for most of the results reported here (unless otherwise stated). The physical significance of the special treatment of the two-body terms is simple: Since the occupations ( ) and ( ) y p x p y p x p − − can assume x y x y , , , , ↓ ↓ ↓ ↓ ↑ ↑ ↑ ↑ any value between 0 and 1, using higher values of 1,2λ will result in a sharp change of the Coulomb re- pulsion as the occupations approach the value of 1. 3 x  σ y ε = − σ σ the blockade is We find numerically that for 1,2 20 λ < not significant enough while for larger values the re- sults are numerically converged (see Figure 4 below). C. Quasi-classical approximation In the quasi-classical procedure, the dynamics of the Cartesian variable follow from Hamilton's equations of motions. For the Anderson impurity model, these are (assuming kt is real): ∑ L R k , ∈ ∑ t x k k σ L R k , ∈ ∑ ε − , σ σ L R k , ∈ ∑ t p k L R k , ∈ t y − σ k t x σ k t p − k t p k ( y p x p Uy − x y , , σ σ σ σ σ ( ) y p x p Ux − x y , , σ σ σ σ σ ( y p x p − − x y , , , σ σ σ σ σ ( ) y p x p − x y , , , σ σ σ σ σ p ε = x , , σ σ σ y ε = − k k σ x ε = + k k σ p = − ε k p ε = k y x ε =  σ σ σ y k , σ + t p k t y k Up Up = − y k , σ x k , σ x k , σ k σ p  p  + + − − + + p , σ , σ , σ ) ) x y y y x y x x  k σ y  k σ p  p  y k , σ (7) k σ x , where ,σ =↑ ↓ for ,σ =↓ ↑ . To complete the mapping procedure we need to define the initial conditions for all classical phase-space var- iables. We follow a similar route discussed in Ref. 13 for the action-angle mapping, which recovers the cor- t = ). Since we are inter- 0 rect statistical behavior (at ested in a non-correlated initial state with thermally populated leads and an unpopulated quantum dot, we can populate each degree of freedom independently. We enforce quantum statistics on the initial conditions for each degree of freedom by setting the initial occu- pation to either 0 or 1 such that its expectation value, averaged over the set of initial conditions, satisfies the Fermi-Dirac distribution. Specifically, we choose a kσξ in the interval [0 1] , and then random number select the occupation of mode kσ according to: ( ) 1 > + ( ) 1 ≤ +     where /L Rµ is the chemical potential of the left/right lead. The phase space variables are then sampled ac- cording to:42-45 ( β µ − e  k L R / ( β µ − e  k L R / ξ k σ ξ k σ n k σ (8) 0 1 = 1 − 1 − ) ) x k σ = r k σ cos θ k σ p xk = p r k , σ cos θ k σ − n k σ θ k σ sin r k σ in the x direction, and y r sin = θ k k σ σ k σ p yk = p r k , σ sin θ k σ + n k σ cos θ k σ r k σ (9) (10) in the y direction. Since the Cartesian mapping introduces 2 classical phase space variables for each creation and annihila- tion operators, this leads to freedom in choosing the initial conditions, as long as the population given by n x p y p satisfies Eq. (8). The rela- − = k k k y k k x , , σ σ σ σ σ tions defined by Eqs. (9) and (10) guarantees that this constraint is not violated regardless of the values of ,r kp σ . In the applications reported below we kr σ and take r kp σ = . This choice provides rapid kr σ = and 0 1 , convergence of the calculated currents with respect to the number of trajectories. Other choices may be based on a Gaussian sampling of the radial position and momentum, etc. D. Spectral density and parameters We adopt a wide band limit to model the spectral den- sity of the leads with a sharp cutoff at high and low energy values: A ) − 1 2 B ) ) − A + B 1 2 J e e + + = ( ε k ( ε k . (11) / ( ε k L R Γ L R / ( ) ) ( 1 1 In the above, B is the width spectral function, /L RΓ determines the strength of coupling to the left or right leads, and A relates to the sharpness of the cutoff. For 1 2 Γ . In terms of the are given by: 5A = Γ , and Γ = Γ + Γ , B = 20 L R spectral function, the couplings kt the results described below we use Γ = Γ = L R , k J t = , ( ) ε ε ∆ k 2 π where ε∆ is the discretization interval. In our study of the Anderson impurity model we limit ourselves (without loss of generality) to the common (12) 4 dot states are given by case referred to as the symmetric version, in which the 1 U ε ε↑ ↓= 2 the transient current as a function of temperature ( ) , source-drain bias ( ) 1 β− = Bk T eV µ µ = − L R SD and the value of Γ . U = , 8 ] [0, . We study = − III. RESULTS We solve Hamilton's equation of motion for the mapped variables with initial conditions sampled from the Fermi distribution and the quasi-classical proce- dure described above. The occupation of the dot is taken to be 0 initially. We use an adaptive time step Runge-Kutta algorithm for the propagation of trajec- tories. For each lead, 800 spin orbitals are used (400 modes for the ↑ spin and 400 modes for the ↓ spin). For most of the results presented below, convergence is achieved with 106 trajectories. A. The resonant level model Figure 1: Transient currents for the resonant level model. Left, middle and right panels show results for different source-drain voltages, different gate voltages, and different temperatures, respectively. Lower, middle and upper panels correspond to the left, right and total currents, respectively. Solid lines (hidden by the symbols) are the exact results derived in Ref. 13. Symbols are the results obtained from the Cartesian mapping. In Figure 1 we plot the left current (lower panels), right current (middle panels), and total current (upper panels) for the case where 0U = , corresponding to the resonant level model. Since the different spins do not interact in this limit, we only consider one spin and ignore the other. Left, middle and right panels show the results for different source-drain voltages, gate voltages and temperatures, respectively. The choice of parameters is identical to the case studied by us using the action-angle mapping.13 As clearly can be seen, the Cartesian mapping pro- vides accurate results for both the right/left currents and for the total current in comparison to exact re- sults.13 In fact, the Cartesian mapping is accurate within the statistical error for all range of source-drain and gate voltages studied and for all temperatures studied. The mapping captures the transient currents as well as their steady-state values as time approaches . tΓ → 5 In comparison to the results obtained using the action- angle mapping presented in Ref. 13, this is a signifi- cant improvement, particularly when a gate voltage is applied. This is partially expected given the perfor- mance of the Cartesian mapping for the transition probability and energy spectrum of a simple 2-spin orbital and 4-spin orbital models.41 B. Anderson impurity model The time-dependent total current for the Anderson impurity model obtained from the Cartesian mapping is shown in Figure 2 for two temperatures T = Γ (left (right panel). Each panel con- panels) and T = Γ / 5 tains results for a set of different source-drain voltag- es. In all cases, the current saturates at high values of sdV , as it should. The behavior of the current is qualitatively different as we change the onsite repulsionU . For small values U the current resembles that of the non-interacting case, shown in Figure 1. As U increases, a pro- nounced oscillation in the current at early times is ob- served even at small source-drain voltage, signifying the repulsion between electrons of different spins. This is translated to a suppression of the steady-state current (inferred from the limit ) at small tΓ → 5 bias voltages. The suppression is more pronounced at lower temperatures. Comparing the results at different temperatures for a given source-drain bias, we find that at low values of sdV the steady state current is higher for the higher temperatures for large values of U . Above a blockade voltage this behavior is reversed and the steady state value of the current at low temperatures is larger than the corresponding value at high temperatures. This 5 crossover behavior captured by the quasi-classical approximation (and to a lesser extent by the action- angle mapping not shown here) is consistent with the quantum mechanical result. In fact, this is one of the hallmarks of Coulomb blockade, as further discussed below. Figure 2: The total current as a function of time for the Anderson impurity model obtained from the Cartesian mapping. Left panels show the results T = Γ for T = Γ and right panels for / 5 . Black, red, green, blue, ma- eV = 2, 4, 6, 8, 10, 12 and 14 genta, cyan and orange correspond to sd in units of Γ , respectively. In Figure 3 we compare the steady state values of the current calculated from the quasi-classical mappings to quantum mechanical results obtained within an equation of motion approach to the nonequilibrium Green's function formalism, where all 2-body correla- tions between the system operators were included.46 This approach is known to provide accurate results in the present regime of parameters and temperatures.47 In fact, it can also capture the Kondo resonances when higher order correlations in the leads are included.48 When the onsite interaction is turned on the quasi- classical Cartesian mapping is not as accurate as it was in the non-interacting case (Figure 1). However, it captures qualitatively the dependence of the current on the source-drain bias for both temperatures studied. At the higher temperature, we find that the current increases roughly linearly with sdV before it bends and saturates at high values of sdV . The quasi-classical Cartesian mapping performs better at low values of .U It slightly underesti- sdV and at higher values of mates the current at saturation. In this regime of pa- rameters the action-angle mapping also performs well, .U particularly for small values of It is important to note that the special treatment of the two-body terms as described in Section II.B is essen- tial to describe the blockade phenomena. In fact, when we employ the mapped Hamiltonian given by Eq. (4) rather than that given by Eq. (6), the quasi-classical approach fails to reproduce the Coulomb blockade U< . The and the current increases even below sdV effect of changing 1λ 2λ on the steady state cur- and rent is shown in Figure 4 for the special case . As is clearly evident, only when λ is 2λ λ λ = ≡ 1 sufficiently large, a clear blockade is developed. The results converge and change very slightly at . 20λ> Figure 3: Steady state values of the current for the Anderson impurity T = Γ / 5 model. Upper and lower panels are for T = Γ and , respec- tively. Solid lines represent quantum mechanical results based on a nonequilibrium Green's function approach, circles correspond to results of the quasi-classical Cartesian mapping, and triangles are the results of the action-angle mapping of Ref. 13, both calculated with 20 λ λ λ≡ = = . 1 2 When the temperature is reduced, the current assumes an "S"-shape voltage dependence, with a blockade that increases with U The mid-point in the current U= . Below this versus bias voltage occurs when sdV value, the current is blocked and only when the bias U= increases above a conducting channel opens sdV , up and the current increases rapidly. This Coulomb blockade is qualitatively captured by the quasi- classical Cartesian mapping, and, to a lesser extent, by the action-angle mapping. Importantly, the Cartesian mapping captures the blockade and the overall shape of the current versus bias voltage. Again, the Carte- sian approach slightly underestimates the current as it levels off. 6 Figure 4: Steady state current versus the source-drain bias voltage for 2λ λ λ = ≡ calculated from the quasi-classical different values of 1 8U = Γ and Cartesian mapping. The remaining parameters are: T = Γ / 5 . IV. CONCLUSIONS We have presented an approach based on a Cartesian mapping of the many-electron Hamiltonian41 to calcu- late the nonequilibrium properties of the resonant lev- el model and the Anderson impurity model. The map- ping keeps track of the anti-commutation relation of the creation/annihilation operators, required for Fer- mi-Dirac statistic. It also provides a suitable frame- work for a quasi-classical approximation which ac- counts for the correct thermodynamics at t = and 0 for a semiclassical initial value treatment, which will be the subject of future study. The approach provides excellent agreement for the resonant level model (non-interacting limit) in com- parison to exact quantum mechanical calculations for a wide range of model parameters, including different temperatures and bias voltages. It also captures quan- titatively the behavior of the current at different gate voltages, correcting for the flaws of our previous mapping based on the Miller-White37 action-angle approach. When we turn on the two-electron interactions the approach is no longer numerically exact. It performs quite well at high temperatures (but so does the ac- tion-angle mapping of Ref. 13). To capture the Cou- lomb blockade at lower temperatures and higher val- ues of the onsite repulsion U , we have modified the two-body mapped interaction, using identities of crea- tion/annihilation operators. The modified Hamiltonian captures the Coulomb blockade qualitatively and more importantly, the temperature dependence of the current below and above the blockade. Acknowledgments We would like to thank Guy Cohen for fruitful dis- cussions. This work was supported by the National Science Foundation Grant No. CHE-1148645 and by the Director, Office of Science, Office of Basic Ener- gy Sciences, Chemical Sciences, Geosciences, and Biosciences Division, U.S. Department of Energy un- der Contract No. DE-AC02-05CH11231, by the FP7 Marie Curie IOF project HJSC, and by the US-Israel Binational Science Foundation. TJL is grateful to the Azrieli Foundation for the award of an Azrieli Fel- lowship. We also acknowledge a generous allocation of supercomputing time from the National Energy Research Scientific Computing Center (NERSC) and the use of the Lawrencium computational cluster re- source provided by the IT Division at the Lawrence Berkeley National Laboratory. References 1 S. Datta, Electronic transport in mesoscopic systems (Cambridge University Press, 1995). 2 Y. Imry, Introduction to Mesoscopic Physics, 2nd ed. (Oxford University Press, 2002). 3 S. Datta, Quantum transport : atom to transistor (Cambridge University Press, 2005). 4 H. Basch and M. A. Ratner, J. Chem. Phys. 119, 11926 (2003). 5 H. Basch, R. Cohen, and M. A. Ratner, Nano Lett. 5, 1668 (2005). 6 M. Galperin, A. Nitzan, and M. A. Ratner, Phys. Rev. B 76, 035301 (2007). 7 M. Dorogi, J. Gomez, R. Osifchin, R. P. Andres, and 7 R. Reifenberger, Phys. Rev. B 52, 9071 (1995). 8 V. Mujica, M. Kemp, A. Roitberg, and M. Ratner, J. Chem. Phys. 104, 7296 (1996). 9 M. A. Ratner, A. Nitzan, and M. Galperin, J. Phys.: Condens. Matter 19, 103201 (2007). 10 A.-P. Jauho, N. S. Wingreen, and Y. Meir, Phys. Rev. B 50, 5528 (1994). 11 A. Schiller and S. Hersheld, Phys. Rev. B 51, 12896 (1995). 12 Y. Wang and J. Voit, Phys. Rev. Lett. 77, 4934 (Dec 1996). 13 D. W. H. Swenson, T. Levy, G. Cohen, E. Rabani, and W. H. Miller, J. Chem. Phys. 134, 164103 (2011). 14 L. Mühlbacher and E. Rabani, Phys. Rev. Lett. 100, 176403 (2008). 15 P. Werner, T. Oka, and A. J. Millis, Phys. Rev. B 79, 035320 (2009). 16 P. Werner, T. Oka, M. Eckstein, and A. J. Millis, Phys. Rev. B 81, 035108 (2010). 17 M. Schiró and M. Fabrizio, Phys. Rev. B 79, 153302 (2009). 18 E. Gull, D. R. Reichman, and A. J. Millis, Phys. Rev. B 82, 075109 (2010). 19 G. Cohen and E. Rabani, Phys. Rev. B 84, 075150 (2011). 20 H. Wang and M. Thoss, J. Chem. Phys. 131, 024114 (2009). 21 H. Wang, I. Pshenichnyuk, R. Haertle, and M. Thoss, J. Chem. Phys. 135, 244506 (2011). 22 H. Haug and A. P. Jauho, Quantum Kinetics in Transport and Optics of Semiconductors (Springer, Germany, 1996). 23 S. Datta, Superlattices Microstruct. 28, 253 (2000). 24 J. Lehmann, S. Kohler, V. May, and P. Hanggi, J. Chem. Phys. 121, 2278 (2004). 25 J. N. Pedersen and A. Wacker, Phys. Rev. B 72, 195330 (2005). 26 L. Siddiqui, A. W. Ghosh, and S. Datta, Phys. Rev. B 76, 085433 (2007). 27 G. Cohen and E. Rabani, Mol. Phys. 106, 341 (2008). 28 V. May and O. Kühn, Phys. Rev. B 77, 115439 (2008). 29 V. May and O. Kühn, Phys. Rev. B 77, 115440 (2008). 30 M. Leijnse and M. R. Wegewijs, Phys. Rev. B 78, 235424 (2008). 31 M. Esposito and M. Galperin, Phys. Rev. B 79, 205303 (2009). 32 M. Esposito and M. Galperin, J. Phys. Chem. C 114, 20362 (2010). 33 W. H. Miller, Adv. Chem. Phys. 25, 69 (1974). 34 W. H. Miller, J. Phys. Chem. A 105, 2942 (2001). 35 W. H. Miller, J. Chem. Phys. 125, 132305 (2006). 36 D. W. H. Swenson, G. Cohen, and E. Rabani, Mol. Phys. 110, 743 (2012). 37 W. H. Miller and K. A. White, J. Chem. Phys. 84, 5059 (1986). 38 C. W. J. Beenakker, Phys. Rev. B 44, 1646 (1991). 39 R. Landauer, IBM J. Res. Dev. 1, 223 (1957). 40 P. W. Anderson, Phys. Rev. 124, 41 (1961). 41 B. Li and W. H. Miller, J. Chem. Phys. 137, 154107 (2012). 42 H.D. Meyer, W.H. Miller, J. Chem. Phys. 71, 2156- 2169 (1979). 43 H.D. Meyer, W.H. Miller, J. Chem. Phys 70, 3214- 3223 (1979). 44 G. Stock, M. Thoss, Phys. Rev. Lett. 78. 578-581 (1997). 45 W.H. Miller, J. Phys. Chem. A 105, 2942-2955 (2001). 46 T.J. Levy and E. Rabani, unpublished. 47 B. Song B, D.A. Ryndyk and G. Cuniberti, Phys. Rev. B 76, 045408 (2007). 48 Y. Meir, N.S. Wingreen and P.A. Lee, Phys. Rev. Lett. 66 3048 (1991). 8 

1108.5917

1

1108

Ab Initio Theory of Scattering-Independent Anomalous Hall Effect

We report on first-principles calculations of the side-jump contribution to the anomalous Hall conductivity (AHC) directly from the electronic structure of a perfect crystal. We implemented our approach for a short-range scattering disorder model within the density functional theory and computed the full scattering-independent AHC in elemental bcc Fe, hcp Co, fcc Ni, and L1o FePd and FePt alloys. The full AHC thus calculated agrees systematically with experiment to a degree unattainable so far, correctly capturing the previously missing elements of side-jump contributions, hence paving the way to a truly predictive theory of the anomalous Hall effect and turning it from a characterization tool to a probing tool of multi-band complex electronic band structures.

cond-mat.mes-hall

cond-mat

Ab Initio Theory of Scattering-Independent Anomalous Hall Effect Jurgen Weischenberg,1 Frank Freimuth,1 Jairo Sinova,2 Stefan Blugel,1 and Yuriy Mokrousov1, ∗ 1Peter Grunberg Institut & Institute for Advanced Simulation, Forschungszentrum Julich and JARA, 52425 Julich, Germany 2Department of Physics, Texas A&M University, College Station, Texas 77843-4242, USA (Dated: June 2, 2018) We report on first-principles calculations of the side-jump contribution to the anomalous Hall conductivity (AHC) directly from the electronic structure of a perfect crystal. We implemented our approach for a short-range scattering disorder model within the density functional theory and computed the full scattering-independent AHC in elemental bcc Fe, hcp Co, fcc Ni, and L10 FePd and FePt alloys. The full AHC thus calculated agrees systematically with experiment to a degree unattainable so far, correctly capturing the previously missing elements of side-jump contributions, hence paving the way to a truly predictive theory of the anomalous Hall effect and turning it from a characterization tool to a probing tool of multi-band complex electronic band structures. PACS numbers: 72.25.Ba, 72.15.Eb, 71.70.Ej The anomalous Hall effect (AHE) in ferromagnets is one of the most celebrated transport phenomena in solid- state physics [1]. It has been researched intensely in the past decade after it was realized that the intrinsic con- tribution (IC) could be interpreted in terms of the Berry phases of Bloch electrons in a solid [2, 3]. For almost ten years the IC was the only one, which could be ac- cessed in density functional theory (DFT) calculations of the AHE [4–6]. The ability to estimate the impurity- driven (i.e. extrinsic) contributions to the AHE has been remarkably limited so far, thus hindering the predictive power in understanding and engineering the AHE trans- port properties of real materials. Besides the IC, σint, there are two extrinsic disorder- driven contributions to the anomalous Hall conductivity (AHC). In metallic systems they can be distinguished ac- cording to their parametric dependency on the impurity concentration n, with skew-scattering conductivity, σsk, proportional to 1/n [7], and the side-jump contribution (SJC), σsj, independent of impurity concentration [8]. The fact that the SJC, although originating from im- purities, does not depend on their concentration, makes it one of the most challenging electron scattering mecha- nisms to understand and suggests a close relation to the intrinsic contribution of the AHE. This behavior arises in metallic systems from the 1/ǫF τ expansion of the trans- port coefficients in linear response for coupled multi-band systems. Consequently, since the SJC does not depend on the disorder strength, it cannot be easily separated from the IC in low temperature dc measurements [3]. So far, only for an L10-ordered FePd ferromagnetic al- loy clear evidence has been presented that the side-jump contribution can dominate over other mechanisms of the AHE in a wide temperature range through a comparison of theoretical IC and experimentally measured AHC [9]. Previous numerical ab initio treatments have mainly concentrated on the IC [3] and defined the SJC by tak- ing the zero disorder limit of coherent potential approxi- mation disordered alloys calculations [10]. Within such a treatment the indirect computation of the SJC presents a significant computational challenge considering also that the exact knowledge of the disorder potential in the sys- tem is necessary in this case. Since very often the exper- imental data are obtained on samples with unknown im- purity content and disorder type, it is highly desirable to be able to evaluate the SJC explicitly from the electronic structure of a perfect crystal. A direct computation of the SJC in models with disorder was not feasible until recently when it was shown that, assuming short-range uncorrelated disorder model, the SJC may be indeed cal- culated directly from the ideal electronic structure of a crystal without any disorder [11], rendering possible a rigorous numerical study that can be fully compared to experimental results and can set the stage to a truly pre- dictive theory of the AHE. While the validity of derived expressions has been demonstrated for simple models, values for the scattering-independent SJC in fundamen- tal ferromagnetic materials such as Fe, Co or Ni, have not been obtained so far. In this Letter we report on calculations from first prin- ciples of the values of the scattering-independent side- jump conductivity in elemental bcc Fe, hcp Co, fcc Ni, as well as ordered FePd and FePt alloys directly from the electronic structure of their pristine crystals. Our calculations unambiguously show that the calculated val- ues of the total scattering-independent contributions, IC and SJC, agree systematically with experiments to a level that was not reached before in these materials. More importantly, the inclusion of the scattering-independent SJC accounts consistently for the discrepancy between the IC and the measured values in a very non-trivial fash- ion. We analyze the side-jump as a Fermi surface prop- erty and demonstrate that it shows a strong anisotropy with respect to the magnetization direction in the crystal, even more pronounced than that found for the intrinsic AHC [12, 13]. Additionally, we demonstrate the impor- tance of correlation effects in describing the AHE in fcc Ni correctly. The starting point for the theory of the scattering- independent side-jump [11] are the retarded Green's func- tion in equilibrium and the Hamiltonian H of a general multiband noninteracting system in three spatial dimen- sions. At the first step we expand the self-energy of the system Σeq in powers of potential V (r), which describes scattering at impurities. For a short-range scattering dis- order model, scalar delta-correlated Gaussian disorder or delta-scattering uncorrelated disorder, the contribution to the self-energy which is of first order in V (r) vanishes, since one can assume that hV (r)i = 0 or else absorb hV (r)i into the Hamiltonian, a procedure which results in a simple shift of the energy levels. Further, inserting the expression for the self-energy within these simple dis- order models into appropriate equations for the current densities derived following the Kubo-Streda formalism, rotating into eigenstate representation and keeping only the leading order terms in the limit of vanishing disorder parameter V, i.e. ignoring skew-scattering contributions, the scattering-independent part of the AHE conductivity may be written as σ(0) = σint + σsj, where σint ij = e2  Z d3k (2π)3 ImXn6=m (fn − fm) vnm,i(k)vmn,j(k) (ωn − ωm)2 (1) can be recovered as the intrinsic contribution [4]. In this expression indices n and m run over all bands with occupations fn and fm, respectively, vnm,i are the ma- H, and trix elements of the velocity operator vi = ∂ki ωn(k) = εn(k)/ with εn(k) as band energies. The scattering-independent SJC to conductivity σ(0) reads for inversion-symmetric systems: σsj ij = e2  Xn Z d3k (2π)3 Re Tr(cid:26)δ(εF − εn) γc [γc]nn × ×(cid:20)SnAki (1 − Sn) ∂εn ∂kj − SnAkj (1 − Sn) ∂εn ∂ki(cid:21)(cid:27). (2) Here εF is the Fermi energy and the imaginary part of the self-energy ImΣeq = −Vγ is taken to be in the eigenstate representation, i.e. γc = U †γU , with γ = 1 2 Xn Z d3k (2π)2 U SnU † δ(εF − εn), (3) U as the k-dependent unitary matrix that diagonalizes the Hamiltonian at point k, [U †H(k)U ]nm = εn(k)δnm, (4) Sn is a matrix that is diagonal in the band indices, [Sn]ij = δij δin, and the so-called Berry connection ma- trix is given by Ak = iU †∂kU [11]. Not included in expression (2) are the vertex corrections, which vanish 2 TABLE I: Anomalous Hall conductivities for bcc Fe and hcp Co in units of S/cm for selected high-symmetry orientations of the magnetization. σint, σsj and σint+sj stand for IC, SJC and their sum, respectively. The experimental values are for the scattering-independent conductivity. Fe σint σsj σint+sj Exp. [20] [001] 767 111 878 1032 [111] 842 178 1020 [110] Co σint 810 σsj 141 σint+sj 951 Exp. [12] c axis ab plane 100 −30 70 150 477 217 694 813 for an inversion-symmetric system in the Gaussian dis- order model. However, note that in contrast to the orig- inal formula as presented in Eq. (3) of Ref. [11], our expression for σsj ij is manifestly antisymmetric. For the Rashba model it reduces to the original form of Eq. (3) in Ref. [11]. It is important to note that the SJC in the short-range disorder model, Eq. (2), is solely determined by the electronic structure of the pristine crystal and thus directly accessible by ab initio methods. In practice, we replace the integrals in Eqs. (1) and (2) by a discrete sum over a finite number of k-points in the Brillouin zone (BZ). To reduce the computational cost we adopt the method of Wannier interpolation [5, 14], which employs the description of the electronic structure in terms of maximally-localized Wannier functions (ML- WFs), to evaluate Eqs. (1) and (2) for bcc Fe, hcp Co, fcc Ni, as well as L10 FePd and FePt. The electronic structure calculations were performed with full-potential linearized augmented plane-wave method as implemented in the Julich DFT code FLEUR [15] within the generalized gradient approximation (GGA). We used the plane-wave cut-off Kmax of 4.0 bohr−1 and 16000 k-points for self- consistent calculations. Spin-orbit coupling was included in the calculations in second variation. We constructed a set of 18 MLWFs per atom using the Wannier90 code [16] and our interface between FLEUR and Wannier90 [17]. We present the results of our calculations of the in- trinsic and side-jump AHC for Fe, Co, Ni, FePd and FePt in Tables I, II, and Fig. 1 for high-symmetry di- rections of the magnetization M in the crystal. These results are compared to experimental values, from which the skew scattering contribution was either explicitly sub- stracted [9, 18], or can be safely ignored at higher tem- peratures [12, 19, 20]. We first analyze the results for bcc Fe. For M along [001], the IC in Fe accounts to roughly 75% of the known experimental value of ≈1000 S/cm [6, 20]. This compar- ison gets slightly improved considering that the experi- mental value averages over crystals with different orien- tation. Taking the SJC into consideration improves the value of the AHC in Fe significantly for all magnetization directions, with the angle-averaged σint + σsj of about 90% of the experimental conductivity. In hcp Co, the magnitude of the SJC for M along the c axis is as large as 217 S/cm, with the total AHC of 694 S/cm, very close to the experimental value of about 800 S/cm. For M in the basal ab plane the SJC is small and negative, bring- ing thus the intrinsic value down to ≈70 S/cm, somewhat away from the experimental value of about 150 S/cm. One has to keep in mind, however, that the experimental values for hcp Co are approximate [12]. A significant improvement upon including the SJC is also evident for the more complex ordered FePd and FePt alloys in their L10 phase with M along the [001] axis. For FePd, the IC is very small, of about 130 S/cm, while the side-jump AHC is twice as large and of the same sign, resulting in a value of the total AHC much closer to ex- periment, Table II. As follows from our calculations, in FePd the AHC is dominated by σsj, in accordance to an earlier indirect prediction [9]. On the other hand, in FePt with Mk[001], the IC is much larger, while the SJC is half the value of that in FePd. This is again in agreement to Ref. [9], in which such a crossover between the intrin- sic and side-jump conductivities, appearing within the Dirac model as well [21], was attributed to different SOI strength of Pd and Pt atoms. As far as the comparison to experiments is concerned, also in FePt adding the cal- culated SJC to the IC brings the total AHC within the range of experimentally observed values for samples of [001]-magnetized L10 FePt with high degree of ordering, S > 0.7, and different sample thickness [9, 18]. The case of fcc Ni presents a special challenge, since in this material the GGA value of the intrinsic AHC is much larger than the measured scattering-independent value, implying a sizable σsj with the sign opposite to the IC [6]. From our calculations of the IC in fcc Ni we obtain values which lie between −2000 and −2500 S/cm (Fig. 1 at U = 0), depending on the direction of M, while the experi- mental value resides in the vicinity of −640 S/cm [6, 22]. The calculated values for the scattering-independent SJC in fcc Ni presented in the same figure (at U = 0) lie be- tween −100 and 400 S/cm, and thus cannot explain the large discrepancy between theory and experiment. The description of the electronic structure of Ni within con- ventional DFT is well-known to be inaccurate, however. Several attempts aiming at improving the GGA values for quantities such as magnetocrystalline anisotropy en- ergy, spin-wave dispersion etc., were made in the past (e.g. Refs. [23], [24] and references therein), proving the importance of correlation effects in this material and the sensitivity of calculated quantities on the shape of its Fermi surface. TABLE II: Same as in Table I for L10 FePd and FePt alloys. FePd σint σsj σint+sj Exp. [9] [001] 133 263 396 806 [110] FePt 280 280 560 σint σsj σint+sj Exp. [9, 18] [001] 818 128 946 900 ÷ 1267 [110] 409 220 629 3 -800 -1200 -1600 -2000 -2400 ) m c / S ( C H A [001] [110] [111] 500 400 300 200 100 0 -100 -200 -300 0 1 2 3 4 0 1 2 U (eV) 3 4 FIG. 1: (Color online) The dependence of the intrinsic and side-jump (inset) conductivities on the strength of the Coulomb repulsion U of valence d-electrons in fcc Ni for dif- ferent directions of the magnetization in the crystal. In our work, we choose the GGA+U approach in order to study the effect of correlations on the AHE in fcc Ni, following the implementation of Ref. [25] and treating the double counting corrections within the atomic limit [26]. For this purpose, we scan the strength of intra-atomic re- pulsion parameter U , keeping at the same time the intra- atomic exchange parameter J such that the value of the spin moment of Ni stays roughly constant [23]. As can be seen from our calculations, presented in Fig. 1, the values of the σint upon including U change drastically and come closer to experiment, approaching a value of −800 S/cm when U is changed in the range of 0−4 eV, commonly used for calculations of other properties of Ni [24, 27]. This suggests that the main reason for the discrepancy between the intrinsic AHC values obtained from DFT and experiment might lie in the improper description of Ni's electronic structure from first principles. On the other hand, the values of the σsj are affected differently by the modifications in the electronic structure, almost not changing for M along the [001] axis, and displaying a non-monotonous behavior within the range of 100–300 S/cm in the absolute value as a function of U for two other magnetization directions. Such different sensitivity to the band structure can be understood by analyzing the structure of the SJC and the IC on the Fermi surface (FS). To simplify things, when taking into account the very complicated FSs of the ferromagnets considered in this study, we sum up the contributions to σint and σsj over all bands and all sheets of the FS, respectively, following Eqs. (1) and (2), when going along a certain direction in the BZ until the inner "Fermi" sphere in the BZ with the center at its 4 ence in the absolute change in the SJC and IC in FePd and FePt upon rotating the magnetization direction can be probably related to different SOI strength of the two materials [9], in FePt the corresponding trend of the SJC and the IC is opposite owing to the different Fermi sur- face topology of the two contributions. Surprisingly, the strong anisotropy of the SJC can be also observed in bcc Fe and fcc Ni. In Fe this anisotropy reaches as much as 70%, while in Ni the SJC anisotropy is striking as compared to the IC anisotropy, with changes in sign and order of magnitude of σsj as a function of the magneti- zation direction. This can be perhaps intuitively under- stood considering that σsj is given almost entirely by sin- gular "hot spots" at the FS, which change their position and the magnitude of their contribution depending on the matrix elements of the SOI, controlled in turn by the magnetization direction [12, 28], compare e.g. Fig. 2(b) and (c) for Ni. Despite the unprecedented improvement of the values of the AHC in several ferromagnets when compared to the experimentally measured numbers, the scattering- independent SJC considered here cannot describe the en- tire physics of the complex side-jump scattering and will likely fail to describe it in certain systems where long- range scattering and spin-dependent scattering domi- nate. This can be particularly important for the case of low-doped, i.e., having long screening length, magnetic systems. Also, within our approach, we consider only the leading-order in impurity strength correction to the self- energy, which is justified within the weak scattering limit. This approximation might fail, however, when the per- turbation in the crystal potential due to the presence of disorder or impurities is very strong. It would be highly desirable to extend the current model for the scattering- independent side-jump conductivity beyond the short- range disorder and weak scattering limit. In summary, we have implemented a method to cal- culate the scattering-independent SJC within DFT. We found that for fundamental ferromagnets, such as Fe, Co, FePd and FePt the agreement between theory and exper- iment can be essentially improved upon considering the scattering-independent SJC. This SJC can be calculated from the electronic structure of the pristine crystal only, which encourages the application of the considered model for the side-jump scattering to wider classes of materials with the goal of extending the applicability of the DFT in treating complex transverse scattering phenomena and comparison to experiments that are commonly performed on samples with unknown disorder and impurity content. We gratefully acknowledge Julich Supercomputing Centre for computing time and funding under the HGF- YIG Programme VH-NG-513. JS was supported under Grant Nos. onr-n000141110780 and NSF-DMR-1105512 and by the Research Corporation Cottrell Scholar Award. FIG. 2: (Color online) Angle-resolved conductivity dσ/dΩ in units of S/cm as a function of direction in the BZ. dσ/dΩ corresponds to all contributions to σ from inside the inner sphere in the BZ within the solid angle element dΩ. (a) σint for Ni [001], (b) σsj for Ni [001], (c) σsj for Ni [110], (d) σsj for Fe [001]. origin (Γ-point) is reached. In the case of such angle- resolved IC in Ni (U = 0), shown in Fig. 2(a), large contributions can be seen along the "hot loops" in the BZ, which are situated in the vicinity of the intersections between different sheets of bands [13], while the IC in the region away from such band crossings is also signifi- cant [6]. This is rather different from the topology of the SJC on the Fermi sphere. The SJC in Ni (U = 0) and Fe, presented in Fig. 2(b)-(d) manifests that the main con- tribution to σsj comes from certain isolated "hot spots", distributed rather sparsely over the Fermi sphere, while the SJC decays very quickly with the distance from such points. Such a strong difference in the distribution of the σsj and σint on the FS arises from the effective mag- netic monopole nature of the IC Berry's phase contri- bution near the band crossings, resulting in a more pro- nounced sensitivity of the σint to the parameters of the electronic structure, such as Coulomb repulsion U , Fermi energy etc., whereas the SJC does not contain such sin- gularities near those crossings. From our calculations presented above in Tables I, II, Figs. 1 and 2, it is evident that σsj exhibits large changes when the direction of the magnetization in the crystal is varied. In uniaxial crystals, such as FePt and hcp Co such anisotropy is not surprising, given that in these ma- terials also the anisotropy of the intrinsic AHC appears already in the first order with respect to the directional cosines of the magnetization [12]. And while the differ- ∗ corresp. author: y.mokrousov@fz-juelich.de [1] E. H. Hall, Phylos. Mag. 12, 157 (1881) [2] N. Nagaosa, J. Phys. Soc. Jpn. 75, 042001 (2006) [3] N. Nagaosa et al., Rev. Mod. Phys. 82, 1539 (2010) [4] Y. Yao et al., Phys. Rev. Lett. 92, 037204 (2004) [5] X. Wang et al., Phys. Rev. B 74, 195118 (2006) [6] X. Wang et al., Phys. Rev. B 76, 195109 (2007) [7] J. Smit, Physica 24, 39 (1958) [8] L. Berger, Phys. Rev. B 2, 4559 (1970) [9] K. M. Seemann et al., Phys. Rev. Lett. 104, 076402 (2010) [10] S. Lowitzer et al., Phys. Rev. Lett. 105, 266604 (2010) [11] A. A. Kovalev et al., Phys. Rev. Lett. 105, 036601 (2010) [12] E. Roman et al., Phys. Rev. Lett. 103, 097203 (2009) [13] H. Zhang et al., Phys. Rev. Lett. 106, 117202 (2011) [14] J. R. Yates et al., Phys. Rev. B 75, 195121 (2007) [15] For description of the code see http://www.flapw.de [16] A. A. Mostofi et al., Comp. Phys. Comm. 178, 685 (2008) 5 [17] F. Freimuth et al., Phys. Rev. B 78, 035120 (2008) [18] M. Chen et al., Appl. Phys. Lett. 98, 082503 (2011) [19] T. Miyasato et al., Phys. Rev. Lett. 99, 086602 (2007) [20] P. N. Dheer, Phys. Rev. 156, 637 (1967) [21] N. A. Sinitsyn et al., Phys. Rev. B 75, 045315 (2007) [22] J. M. Lavine, Phys. Rev. 123, 1273 (1961) [23] I. Yang et al., Phys. Rev. Lett. 87, 216405 (2001) [24] E. S¸a¸sıoglu et al., Phys. Rev. B 81, 054434 (2010) [25] A. Shick et al., Phys. Rev. B 60, 10763 (1999) [26] For several values of the parameter U we compared our results to the values of the AHC obtained in the "around mean field" limit, and did not find significant changes in the AHC. [27] For bcc Fe applying a GGA+U approach with parameter U of up to 2 eV results in a smooth positive shift of the values of the IC and SJC by up to approximately 60 and 40 S/cm, respectively, for all magnetization directions, bringing thus the total value of the calculated AHC even closer to experiment. [28] K. Hals et al., Phys. Rev. Lett. 105, 207204 (2010) 

1902.04933

2

1902

Phonon-induced enhancement of photon entanglement in quantum dot-cavity systems

We report on simulations of the degree of polarization entanglement of photon pairs simultaneously emitted from a quantum dot-cavity system that demand revisiting the role of phonons. Since coherence is a fundamental precondition for entanglement and phonons are known to be a major source of decoherence, it seems unavoidable that phonons can only degrade entanglement. In contrast, we demonstrate that phonons can cause a degree of entanglement that even surpasses the corresponding value for the phonon-free case. In particular, we consider the situation of comparatively small biexciton binding energies and either finite exciton or cavity mode splitting. In both cases, combinations of the splitting and the dot-cavity coupling strength are found where the entanglement exhibits a nonmonotonic temperature dependence which enables entanglement above the phonon-free level in a finite parameter range. This unusual behavior can be explained by phonon-induced renormalizations of the dot-cavity coupling $g$ in combination with a nonmonotonic dependence of the entanglement on $g$ that is present already without phonons.

cond-mat.mes-hall

cond-mat

a Phonon-induced enhancement of photon entanglement in quantum dot-cavity systems T. Seidelmann,1 F. Ungar,1 A. M. Barth,1 A. Vagov,1, 2 V. M. Axt,1 M. Cygorek,3 and T. Kuhn4 1Universitat Bayreuth, Lehrstuhl fur Theoretische Physik III, Universitatsstrasse 30, 95447 Bayreuth, Germany 2ITMO University, St. Petersburg, 197101, Russia 3Department of Physics, University of Ottawa, Ottawa, Ontario, Canada K1N 6N5 4Institut fur Festkorpertheorie, Universitat Munster, 48149 Munster, Germany We report on simulations of the degree of polarization entanglement of photon pairs simulta- neously emitted from a quantum dot-cavity system that demand revisiting the role of phonons. Since coherence is a fundamental precondition for entanglement and phonons are known to be a major source of decoherence, it seems unavoidable that phonons can only degrade entanglement. In contrast, we demonstrate that phonons can cause a degree of entanglement that even surpasses the corresponding value for the phonon-free case. In particular, we consider the situation of com- paratively small biexciton binding energies and either finite exciton or cavity mode splitting. In both cases, combinations of the splitting and the dot-cavity coupling strength are found where the entanglement exhibits a nonmonotonic temperature dependence which enables entanglement above the phonon-free level in a finite parameter range. This unusual behavior can be explained by phonon-induced renormalizations of the dot-cavity coupling g in combination with a nonmonotonic dependence of the entanglement on g that is present already without phonons. The appearance of entangled states is one of the show- case effects that highlights most impressively the dra- matic conceptual changes brought forth by going over from classical to quantum physics [1, 2]. Moreover, real- izations of entangled states, mostly with photons, have paved the way toward many innovative applications [3], e.g., in quantum cryptography [4, 5], quantum teleporta- tion [6], quantum information processing [7 -- 10], and pho- tonics [11]. In particular, quantum dot (QD) cavity sys- tems have attracted a lot of attention as sources for trig- gered entangled photon pairs [12 -- 19], not only because these systems hold the promise of a natural integration in solid-state devices. Embedding a QD in a microcavity enables the manipulation of few-electron and few-photon states in a system with high optical nonlinearities, which can be used for realizing a few-photon logic in quantum optical networks [20]. Furthermore, the cavity boosts the quantum yield due to the Purcell effect [14, 21] and, for high cavity quality factors Q, it reduces the detrimental effects of phonons on the photon indistinguishability [22]. The essence of entanglement in a bipartite system is the creation of a state that cannot be factorized into parts referring to the constituent subsystems, which re- quires the buildup of a superposition state. Polariza- tion entanglement between horizontally (H) or vertically (V ) polarized photon pairs is established, e.g., by cre- ating superpositions of the states HH(cid:105) and V V (cid:105) with two photons with either H or V polarizations exploiting the biexciton cascade [12 -- 18]. Starting from the biexci- ton, the system can decay first into one of the two ex- citons and a photon with the corresponding polarization (H or V ). The excitons then decay further to the QD ground state emitting a second photon with the same po- larization as in the biexciton decay. Ideally, the resulting quantum state is a coherent superposition and maximally entangled. Which-path information introduced, e.g., by the fine-structure splitting of the excitons, leads to an asymmetric superposition and decreased entanglement. The system can also decay from the biexciton directly to the ground state by simultaneous two-photon emission, a process which is much less affected by which-path infor- mation than the sequential single-photon decay [23 -- 25]. Obviously, maintaining a coherent superposition re- quires stable relative phases between the involved states. However, in a solid-state system, the interaction with the environment unavoidably leads to a loss of phase co- herence. In particular, phonons are known to provide a major source of decoherence [26 -- 35], which led to the expectation that phonons should always degrade the en- tanglement. Indeed, recent simulations [24, 36, 37] are in line with this expectation. In this Letter, we demonstrate that the phonon in- fluence is not necessarily destructive. On the contrary, phonons can increase the degree of photon entangle- ment when the destructive effect resulting from phonon- induced decoherence is overcompensated by phonon- related renormalizations of the QD-cavity coupling that shift the system into a regime of higher photon entan- glement. A precondition of this mechanism is a decrease of the degree of entanglement with rising QD-cavity cou- pling g in the phonon-free case in a finite g range. This is realized, e.g., in the limit of weak biexciton binding and finite exciton or cavity mode splitting. In both cases, the phonon-induced enhancement is found in a finite range of binding energies and couplings g. Our studies are based on the Hamiltonian [24, 37]: H = ωHXH(cid:105)(cid:104)XH + ωV XV (cid:105)(cid:104)XV (cid:88) + (ωH + ωV − ωB)B(cid:105)(cid:104)B + (cid:88) (cid:16) + ωq b† q bq + nχ γq q q,χ (cid:88) b† q + γ∗ (cid:96)=H,V † ωc (cid:96) a (cid:96)a(cid:96) (cid:17)χ(cid:105)(cid:104)χ + X , bq q (1) where B(cid:105) is the biexciton state with energy (ωH + ωV − ωB) and a biexciton binding energy EB = ωB, while XH/V (cid:105) denote the two exciton states with energies ωH/V that couple to H or V polarized cavity modes with † H/V ) and mode destruction (creation) operators aH/V (a energies ωc q) are operators that destroy (cre- ate) longitudinal acoustic phonons with wave vector q and energy ωq. We consider bulk phonons with a lin- ear dispersion and account for the deformation potential coupling γq. nχ is the number of electron-hole pairs con- tained in the states χ(cid:105) ∈ {B(cid:105),XH/V (cid:105)}. Finally, the Jaynes-Cummings type coupling of the cavity modes to the QD with coupling constant g is given by: H/V . bq (b† (cid:16)G(cid:105)(cid:104)XHa X = −g † † H + XH(cid:105)(cid:104)Ba H † V − XV (cid:105)(cid:104)Ba +G(cid:105)(cid:104)XV a † V (cid:17) (2) where H.c. stands for the Hermitian conjugate and G(cid:105) is the QD ground state, the energy of which is taken as the zero of energy. In addition, we account for cavity losses with a rate κ by the Lindblad operator: + H.c. , Lcav [ρ] = † † † (cid:96) − ρa (cid:96)a(cid:96) − a 2 a(cid:96) ρa (cid:96)a(cid:96) ρ . (3) (cid:17) (cid:88) (cid:96)=H,V (cid:16) κ 2 We assume that the system is initially prepared in the biexciton state, without photons and that the phonons are initially in equilibrium at a temperature T . This can be achieved, e.g., by using two-photon resonant or near- resonant excitation with short coherent pulses [16, 38 -- 41], which introduces much less decoherence and time jitter than, e.g., pumping the wetting layer and subse- quent relaxation to the biexciton. The dynamics of the reduced density matrix ρ is determined by the equation: (cid:104) H, ρ (cid:105) − + Lcav [ρ] , d dt ρ = − i  (4) where [, ]− denotes the commutator. As in Ref. [37], we evaluate ρ numerically in the subspace spanned by the five states B, 0, 0(cid:105), XH , 1, 0(cid:105), XV , 0, 1(cid:105), G, 2, 0(cid:105), and G, 0, 2(cid:105), where the numbers nH/V in χ, nH , nV (cid:105) denote the number of H/V photons. We use a path-integral approach that does not introduce approximations to the model. This is made possible by recent methodologi- cal advances that allow for a natural inclusion of non- Hamiltonian parts of the dynamics (e.g. represented by 2 Lindblad operators) in the path-integral formalism [42] as well as huge improvements of the performance by iterat- ing instead of the augmented density matrix, introduced in the pioneering work of Makri and Makarov [43, 44], a partially summed augmented density matrix [45]. We quantify the degree of entanglement by the concurrence, a quantity which has a one-to-one correspondence to the entanglement of formation [46]. To be precise, we use the concurrence of simultaneously emitted photon pairs C = 2 ¯ρHV ¯ρHH + ¯ρV V (5) (see the Supplemental Material [47] for further details) that can be calculated directly from the time-averaged occupations ¯ρHH , ¯ρV V and coherence ¯ρHV of the states HH(cid:105) and V V (cid:105) [25, 37, 57]. We focus on simultaneously emitted photon pairs since experiments [58, 59] agree with theory [15, 37] that this case is favorable for the entanglement. First, we present results for the situation sketched in Fig. 1(a) where the excitons have a finite fine-structure splitting δ = (ωH − ωV ), the biexciton binding energy is zero and both cavity modes are tuned to the two-photon V = ωH + ωV − ωB. In the situa- resonance 2ωc tion with phonons, these QD energies denote the polaron- shifted ones. To compare QD-cavity systems with identi- cal energy relations, the energy values are kept the same in the corresponding phonon-free calculations thus keep- ing the polaron shifts. H = 2ωc Figure 1(b) displays the temperature dependence of the concurrence for three values of the QD-cavity cou- pling. Only the result for g = 130 µeV agrees with the common expectation that the entanglement should monotonically decrease with temperature. In contrast, for g = 60 µeV and g = 35 µeV, unusual nonmono- tonic T dependences are found. Most interestingly, for g = 35 µeV, the concurrence is noticeably higher than the corresponding value obtained without phonons in the entire T range that we consider (T ∈ [1 K, 100 K]); i.e., for certain values of g we find indeed a phonon-induced enhancement of entanglement while in other cases the ex- pectation that phonons reduce the entanglement is con- firmed. The reason for this remarkable behavior becomes ap- parent when looking at the g dependence of the concur- rence in Fig. 1(c). Already without phonons, the concur- rence is a nonmonotonic function of g (purple curve) with a pronounced minimum reached roughly for g (cid:39) δ/2. Di- viding Eq. (4) by the coupling strength g and leaving out the coupling to phonons, the system dynamics is de- scribed by the rescaled quantities t(cid:48) = gt, g(cid:48) = g/g = 1, δ(cid:48) = δ/g, and κ(cid:48) = κ/g. Since the concurrence is the asymptotic value of the normalized coherence at long av- eraging times [37], the rescaling of the time is irrelevant. For large values of g, both parameters δ(cid:48) and κ(cid:48) tend to zero. This implies that the concurrence approaches unity 3 FIG. 1. (a) Sketch of the level scheme of a QD-cavity system with finite fine-structure splitting, zero biexciton binding energy and two-photon resonant cavity modes. (b) Concurrence as a function of the temperature for three selected values of the QD- cavity coupling. The corresponding values obtained without phonons are drawn as straight (faded) lines with the same linetype. Inset: difference ∆C between the maximum concurrence value at finite temperature and the corresponding phonon-free value normalized by the latter as a function of the biexciton binding energy EB for g1 = 35 µeV. (c) Concurrence as a function of the QD-cavity coupling for three temperatures together with the phonon-free result. In addition C(g(g)) is plotted using the phonon-renormalized coupling g(g) for T = 30 K (indicated on the upper axis), where C(g) is the phonon-free concurrence. The values of the QD-cavity coupling used in (b) are marked in (c) by vertical lines. Parameters: δ = 0.1 meV, κ = 0.025 ps−1, electron (hole) confinement length ae = 3 nm, ah = ae/1.15 where we assume a spherical GaAs-type QD with harmonic confinement. All other parameters, e.g., concerning the phonon coupling are taken from Ref. [60]. for large coupling strengths because the which-path in- formation disappears for a vanishing splitting and thus the concurrence is one [37, 57]. For very small QD-cavity couplings, κ(cid:48) and δ(cid:48) become arbitrarily large. Therefore, the sequential single photon decay via the intermediate exciton states becomes strongly off-resonant and is thus negligible compared with contributions from a direct two- photon transition which is always resonant in the present case [25]. Since the which-path information is contained only in the sequential decay the concurrence approaches unity again. But for finite splittings, the concurrence is smaller than one and thus a minimum must appear at a certain coupling strength g. When phonons are accounted for, the minimum is low- ered and shifted to a higher coupling strength depend- ing on the temperature. We attribute the shift to the well known effect of phonon-induced renormalization of the light-matter coupling [61]. To support this assign- ment we have estimated the renormalized coupling g(g) as in Ref. [62] by fitting equations with phenomenological renormalizations of a resonantly driven two-level system to path-integral calculations. The results are shown in the Supplemental Material [47]. If the only effect intro- duced by phonons was the g renormalization, then the value of the concurrence found without phonons at a par- ticular value of g should be shifted by phonons to g(g). Indeed, in Fig. 1(c) we have plotted C(g(g)) using the phonon-renormalized coupling g(g) for T = 30 K, where C(g) is the concurrence in the phonon-free case (green curve with circles). We find that, despite the crudeness of the estimation for g(g), the minimum of the shifted curve agrees even quantitatively well with the minimum found in the full path-integral simulation for this temperature (red dotted curve). Since the shift is larger for higher temperatures, displacing the phonon-free curve necessar- ily leads to higher values of the shifted curves in regions where the phonon-free concurrence is monotonically de- creasing with g. Consequently, in this region, phonon- induced enhancement appears for a finite g range. The total effect of phonons is, however, not merely a shift but also a lowering of the curves with rising tem- perature, which is indeed due to the dephasing action of phonons. It is important for obtaining a phonon-induced entanglement that the gain in entanglement resulting from the shift of the phonon-free curve due the phonon- induced g renormalization is not destroyed by the over- 4 FIG. 2. (a) Sketch of the level scheme of a QD-cavity system with finite fine-structure splitting, biexciton binding energy EB = 1 meV and two-photon resonant cavity modes. (b) Concurrence as a function of the temperature for three selected values of the QD-cavity coupling. The corresponding values obtained without phonons are drawn as straight (faded) lines with the same linetype. (c) Concurrence as a function of the QD-cavity coupling for three temperatures together with the phonon-free result. The values of the QD-cavity coupling used in (b) are marked in (c) by vertical lines. Apart from EB, the same parameters are used as in Fig. 1. all lowering of the concurrence caused by the decoher- ence. Figure 1(c) demonstrates that it is indeed possible that the renormalization-induced shift overcompensates the dephasing action. Additionally, when accounting for pure dephasing by introducing a phenomenological rate [23], the phonon-induced enhancement disappears (see the Supplemental Material [47]). This result reaffirms the g renormalization as the main origin of the effect, since it is absent in the phenomenological model. It is instructive to contrast the above findings with simulations for the more commonly considered situation sketched in Fig. 2(a), where the biexciton binding energy has the finite value EB = 1 meV and the cavity modes are in resonance with the two-photon transition to the biex- citon. Again, the phonon-free curve exhibits a minimum which is, however, rather flat [purple line in Fig. 2(c)]. In the limit g → ∞ the concurrence approaches unity since the argument given for the case of vanishing biexciton binding energy applies here as well. For the case that both g/( 1 2 EB) and δ/EB are small parameters, it has been shown analytically in Ref. [37] that the phonon-free concurrence approaches (cid:2)(cid:0)E2 is smaller than one for a finite δ. Including phonons, the reduction of the concurrence for small g values is strongly magnified as seen in Fig. 2(c). Overall, the dephasing action induced by phonons is so strong that the line shape of the concurrence as a function of g B − δ2(cid:1) /(cid:0)E2 B + δ2(cid:1)(cid:3), which is significantly deformed, and the effects related to a renormalization of g cannot be identified. As a con- sequence, the concurrence monotonically decreases with rising temperature and always stays below the phonon- free calculation for all values of g as exemplarily shown in Fig. 2(b). This demonstrates that the phonon-induced enhancement of entanglement described above can only occur when the g-renormalization effects dominate over the phonon-induced dephasing. The stronger phonon- induced dephasing for EB on the order of a few meV com- pared with vanishing EB has been explained recently [25] by noting that the energies bridged by phonon-assisted processes are closer to the maximum of the phonon spec- tral density in the former case. It is worthwhile to note that phonon-induced enhance- ment of photon entanglement is not restricted to the sin- gular case of vanishing EB but rather appears for a finite range of binding energies as demonstrated in the inset of Fig. 1(b). The difference ∆C between the maximum con- currence value at finite temperatures and the correspond- ing phonon-free value is positive clearly for an extended range. Further analysis (shown in the Supplemental Ma- terial [47]) reveals that the effect can be observed as long as EB (cid:46) δ/2 holds for our realistic parameters. We note in passing that the situation considered in Fig. 1 is not the only one where the conditions for phonon-induced entanglement are realized. This phe- nomenon can also be observed in a system with weak biexciton binding and degenerate excitons where which- path information is introduced by a finite splitting of the cavity modes (see the Supplemental Material [47]). There the concurrence calculated without phonons is again a nonmonotonic function of g, which exhibits even more than one extremum. Also in this case, the phonon-induced renormalization is strong enough to evoke a phonon-induced entanglement for finite parame- ter ranges. In conclusion, we demonstrate that phonon-induced renormalizations of the dot-cavity coupling can overcom- pensate decoherence effects and shift the system to a re- gion of higher entanglement. In combination with a non- monotonic dependence of the phonon-free concurrence, this can result in a nonmonotonic temperature depen- dence of the concurrence. Most interestingly, the concur- rence can even reach values above the phonon-free level, thus causing phonon-induced photon entanglement. M. Cygorek thanks the Alexander-von-Humboldt foun- dation for support through a Feodor Lynen fellowship. A. Vagov acknowledges the support from the Russian Science Foundation under the Project No. 18-12-00429, which was used to study dynamical processes nonlocal in time by the path-integral approach. This work was also funded by the Deutsche Forschungsgemeinschaft (DFG, German Research Foundation) - project Nr. 419036043. 5 9, 315 (2007). [14] A. Dousse, J. Suffczy´nski, A. Beveratos, O. Krebs, A. Lemaıtre, I. Sagnes, J. Bloch, P. Voisin, and P. Senel- lart, Nature 466, 217 (2010). [15] E. del Valle, New J. Phys. 15, 025019 (2013). [16] M. Muller, S. Bounouar, K. D. Jons, M. Glassl, and P. Michler, Nat. Photon. 8, 224 (2014). [17] N. Akopian, N. H. Lindner, E. Poem, Y. Berlatzky, and P. M. J. Avron, D. Gershoni, B. D. Gerardot, Petroff, Phys. Rev. Lett. 96, 130501 (2006). [18] A. Orieux, M. A. M. Versteegh, K. D. Jons, and S. Ducci, Reports on Progress in Physics 80, 076001 (2017). [19] C. S´anchez Munoz, F. P. Laussy, C. Tejedor, and E. del Valle, New Journal of Physics 17, 123021 (2015). [20] A. Faraon, A. Majumdar, D. Englund, E. Kim, M. Ba- jcsy, and J. Vukovi, New Journal of Physics 13, 055025 (2011). [21] A. Badolato, K. Hennessy, M. Atature, J. Dreiser, E. Hu, and A. Imamoglu, Science 308, 1158 P. M. Petroff, (2005). [22] T. Grange, N. Somaschi, C. Ant´on, L. De Santis, G. Cop- pola, V. Giesz, A. Lemaıtre, I. Sagnes, A. Auff`eves, and P. Senellart, Phys. Rev. Lett. 118, 253602 (2017). [23] S. Schumacher, J. Forstner, A. Zrenner, M. Florian, C. Gies, P. Gartner, and F. Jahnke, Opt. Express 20, 5335 (2012). [24] D. Heinze, A. Zrenner, and S. Schumacher, Phys. Rev. B 95, 245306 (2017). [25] T. Seidelmann, F. Ungar, M. Cygorek, A. Vagov, A. M. Barth, T. Kuhn, and V. M. Axt, Phys. Rev. B 99, 245301 (2019). [26] J. Forstner, C. Weber, J. Danckwerts, and A. Knorr, Phys. Rev. Lett. 91, 127401 (2003). [27] A. Vagov, V. M. Axt, T. Kuhn, W. Langbein, P. Borri, and U. Woggon, Phys. Rev. B 70, 201305(R) (2004). [28] V. M. Axt, T. Kuhn, A. Vagov, and F. M. Peeters, Phys. [1] R. Horodecki, P. Horodecki, M. Horodecki, and Rev. B 72, 125309 (2005). K. Horodecki, Rev. Mod. Phys. 81, 865 (2009). [2] J. Audretsch, Entangled Systems: New Directions in Quantum Physics (Whiley-VCH, Weinheim, 2007). [3] A. Zeilinger, Physica Scripta 92, 072501 (2017). [4] R. M. Stevenson, R. M. Thompson, A. J. Shields, I. Far- rer, B. E. Kardynal, D. A. Ritchie, and M. Pepper, Phys. Rev. B 66, 081302(R) (2002). [5] N. Gisin, G. Ribordy, W. Tittel, and H. Zbinden, Rev. Mod. Phys. 74, 145 (2002). [6] D. Bouwmeester, J.-W. Pan, K. Mattle, M. Eibl, H. We- infurter, and A. Zeilinger, Nature 390, 575 (1997). [7] J.-W. Pan, Z.-B. Chen, C.-Y. Lu, H. Weinfurter, Zukowski, Rev. Mod. Phys. 84, A. Zeilinger, and M. 777 (2012). [8] C. H. Bennett and D. P. DiVincenzo, Nature 404, 247 (2000). [9] J. Audretsch, ed., Entangled World: The Fascination of Quantum Information and Computation (Whiley-VCH, Weinheim, 2006). [10] S. C. Kuhn, A. Knorr, S. Reitzenstein, and M. Richter, Opt. Express 24, 25446 (2016). [11] J. L. O'Brian, A. Furusawa, and J. Vuckovi´c, Nature Photonics 3, 687 (2009). [12] R. M. Stevenson, R. J. Young, P. Atkinson, K. Cooper, D. A. Ritchie, and A. J. Shields, Nature 439, 179 (2006). [13] R. Hafenbrak, S. M. Ulrich, P. Michler, L. Wang, A. Rastelli, and O. G. Schmidt, New Journal of Physics [29] P. Machnikowski, Phys. Rev. B 78, 195320 (2008). [30] A. J. Ramsay, A. V. Gopal, E. M. Gauger, A. Nazir, B. W. Lovett, A. M. Fox, and M. S. Skolnick, Phys. Rev. Lett. 104, 017402 (2010). [31] C. Roy and S. Hughes, Phys. Rev. Lett. 106, 247403 (2011). [32] T. Close, E. M. Gauger, and B. W. Lovett, New Journal of Physics 14, 113004 (2012). [33] P. Kaer and J. Mørk, Phys. Rev. B 90, 035312 (2014). [34] D. E. Reiter, Phys. Rev. B 95, 125308 (2017). [35] A. Reigue, J. Iles-Smith, F. Lux, L. Monniello, M. Bernard, F. Margaillan, A. Lemaitre, A. Martinez, D. P. S. McCutcheon, J. Mørk, R. Hostein, and V. Vo- liotis, Phys. Rev. Lett. 118, 233602 (2017). [36] M. B. Harouni, Laser Physics 24, 115202 (2014). [37] M. Cygorek, F. Ungar, T. Seidelmann, A. M. Barth, A. Vagov, V. M. Axt, and T. Kuhn, Phys. Rev. B 98, 045303 (2018). [38] L. Hanschke, K. A. Fischer, S. Appel, D. Lukin, J. Wierzbowski, S. Sun, R. Trivedi, J. Vuckovi´c, J. J. Finley, and K. Muller, npj Quantum Inf. 4, 43 (2018). [39] S. Bounouar, M. Muller, A. M. Barth, M. Glassl, V. M. Axt, and P. Michler, Phys. Rev. B 91, 161302(R) (2015). [40] D. Huber, M. Reindl, Y. Huo, H. Huang, J. S. Wildmann, O. G. Schmidt, A. Rastelli, and R. Trotta, Nature Com- munications 8, 15506 (2017). [41] M. Reindl, K. D. Jons, D. Huber, C. Schimpf, Y. Huo, 6 V. Zwiller, A. Rastelli, and R. Trotta, Nano Lett. 17, 4090 (2017). M. H. M. van Weert, E. P. A. M. Bakkers, L. P. Kouwen- hoven, and V. Zwiller, Nano Letters 11, 645 (2011). [42] A. M. Barth, A. Vagov, and V. M. Axt, Phys. Rev. B 94, 125439 (2016). [43] N. Makri and D. E. Makarov, J. Chem. Phys. 102, 4600 (1995). [44] N. Makri and D. E. Makarov, J. Chem. Phys. 102, 4611 (1995). [45] M. Cygorek, A. M. Barth, F. Ungar, A. Vagov, and V. M. Axt, Phys. Rev. B 96, 201201(R) (2017). [46] W. K. Wootters, Quantum Inf. Comput. 1, 27 (2001). [47] See Supplemental Material at [URL will be inserted by publisher] for a precise definition of the concurrence of simultaneously emitted photon pairs, details how the renormalization of the coupling g is estimated, some nu- merical remarks, and for an additional configuration ex- hibiting phonon-induced entanglement. It also includes Refs. [48-56] in addition to references already cited in the main text. Results obtained using an approximate phe- nomenological rate model are shown as well as results regarding the dependence on the binding energy. [48] W. K. Wootters, Phys. Rev. Lett. 80, 2245 (1998). [49] G. Pfanner, M. Seliger, and U. Hohenester, Phys. Rev. B 78, 195410 (2008). [53] F. Ding, R. Singh, J. D. Plumhof, T. Zander, V. Kr´apek, Y. H. Chen, M. Benyoucef, V. Zwiller, K. Dorr, G. Bester, A. Rastelli, and O. G. Schmidt, Phys. Rev. Lett. 104, 067405 (2010). [54] R. Trotta, P. Atkinson, J. D. Plumhof, E. Zallo, R. O. Rezaev, S. Kumar, S. Baunack, J. R. Schrter, A. Rastelli, and O. G. Schmidt, Advanced Materials 24, 2668 (2012). and [55] R. Trotta, E. Zallo, E. Magerl, O. G. Schmidt, A. Rastelli, Phys. Rev. B 88, 155312 (2013). [56] F. Troiani, J. I. Perea, and C. Tejedor, Phys. Rev. B 74, 235310 (2006). [57] A. Carmele and A. Knorr, Phys. Rev. B 84, 075328 (2011). [58] R. M. Stevenson, A. J. Hudson, A. J. Bennett, R. J. Young, C. A. Nicoll, D. A. Ritchie, and A. J. Shields, Phys. Rev. Lett. 101, 170501 (2008). [59] S. Bounouar, C. de la Haye, M. Strau, P. Schnauber, A. Thoma, M. Gschrey, J.-H. Schulze, A. Strittmatter, S. Rodt, and S. Reitzenstein, Appl. Phys. Lett. 112, 153107 (2018). [60] B. Krummheuer, V. M. Axt, T. Kuhn, I. D'Amico, and F. Rossi, Phys. Rev. B 71, 235329 (2005). [50] M. Cosacchi, M. Cygorek, F. Ungar, A. M. Barth, A. Vagov, and V. M. Axt, Phys. Rev. B 98, 125302 (2018). [61] A. J. Ramsay, T. M. Godden, S. J. Boyle, E. M. Gauger, A. Nazir, B. W. Lovett, A. M. Fox, and M. S. Skolnick, Phys. Rev. Lett. 105, 177402 (2010). [51] D. F. V. James, P. G. Kwiat, W. J. Munro, and A. G. [62] A. Vagov, M. Glassl, M. D. Croitoru, V. M. Axt, and White, Phys. Rev. A 64, 052312 (2001). T. Kuhn, Phys. Rev. B 90, 075309 (2014). [52] M. E. Reimer, M. P. van Kouwen, A. W. Hidma, 

1701.07726

1

1701

Spectroscopic evidence for bulk-band inversion and three-dimensional massive Dirac fermions in ZrTe5

Three-dimensional topological insulators (3D TIs) represent novel states of quantum matters in which surface states are protected by time-reversal symmetry and an inversion occurs between bulk conduction- and valence-bands. However, the bulk-band inversion which is intimately tied to the topologically nontrivial nature of 3D TIs has rarely been investigated by experiments. Besides, 3D massive Dirac fermions with nearly near band dispersions were seldom observed in TIs. Recently, a van der Waals crystal, ZrTe5, was theoretically predicted to be a TI. Here, we report an infrared transmission study of a high-mobility (~33,000 cm$^2$/(Vs)) multilayer ZrTe5 flake at magnetic fields ($B$) up to 35 T. Our observation of a linear relationship between the zero-magnetic-field optical absorption and the photon energy, a bandgap of ~10 meV and a $\sqrt{B}$ dependence of the Landau level (LL) transition energies at low magnetic fields demonstrates 3D massive Dirac fermions with nearly linear band dispersions in this system. More importantly, the reemergence of the intra-LL transitions at magnetic fields higher than 17 T reveals the energy cross between the two zeroth LLs, which reflects the inversion between the bulk conduction- and valence-bands. Our results not only provide spectroscopic evidence for the TI state in ZrTe5 but also open up a new avenue for fundamental studies of Dirac fermions in van der Waals materials.

cond-mat.mes-hall

cond-mat

Spectroscopic evidence for bulk-band inversion and three- dimensional massive Dirac fermions in ZrTe5 Zhi-Guo Chena,1, R. Y. Chenb, R. D. Zhongc, John Schneelochc, C. Zhangc, Y. Huangd, Fanming Qua,e, Rui Yuf, Q. Lic, G. D. Guc and N. L. Wangb,g,1 aBeijing National Laboratory for Condensed Matter Physics and Institute of Physics, Chinese Academy of Sciences, Beijing 100190, China bInternational Center for Quantum Materials, School of Physics, Peking University, Beijing 10087, China cCondensed Matter Physics and Materials Science Department, Brookhaven National Lab, Upton, New York 11973, USA dCenter for Functional Nanomaterials, Brookhaven National Lab, Upton, New York 11973, USA eQuTech, Delf University of Technology, Delf 2600 GA, The Netherlands fSchool of Physics and Technology, Wuhan University, Wuhan 430072, China gCollaborative Innovation Center of Quantum Matter, Beijing, China 1To whom correspondence may be addressed. E-mail: zgchen@iphy.ac.cn or nlwang@pku.edu.cn Keywords: band inversion  Dirac fermions  topological insulators  Landau levels  Zeeman splitting  magneto-infrared spectroscopy  van der Waals materials 1 Three-dimensional topological insulators (3D TIs) represent novel states of quantum matters in which surface states are protected by time-reversal symmetry and an inversion occurs between bulk conduction- and valence-bands. However, the bulk-band inversion which is intimately tied to the topologically nontrivial nature of 3D TIs has rarely been investigated by experiments. Besides, 3D massive Dirac fermions with nearly near band dispersions were seldom observed in TIs. Recently, a van der Waals crystal, ZrTe5, was theoretically predicted to be a TI. Here, we report an infrared transmission study of a high-mobility ( 33,000 cm2/(Vs)) multilayer ZrTe5 flake at magnetic fields (B) up to 35 T. Our observation of a linear relationship between the zero-magnetic- field optical absorption and the photon energy, a bandgap of  10 meV and a √𝑩-dependence of the Landau level (LL) transition energies at low magnetic fields demonstrates 3D massive Dirac fermions with nearly linear band dispersions in this system. More importantly, the reemergence of the intra-LL transitions at magnetic fields higher than 17 T reveals the energy cross between the two zeroth LLs, which reflects the inversion between the bulk conduction- and valence-bands. Our results not only provide spectroscopic evidence for the TI state in ZrTe5 but also open up a new avenue for fundamental studies of Dirac fermions in van der Waals materials. Significance Experimental verifications of the theoretically predicted topological insulators (TIs) are essential steps towards the applications of the topological quantum phenomena. In the past, theoretically predicted TIs were mostly verified by the measurements of the topological surface states. However, as another key feature of the nontrivial topology in TIs, an inversion between the bulk-bands has rarely been observed by experiments. Here, by studying the optical transitions between the bulk LLs of ZrTe5, we not only offer spectroscopic evidence for the bulk-band inversion-the crossing of the two zeroth LLs in a magnetic field but also quantitatively demonstrate three-dimensional massive Dirac fermions with nearly linear band dispersions in ZrTe5. Our investigation provides a new paradigm for identifying TI states in candidate materials. Introduction Topologically nontrivial quantum matters, such as topological insulators (1-8), Dirac semimetals (9-19) and Weyl semimetals (20-27), have sparked enormous interest owing both to their exotic electronic properties and potential applications in spintronic devices and quantum computing. Therein, intrinsic topological insulators have insulating bulk states with odd Z2 topological invariants and metallic surface or edge states protected by time-reversal symmetry (4-6, 28). Most of the experimental evidence to date 2 for TIs is provided by the measurements of the spin texture of the metallic surface states. As a hallmark of the nontrivial Z2 topology of TIs (4-6, 28), an inversion between the characteristics of the bulk conduction- and valence-bands occurring at an odd number of time-reversal invariant momenta has seldom been probed by experiments. An effective approach for identifying the bulk-band inversion in TIs is to follow the evolution of two zeroth Landau levels (LLs) which arise from the bulk conduction- and valence-bands, respectively. As shown in Fig. 1A, for TIs, due to the bulk-band inversion and Zeeman effects, the two zeroth bulk Landau levels are expected to intersect in a critical magnetic field and then separate (3, 29), while for trivial insulators, the energy difference between their two zeroth Landau levels would become larger with increasing magnetic field. Therefore, an intersection between the two zeroth bulk LLs is a significant signature of the bulk-band inversion in TIs. However, a spectroscopic study of the intersection between the two zeroth bulk LLs in 3D TIs is still missing. In addition, many typical 3D TIs, such as Bi2Se3, show massive bulk Dirac fermions with parabolic band dispersions, which are effectively described by massive Dirac models (6, 28, 29). By contrast, 3D massive Dirac fermions with nearly linear bulk band dispersions (7), which are interesting topics following 2D massive Dirac fermions in gapped graphene (30, 31), were rarely observed in 3D TIs. A transition-metal pentatelluride, ZrTe5, embodies both 1D chain and 2D layer features (32), shown in Fig. 1B. One Zr atom together with three Te (1) atoms forms a quasi-1D prismatic chain ZrTe3 along the a-axis (x-axis). These prismatic ZrTe3 chains are connected through zigzag chains of Te (2) atoms along the c-axis (y-axis) and then construct quasi-2D ZrTe5 layers. The bonding between ZrTe5 layers is van der Waals type (33, 34). Thus, as displayed in Fig. 1C, bulk ZrTe5 crystals can be easily cleaved down to a few layers. Recently, the ab initio calculations indicate that monolayer ZrTe5 sheets are great contenders for quantum spin Hall insulators-2D TI and that 3D ZrTe5 crystals are quite close to the phase boundary between strong and weak TIs (33). Scanning tunneling microscopy measurements have shown that edge states exist at the step edges of the ZrTe5 surfaces (35, 36). Nonetheless, further investigations are needed to check whether the observed edge states in ZrTe5 are topologically nontrivial or not. Studying the bulk- band inversion or the intersection between the two zeroth bulk LLs can provide a crucial clue to clarifying the nature of the edge states in ZrTe5. Except the edge states within the energy gap of the bulk bands around the Brillouin zone center (i.e.  point) of ZrTe5 (36, 37), 3D massless Dirac fermions with the linearly dispersing conduction- and valence-band degenerate at the  point were suggested to exist in this material by previous angle-resolved photon emission spectroscopy, transport and optical experiments (38- 41). Considering that (1) our ZrTe5 thick crystals were experimentally shown to be Dirac semimetals hosting 3D massless Dirac fermions, (2) ZrTe5 monolayers were theoretically predicted to be quantum spin Hall insulators and (3) the bulk state of ZrTe5 is very sensitive to its interlayer distance which might 3 be discrepant in different samples (33, 40), it is significant to quantitatively verify whether 3D massive Dirac fermions with a bandgap and nearly linear bulk-band dispersions can be realized in dramatically thinned flakes of our ZrTe5 crystals. Infrared spectroscopy is a bulk-sensitive experimental technique for studying low-energy excitations of a material. Here, in order to investigate the bulk-band inversion and the nature of the bulk fermions in ZrTe5, we measured the infrared transmission spectra T(𝜔, B) of its multilayer flake with thickness d ~ 180 nm at magnetic fields applied along the wave vector of the incident light (see Materials, Methods and Supplementary Information Section 1). A series of intra- and inter-LL transitions are present in the relative transmission spectra T(B)/T(B0 = 0 T) of the ZrTe5 flake. The linear √𝐵-dependence of the LL transition energies at B  4 T and the non-zero intercept of the LL transitions at B = 0 T, combined with the linear relationship between the zero-magnetic-field optical absorption and the photon energy, indicates 3D massive Dirac fermions with nearly linear band dispersions in the ZrTe5 flake. Moreover, a 3D massive Dirac model with a bandgap of  10 meV can quantitatively explain the magnetic-field dependence of the measured LL transition energies very well. At high magnetic fields, we observed fourfold splittings of the LL transitions. Additionally, our analysis of the split LL transitions shows that the intra-LL transitions, which are associated to the two zeroth LLs and disappear at B  2.5 T, reemerge at B  17 T. Considering that the zeroth LL crossing in a Zeeman field would make the two zeroth bulk LLs intersect with the chemical potential here and then alter the carrier occupation on the zeroth LLs, we attribute the reemergence of the intra-LL transitions in the ZrTe5 flake to the energy crossing of its two zeroth bulk LLs which originates from the bulk-band inversion. These results strongly support the theoretically predicted 3D TI states in 3D ZrTe5 crystals. Results 3D massive Dirac fermions. At zero magnetic field, the measured absolute transmission T( 𝜔 ) corresponds to the absorption coefficient: A(𝜔) = -[lnT(𝜔)]/d, where d is the thickness of the sample and 𝜔 is the photon energy (see the Methods of Ref. 42). In solids, the absorption coefficient is determined by the joint density of state D(𝜔): A(𝜔) ∝ D(𝜔)/𝜔. 3D electron systems with linear band dispersions along three momentum directions have the D(𝜔) proportional to 𝜔2, while for 2D linear dispersions, D(𝜔) ∝ 𝜔. Thus, in stark contrast to the 𝜔-independent absorption of 2D Dirac materials like graphene (43), the linear 𝜔-dependence of A(𝜔) in Fig. 1D indicates 3D linear band dispersions in ZrTe5. Moreover, at low energies, the absorption coefficient apparently deviates from the linear relationship with 𝜔, implying the 4 opening of a bandgap at the original Dirac point. The 3D linear band dispersions, together with the possible bandgap, suggest the presence of 3D massive Dirac fermions in the exfoliated ZrTe5 flake. In order to confirm the 3D massive Dirac fermions in ZrTe5, we further performed infrared transmission experiments at magnetic fields applied perpendicular to the ac-plane (xy-plane) of the crystal (Faraday geometry). The low-field relative transmission T(B)/T(B0 = 0 T) spectra in Fig. 2A show seven dip features Tn (1≤ n ≤ 7) directly corresponding to the absorption peaks of LL transitions. All of these dip features systematically shift to higher energies, as the magnetic field increases. Here, we define the energy positions of the transmission minima in T(B)/T(B0) as the absorption energies, which is a usual definition in thin film systems, such as Bi2Se3 films and graphene. Then, we plotted the square of the Tn 2 ) as a function of magnetic field in Fig. 2B. The linear B-dependence of 𝐸T𝑛 energies (𝐸T𝑛 dependence of 𝐸T𝑛) reveals the LL transitions of Dirac fermions (31, 44, 45). In Fig. 2C, the non-zero 2 at zero magnetic-field is an important signature of a non-zero Dirac mass 2 (i.e. linear √𝐵- intercept of the linear fit to 𝐸T1 or a bandgap (31). Therefore, the linear relationship between 𝐸T𝑛 2 and B, together with the non-zero intercept, provides further evidence for 3D massive Dirac fermions in the ZrTe5 flake. In order to quantitatively check the 3D massive Dirac fermions in the ZrTe5 flake, we employ a 3D massive Dirac Hamiltonian which was derived from the low-energy effective k  p Hamiltonian based on the spin-orbital coupling, the point group and time-reversal symmetries in ZrTe5 and includes the spin degree of freedom (40). According to the 3D massive Dirac Hamiltonian expanded to the linear order of momenta, we can obtained the nearly linear band dispersions of ZrTe5 at zero magnetic field: 𝐸(𝑘𝑥,𝑦,𝑧) = ±√ℏ2(𝑘𝑥 2𝜐𝑧 2) + (Δ/2)2, where 𝑘𝑥,𝑦,𝑧 are the wave vectors in momentum space, 𝜐𝑥,𝑦,𝑧 are 2𝜐𝑥 2 + 𝑘𝑦 2𝜐𝑦 2 + 𝑘𝑧 the Fermi velocities along three momentum directions, Δ is the bandgap and describes the mass of Dirac fermions 𝑚𝑥,𝑦,𝑧 𝐷 = Δ/(2𝜐𝑥,𝑦,𝑧 2 ) and ℏ is Planck's constant divided by 2. In a magnetic field applied perpendicular to the ac-plane, the LL spectrum of ZrTe5 without considering Zeeman effects has the form: 𝐸𝑁(𝑘𝑧) = ±𝛿𝑁,0√(∆/2)2 + (𝑣𝑧ℏ𝑘𝑧)2 + sgn(𝑁)√2𝑒ℏ𝑣F 2𝐵𝑁 + (∆/2)2 + (𝑣𝑧ℏ𝑘𝑧)2, (1) where 𝑣F is the effective Fermi velocity of LLs, the integer N is Landau index, 𝛿𝑁,0 is the Kronecker delta function, sgn(N) is the sign function and e is the elementary charge. The √𝐵 dependence is a hallmark of Dirac fermions (31, 44, 45). According to Eq. (1), the magnetic field makes the 3D linear band dispersions evolve into a series of 1D non-equally-spaced Landau levels (or bands) which disperse with the momentum component along the field direction. Specifically, since the 3D massive Dirac Hamiltonian 5 of ZrTe5 involves the spin degree of freedom of this system, two zeroth LLs indexed by 𝑁 = +0 and 𝑁 = −0 locate at the hole and electron band extrema, respectively and have energy dispersions 𝐸±0(𝑘𝑧) = ±√(∆/2)2 + (𝑣𝑧ℏ𝑘𝑧)2, which is different from the case that when the spin degree of freedom was not considered in materials with the hexagonal lattice, only one zeroth non-degenerate Landau level is present in each valley (46). The optical selection rule for ZrTe5 only allows the LL transitions from LLN to LLN': ΔN = 𝑁 – 𝑁′ = ±1 and with the 𝑘𝑧-momentum difference Δ 𝑘𝑧 = 0 (40). Due to the singularities of the density-of-states (DOS) at 𝑘𝑧 = 0, magneto-optical response here, which is determined by the joint DOS, should be mainly contributed by the LL transitions at 𝑘𝑧 = 0 (29, 40). Thus, the energies of the inter-band LL (inter-LL) transitions LL−𝑁 → LL+𝑁−1 (or LL−𝑁−1 → LL+𝑁) and the intra-band LL (intra-LL) transitions LL+𝑁−1 → LL+𝑁 (or LL−𝑁 → LL−𝑁−1), 𝐸𝑁 𝐼𝑛𝑡𝑒𝑟 and 𝐸𝑁 𝐼𝑛𝑡𝑟𝑎 at 𝑘𝑧 = 0, are given by: 𝐸𝑁 𝐼𝑛𝑡𝑒𝑟 = √2𝑒ℏ𝑣F 2𝐵𝑁 + (∆/2)2 + √2𝑒ℏ𝑣F 2𝐵𝑁 − 1 + (∆/2)2 (2) 𝐸𝑁 𝐼𝑛𝑡𝑟𝑎 = √2𝑒ℏ𝑣F 2𝐵𝑁 + (∆/2)2 − √2𝑒ℏ𝑣F 2𝐵𝑁 − 1 + (∆/2)2 (3) From T1 to T7, the slopes of the linear fits to 𝐸T𝑛 2 in Fig. 2B scale as 1 : 5.7 : 9.3 : 13.1 : 16.7 : 20.5 : 24.1, respectively, which is close to the approximate ratio of the theoretical inter-LL transition energies based on Eq. (2), 1 : (√2 + √1)2 (≈ 5.8) : (√3 + √2)2 (≈ 9.9) : (√4 + √3)2 (≈ 13.9) : (√5 + √4)2 (≈ 17.9) : (√6 + √5)2 (≈ 21.9) : (√7 + √6)2 (≈ 25.9). Therefore, the absorption features Tn are assigned as the inter-LL transitions: LL−𝑁−1 → LL+𝑁 (or LL−𝑁 → LL+𝑁−1) (Fig. 2B) and we have n = 𝑁, where 1≤ n ≤ 7. 2 based on Eq. (2) from a least square fit yields a bandgap ∆ ≈ 10  2 meV, the effective Fermi Fitting 𝐸T𝑛 velocities 𝑣F T1 ≈ (4.76  0.04) × 105 m/s and 𝑣F T2≤ 𝑛 ≤ 7 ≈ (5.044.95  0.04) × 105 m/s (see Supplementary Section 2). As another signature of the bandgap or the non-zero Dirac mass, the absorption feature T1 * is present at energies lower than the lowest-energy inter-LL transition T1 in Fig. 2D. The feature T1 * is attributed to the intra-LL transition LL+0 → LL+1 (or LL−1 → LL−0 ), illustrated by the grey arrows in Fig. 2E (see Supplementary Section 3). According to Eq. (2) and (3), the energy difference (ET1−ET1*) between the transitions T1 and T1 * in the inset of Fig. 2C directly gives the bandgap value ∆ = 𝐸T1 − 𝐸T1∗ ≈ 10  2 meV, which is consistent with the value obtained by the above fitting. Furthermore, the field dependence of the T1 * energy in Fig. 2C can be well fitted by Eq. (3) with ∆ ≈ 10  2 meV and 𝑣F T1∗ ≈ (4.63  0.04) × 105 m/s. 6 The carrier-charge mobility 𝜇 in the ZrTe5 flake can be calculated using the general formula (47): 𝜇 = eℏ/(𝛤m*), where 𝛤 is the transport scattering rate and m* is the carrier effective mass on the anisotropic Fermi surface (48, 49). Here, the transport scattering rate 𝛤 within the ac-plane can be roughly estimated from the width of the T1 feature at low fields: 𝛤 ≈ 9 meV at B = 0.5 T. Moreover, considering the absence of Pauli blocking of the T1 transition at B = 0.5 T, we get the Fermi energy in ZrTe5, 𝐸F < 𝐸LL+1 (or −1) = 𝐸T1 ≈ 15 meV (see Supplementary Section 4), which means the Fermi level in our sample is quite close to the band extrema. In this case, the average effective mass m* of the carriers within 𝑎𝑐)2] ≈ 3.54 × 10-33 kg ≈ 0.00389 m0, where m0 is 𝑎𝑐 is approximately equal to the free electron mass and the average Fermi velocity within the ac-plane 𝑣F the ac-plane can be described by (30): mac * ≈ ∆/[2(𝑣F the effective Fermi velocity of the LLs, 𝑣F the carriers within the ac-plane of our ZrTe5 sample: 𝜇 ≈ 33,000 cm2/(Vs), which is comparable to those 𝑎𝑐 ≈ 4.76 × 105 m/s. Finally, we can estimate the mobility of in graphene/h-BN heterostructures (50, 51). Buk-band Inversion. As shown in Fig. 3A, applying a higher magnetic field enables us to observe the splitting of the T1 transition, which indicates a non-negligible Zeeman effect in ZrTe5 (40). For TIs, due to the Zeeman field, each LL except the two zeroth LLs splits into two sublevels with opposite spin states, while the LL-0 and LL+0 are spin-polarized and have spin-up and -down state, respectively (3, 29). The energy of the sublevel has the form (40): 𝐸𝑁,𝜉 = 𝐸𝑁(𝑘𝑧 = 0) + 1/2𝜉𝑔𝑁B (4) where 𝜉 is equal to +1 for spin-up and –1 for spin-down and 𝑔𝑁 is the effective Landé g-factor of LLN. The spin-orbit coupling (SOC) in ZrTe5 mixes the spin states of the two sublevels, so two extra optical transitions between the sublevels with different spin-indices become possible. The inter- and intra-LL transition energies including the Zeeman effect can be written as (40): 𝐸𝑁,𝜉,𝜉′ 𝐼𝑛𝑡𝑒𝑟 = 𝐸𝑁 𝐼𝑛𝑡𝑒𝑟 + 1/2(𝜉𝑔𝑁 − 𝜉′𝑔−(𝑁−1))𝐵 (5) 𝐸𝑁,𝜉,𝜉′ 𝐼𝑛𝑡𝑟𝑎 = 𝐸𝑁 𝐼𝑛𝑡𝑟𝑎 + 1/2(𝜉𝑔𝑁 − 𝜉′𝑔(𝑁−1))𝐵 (6) where 𝜉 and 𝜉′ correspond to the spin states of the two sublevels, respectively. Figure 3B displays the false-color map of the –ln[T(B)/T(B0)] spectra of the ZrTe5 flake. Interestingly, a cusp-like feature around 18 T, which is indicated by a white arrow, can be observed in Fig. 3C (i.e. the magnified image of a region in Fig. 3B). To quantitatively investigate the physical meaning of this cusp- like feature, we plot the energies of the four split T1 transitions (i.e. 1𝛼, 1𝛽, 1𝛾 and 1𝛿 (green dots)) around 16 T in Fig. 3B, which are defined by the onsets of the absorption features due to the Zeeman splitting (see Fig. 3 of the Supplemental Material of Ref. 40 and Supplementary Section 6). As displayed 7 by the green dashed lines in Fig. 3B, fitting the energy traces of the inter-LL transitions, 1𝛼, 1𝛽, 1𝛾 and T1 and the bandgap ∆ yields the g- 1𝛿, based on Eq. (5) with the obtained values of the Fermi velocity 𝑣F factors of the two zeroth LLs and LL±1: geff(LL+0) = geff(LL-0) ≈ 11.1, geff(LL-1) ≈ 31.1 and geff(LL+1) ≈ 9.7 (or geff(LL+1) ≈ 31.1 and geff(LL-1) ≈ 9.7) (see Supplementary Section 5). It is known that as a hallmark of TIs, the band inversion causes the exchange of the characteristics between the valence- and conduction-band extrema (2, 6, 28), so as shown in Fig. 1A and 3D, the LL-0 and LL+0, which come from the inverted band extrema, have reversed spin states and cross at a critical magnetic field (3, 29). According to Eq. (4) with the above values of geff(LL±0) and ∆, we estimated the critical magnetic field Bc ≈ 17 T. In Fig. 2A and D, the disappearance of the intra-LL transition T1 * around B ≈ 2.5 T indicates that LL+0 (or LL-0) becomes fully depleted (or occupied) with increasing magnetic field and that at B > 2.5 T, the chemical potential of ZrTe5 can be considered to be located at zero energy. In this case, the two zeroth LLs intersect with the chemical potential at the same magnetic field Bc. More importantly, this intersection means at B > Bc ≈ 17 T, LL-0 and LL+0 becomes empty and occupied, respectively, which leads to the gradual replacement of the inter-LL transitions T1, 1𝛼, 1𝛽, 1𝛾 and 1𝛿, by the intra-LL transitions T1 *, 1𝜒, 1𝜆, 1𝜃 and 1𝜀, explained in Fig. 3D. Furthermore, in Fig. 3B, the energy traces of the four split transitions (grey dots) observed at B > 17 T are shown to follow the white theoretical curves for the intra-LL transitions T1 *, which are based on Eq. (6). Therefore, the four split transitions observed at B > 17 T in Fig. 3B can be assigned as the intra-LL transitions T1 *, 1𝜒, 1𝜆, 1𝜃 and 1𝜀. Since the energy traces of the split intra-LL transitions T1 * deviate markedly from those of the inter- LL transitions T1, the reemergence of the T1 * transitions at B > 17 T causes the cusp-like feature, which provides experimental evidence for the bulk-band inversion in the ZrTe5 flake. In summary, using magneto-infrared spectroscopy, we have investigated the Landau level spectrum of the multilayer ZrTe5 flake. The magnetic-field-dependence of the LL transition energies here, together with the photon-energy-dependence of the absorption coefficient at zero-field, quantitatively demonstrates 3D massive Dirac fermions with nearly linear dispersions in the ZrTe5 flake. Due to the Zeeman splitting of the LLs, the energy splitting of the LL transitions was observed at B  6 T. Interestingly, the intra-LL transitions T1 * reemerge at B > 17 T. We propose that the reemergence of the T1 * transitions results from the band-inversion-induced crossing of the two zeroth LLs, LL+0 and LL-0. Our results make ZrTe5 flakes good contenders for 3D TIs. Moreover, due to the 3D massive Dirac-like dispersions and the high bulk- carrier mobility ( 33,000 cm2 / (Vs)), the ZrTe5 flake can also be viewed as a 3D analogue of gapped graphene, which enables us to deeply investigate exotic quantum phenomena. 8 Materials and Methods Sample preparation and characterizations. Bulk single crystals of ZrTe5 were grown by Te flux method. The elemental Zr and Te with high purity were sealed in an evacuated double-walled quartz ampule. The raw materials were heated at 900°C and kept for 72 hours. Then they were cooled slowly down to 445°C and heated rapidly up to 505°C. The thermal-cooling cycling between 445 and 505 °C lasts for 21 days to re-melt the small size crystals. The multilayer ZrTe5 flake (ac-plane) for magneto- transmission measurements were fabricated by mechanical exfoliation, and deposited onto double-side- polished SiO2/Si substrates with 300 nm SiO2. The flake thickness ~ 180 nm and the chemical composition were characterized by atomic force microscopy (AFM) and energy dispersion spectroscopy (EDS), respectively (Please see Fig. S1). Infrared transmission measurements. The transmission spectra were measured at about 4.5 K in a resistive magnet in the Faraday geometry with magnetic field applied in parallel to the wave vector of incident light and the crystal b-axis. Non-polarized IR light (provided and analyzed by a Fourier transform spectrometer) was delivered to the sample using a copper light pipe. A composite Si bolometer was placed directly below the sample to detect the transmitted light. The diameter of IR focus on the sample is ~ 0.5 – 1 mm. Owing to the mismatch between the size of the IR focus and the ZrTe5 flake, an aluminium aperture was placed around the sample. The transmission spectra are shown at energies above 10 meV, corresponding to wavelengths shorter than 124 µm. The wavelength of infrared light here is smaller than the size of the measured sample, and thus the optical constants can be used for a macroscopic description of the data. References 1. Kane CL, Mele EJ. (2005) Z2 topological order and the quantum spin Hall effect. Phys Rev Lett 95: 146802. 2. Bernevig BA, Hughes TL, Zhang SC. (2006) Quantum spin Hall effect and topological phase transition in HgTe quantum wells. Science 314:1757–1761. 3. König M, et al. (2007) Quantum spin Hall insulator state in HgTe quantum wells. Science 318:766– 770. 4. Fu L, Kane CL, Mele EJ, (2007) Topological insulators in three dimensions. Phys Rev Lett 98:106803. 5. Moore JE, Balents L, (2007) Topological invariants of time-reversal-invariant band structures. Phys Rev B 75:121306. 6. Zhang H, et al. (2009) Topological insulators in Bi2Se3, Bi2Te3 and Sb2Te3 with a single Dirac cone on 9  the surface. Nat Phys 5:438–442. 7. Hsieh D, et al. (2008) A topological Dirac insulator in a quantum spin Hall phase. Nature 452:970– 974. 8. Chen YL, et al. (2009) Experimental realization of a three-dimensional topological insulator, Bi2Te3. Science 325:178–181. 9. Wan XG, Turner AM, Vishwanath A, Savrasov SY, (2011) Topological semimetal and Fermi-arc surface states in the electronic structure of pyrochlore iridates. Phys Rev B 83:205101. 10. Young SM, et al. (2012) Dirac semimetal in three dimensions. Phys Rev Lett 108:140405. 11. Wang ZJ, et al. (2012) Dirac semimetal and topological phase transitions in A3Bi (A = Na, K, Rb). Phys Rev B 85:195320. 12. Wang ZJ, et al. (2013) Three-dimensional Dirac semimetal and quantum transport in Cd3As2. Phys Rev B 88:125427. 13. Liu ZK, et al. (2014) Discovery of a three-dimensional topological Dirac semimetal, Na3Bi. Science 343:864–867. 14. Neupane M, et al. (2014) Observation of a three-dimensional topological Dirac semimetal phase in high-mobility Cd3As2. Nat Commun 5:3786. 15. Jeon S, et al. (2014) Landau quantization and quasiparticle interference in the three-dimensional Dirac semimetal Cd3As2. Nat Mater 13:851–856. 16. Liang T, et al. (2014) Ultrahigh mobility and giant magnetoresistance in the Dirac semimetal Cd3As2. Nat Mater 14:280–284. 17. He LP, et al. (2014) Quantum transport evidence for a three-dimensional Dirac semimetal phase in Cd3As2. Phys Rev Lett 113:246402. 18. Borisenko S, et al. (2014) Experimental realization of a three-dimensional Dirac semimetal. Phys Rev Lett 113:027603. 19. Liu ZK, et al. (2014) A stable three-dimensional topological Dirac semimetal Cd3As2. Nat Mater 13: 677–681. 20. Burkov AA, Balents L, (2011) Weyl semimetal in a topological insulator multilayer. Phys Rev Lett 107:127205. 21. Weng H, et al. (2015) Weyl semimetal phase in non-centrosymmetric transition metal monophosphides. Phys Rev X 5:011029. 22. Huang SM, et al. (2015) An inversion breaking Weyl semimetal state in the TaAs material class. Nat Commun 6:7373. 23. Xu SY, et al. (2015) Discovery of a Weyl Fermion semimetal and topological Fermi arcs. Science 347: 294. 10 24. Lv BQ, et al. (2015) Observation of Weyl nodes in TaAs. Nat Phys 11:724–727. 25. Yang LX, et al. (2015) Weyl semimetal phase in the non-centrosymmetric compound TaAs. Nat Phys 11:728–732. 26. Xu SY, et al. (2015) Discovery of a Weyl fermion state with Fermi arcs in niobium arsenide. Nat Phys. 11:748–754. 27. Shekhar C, et al. (2015) Extremely large magnetoresistance and ultrahigh mobility in the topological Weyl semimetal candidate NbP. Nat Phys 11:645–649. 28. Liu CX, et al. (2010) Model Hamiltonian for topological insulators. Phys Rev B 82:045122. 29. Orlita M, et al. (2015) Magneto-optics of massive Dirac fermions in bulk Bi2Se3. Phys Rev Lett 114:186401. 30. Hunt B, et al. (2013) Massive Dirac fermions and Hofstadter butterfly in a van der Waals heterostructures. Science 340:1427. 31. Chen Z-G, et al. (2014) Observation of an intrinsic bandgap and Landau level renormalization in graphene/boron-nitride heterostructures. Nat Commun 5:4461. 32. Fjellvåg H, Kjekshus A, (1986) Structural properties of ZrTe5 and HfTe5 as seen by powder Diffraction. Solid State Commun 60:91. 33. Weng H, et al. (2014) Transition-Metal Pentatelluride ZrTe5 and HfTe5: A paradigm for large-gap quantum spin Hall insulators. Phys Rev X 4:011002. 34. Niu J, et al. (2015) Electrical transport in nano-thick ZrTe5 sheets: from three to two dimensions. Preprint at <http://arxiv.org/abs/1511.09315>. 35. Li X-B, et al. (2016) Experimental Observation of Topological Edge States at the Surface Step Edge of the Topological Insulator ZrTe5. Phys Rev Lett 116:176803. 36. Wu R, et al. (2016) Experimental evidence of large-gap two-dimensional topological insulator on the surface of ZrTe5. Phys Rev X 6:021017. 37. Zhang Y, et al. (2016) Electronic evidence of temperature-induced Lifshitz transition and topological nature in ZrTe5. Preprint at <http://arxiv.org/abs/1602.03576>. 38. Li Q, et al. (2016) Chiral magnetic effect in ZrTe5. Nat Phys 12:550. 39. Chen RY, et al. (2015) Optical spectroscopy study of the three-dimensional Dirac semimetal ZrTe5. Phys Rev B 92:075107. 40. Chen RY, et al. (2015) Magnetoinfrared spectroscopy of Landau levels and Zeeman splitting of three- dimensional massless Dirac fermions in ZrTe5. Phys Rev Lett 115:176404. 41. Zheng G, et al. (2016) Transport evidence for the three-dimensional Dirac semimetal phase in ZrTe5. Phys Rev B 93:115414. 42. Orlita M, et al. (2014) Observation of three-dimensional massless Kane fermions in a zinc-blende 11  crystal. Nat Phy 10:233. 43. Li ZQ, et al. (2008) Dirac charge dynamics in graphene by infrared spectroscopy. Nat Phys 4:532-535. 44. Sadowski ML, et al. (2006) Landau level spectroscopy of ultrathin graphite layers. Phys Rev Lett 97:266405. 45. Jiang Z, et al. (2007) Infrared spectroscopy of Landau levels of graphene. Phys Rev Lett 98:197403. 46. Liang T, et al. (2013) Evidence for massive bulk Dirac fermions in Pb1-xSnxSe from Nernst and thermopower experiments. Nature Commun 4:2696. 47. Issi J-P, et al. (2014) Electron and phonon transport in graphene in and out of the bulk. Physics of Graphene, eds Aoki H, Dresselhaus MS (Springer), pp 65-112. 48. Kamm GN, et al. (2014) Fermi surface, effective masses, and Dingle temperatures of ZrTe5 as derived from the Shubnikov–de Haas effect. Phys Rev B 31:7617. 49. Yuan X, et al. (2015). Observation of quasi-two-dimensional Dirac fermions in ZrTe5. Preprint at <http://arxiv.org/abs/1510.00907>. 50. Ponomarenko LA, et al. (2013) Cloning of Dirac fermions in graphene superlattices. Nature 497:594– 597. 51. Dean CR, et al. (2013) Hofstadter's butterfly and the fractal quantum Hall effect in moiré superlattices. Nature 497:598-602. Acknowledgments The authors thank X. C. Xie, F. Wang, Z. Fang, M. Orlita, M. Potemski, H. M. Weng, L. Wang, C. Fang and X. Dai for very helpful discussions. The authors acknowledge support from CAS Pioneer Hundred Talents Program, the National Key Research and Development Program of China (Project No. 2016YFA0300600), the European Research Council (ERC ARG MOMB Grant No. 320590), the National Science Foundation of China (Grant No. 11120101003 and No. 11327806) and the 973 project of the Ministry of Science and Technology of China (Grant No. 2012CB821403). A portion of this work was performed in National High Magnetic Field Laboratory which is supported by National Science Foundation Cooperative Agreement No. DMR-1157490 and the State of Florida. Work at Brookhaven was supported by the Office of Basic Energy Sciences (BES), Division of Materials Sciences and Engineering, U.S. Department of Energy (DOE), through Contract No. DE-SC00112704. Author Contributions Z.G.C. conceived this research project, carried out the optical experiments and wrote the paper. Z.G.C., N.L.W., R.Y.C. and R.Y. analyzed the data. G.D.G., R.D.Z. and J.S. grew the single crystals. Q.L., C.Z., 12 Y.H. and F.Q. performed the basic characterization. All authors discussed the results and commented on the paper. Figure Legends Fig. 1. Bulk-band and crystal structure of ZrTe5. (A) Top row: schematic of the topological phase transition from trivial to topological insulators (TIs). A 3D Dirac semimetal (DS) can be regarded as a quantum critical point with a gapless band structure. Due to the bulk-band inversion in TIs, the conduction and valence band exchange their extrema. Bottom row: energy spectrum for the Landau-index N = +0 and −0 LLs with a Zeeman splitting. (B) Atomic structure of ZrTe5. Each unit cell contains two ZrTe5 layers. (C) Left: AFM image of ZrTe5 flakes. Right: thicknesses along the colored lines on the left. Thicknesses from 5 to 13 unit cells (u.c.) are shown. (D) Absorption coefficient A(𝜔) of the ZrTe5 flake with thickness d ~ 180 nm as a function of photon energy at B = 0 T. The linear energy-dependence of A(𝜔) is mainly associated with inter-band transitions of 3D Dirac electrons. A sudden drop in A(𝜔) at low energies implies the modification of the band structures within an energy range, namely the bandgap. The inset of (D) depicts the inter-band absorption in gapped ZrTe5. Fig. 2. Low-magnetic-field Landau level transitions in ZrTe5. (A) Absorption features Tn (1 ≤ n ≤ 7) in T(B)/T(B0) spectra. The spectra are displaced from one another by 0.1 for clarity. (B and C) Squares of Tn * energies plotted as a function of magnetic field (B ≤ 4 T). Tn correspond to the interband LL and T1 2 (𝐵) at zero-field reveals a transitions: LL−𝑁 → LL+𝑁−1 (or LL−𝑁−1 → LL+𝑁 ). The non-zero 𝐸T1 *, which equals to bandgap ∆. The upper left inset of (C) shows the energy difference between T1 and T1 *, which is located at energies lower the bandgap. (D) Absorption features T1 than T1, arises from the intraband LL transition LL+0 → LL+1 (or LL−1 → LL−0). (E) Schematic of the Landau level spectrum without a Zeeman splitting and the allowed optical transitions, shown in a 𝐸2-𝐵 plot. Fig. 3. Crossing of the two zeroth LLs of ZrTe5. (A) Split T1 transitions in the T(B)/T(B0) spectra at B ≥ 6 T. The energies of the split T1 transitions are defined by the onsets of the dip features indicated by the blue triangles. Around 16 T, four modes are present. Four modes reemerge around 27 T. (B) Color scale map of the –ln[T(B)/T(B0)] spectra as a function of magnetic field and energy. (C) Magnified view of a region in (B) to better present the cusp-like feature which is indicated by a white arrow around 18 T. The * (grey dots) energies are plotted as a function of B in (B). Here, the green measured T1 (green dots) and T1 and grey dots in (B) can be extracted from the lower-energy edges (i.e. the onsets) of the peak-feature traces in the color intensity plots (i.e. (B) and (C)) of the –ln[T(B)/T(B0)] spectra. These dots in (B) have the intensities of the color scale, respectively: 0.025 for 1𝛼, 1𝛽, 1𝛾 and 1𝜃, –0.025 for 1𝛿, 1𝜒 and 1𝜆 and * energies, based on Eq. (5) and (6) with the g-factors: geff(LL-1) = –0.075 for 1𝜀. The theoretical T1 and T1 31.1, geff(LL+1) = 9.7, and geff(LL+0) = geff(LL-0) = 11.1, the bandgap ∆ = 10 meV, the Fermi velocities ∗ T1 T1 = 4.76 × 105 m/s and 𝑣F = 4.63 × 105 m/s, are shown by the green and grey dashed curves, 𝑣F * (grey arrows) transitions. The LL respectively. (D) Schematic of the split T1 (green arrows) and T1 T1. The spectrum is produced with the above values of the g-factors, the bandgap and the Fermi velocity 𝑣F two zeroth LLs cross at a critical magnetic field Bc ≈ 17 T. The chemical potential is roughly at zero energy when the magnetic field is high enough. At B > Bc, the interband LL transitions (1𝛼, 1𝛽, 1𝛾 and * in T(B)/T(B0) spectra. T1 13 1𝛿) are gradually replaced by the intraband LL transitions (1𝜒, 1𝜃, 1𝜆 and 1𝜀), which causes the cusp-like feature shown in (C). Figures Fig. 1 14 Fig. 2 15 Fig. 3 16 

1601.06956

2

1601

GaAs integrated quantum photonics: Towards dense and fully-functional quantum photonic integrated circuits

The recent progress in integrated quantum optics has set the stage for the development of an integrated platform for quantum information processing with photons, with potential applications in quantum simulation. Among the different material platforms being investigated, direct-bandgap semiconductors and particularly gallium arsenide (GaAs) offer the widest range of functionalities, including single- and entangled-photon generation by radiative recombination, low-loss routing, electro-optic modulation and single-photon detection. This paper reviews the recent progress in the development of the key building blocks for GaAs quantum photonics and the perspectives for their full integration in a fully-functional and densely integrated quantum photonic circuit.

cond-mat.mes-hall

cond-mat

September 4, 2018 Abstract The recent progress in integrated quantum optics has set the stage for the development of an integrated platform for quantum information processing with photons, with potential applications in quantum simulation. Among the different material platforms being investigated, direct-bandgap semiconductors and particularly gallium arsenide (GaAs) offer the widest range of functionalities, including single- and entangled-photon gener- ation by radiative recombination, low-loss routing, electro-optic modulation and single-photon detection. This paper reviews the recent progress in the development of the key building blocks for GaAs quantum photonics and the perspectives for their full integration in a fully-functional and densely integrated quantum photonic circuit. GaAs integrated quantum photonics: Towards compact and multi-functional quantum photonic integrated circuits Christof P. Dietrich1,*, Andrea Fiore2, Mark G. Thompson3, Martin Kamp1, and Sven H ofling1,4 1. Introduction Quantum information science using the unique features of quantum mechanics - superposition and entanglement - can greatly enhance computational efficiency, communication security, and measurement sensitivity. The generation of quantum bits (qubits) and the realization of quantum gates are the prerequisites for conducting quantum information science. The use of photons as qubits has been widely con- sidered as one of the leading approaches to encode, transmit and process quantum information because of their low de- coherence, high-speed transmission and compatibility with classical photonic technology [1 -- 3]. Photonic qubits can easily be encoded in many different degrees of freedom, e.g. polarization, path, frequency and orbital angular mo- mentum, making the implementation of qubits using single photons very attractive. However, single photons are subject to substantial losses. Furthermore, nonlinear interactions between single photons are weak and do not provide π cross phase modulation in natural nonlinear materials, which is the requirement for achieving a two-qubit gate. In 2001, a major breakthrough known as the KLM scheme [4] showed that scalable quantum computing is possible by only combining single-photon sources, linear optical circuits and single-photon detectors. In this case, the non-linearity in the detection replaces a direct optical non-linearity. The resulting quantum gates are probabilistic, but the probability of success can be made as close to one as desired by increasing the number of resources (ancilla photons). In addition, it was proposed that quantum logic networks can be simulated by simple one-way quantum computation using a particular class of entangled states, so called cluster states [5]. On a parallel line, a particularly simple class of linear-optics quantum simulators, boson sampling circuits, were theoretically shown to implement computational problems which are classically intractable [6]. By following these approaches of only using linear opti- cal components, two-qubit [7] and three-qubit gates [8], simple quantum algorithms [9, 10] as well as boson sam- pling [11 -- 14] could already be experimentally demonstrated. Early demonstrations of linear-optics quantum processing have relied on inefficient and bulky single photon sources based on spontaneous parametric down conversion (SPDC), modest efficiency (at near-infrared wavelengths) single pho- ton detectors based on avalanche photodiodes (APDs) or superconducting nanowires, and optical circuits with bulk optical elements. However, similar to integrated electron- ics, the practicality and scalability of quantum information technology ultimately requires the integration of individual components on a single chip. Integrated quantum photonics is emerging as a promising approach for future quantum information science, since it enables a substantial improve- ment in performance and complexity of quantum photonic circuits and provides routes to scalability by the on-chip generation, manipulation and detection of quantum states of light. Major progress has been made recently towards 2 COBRA Research Institute, Eindhoven University of Tech- 1 Technische Physik, Universitat Wurzburg, Am Hubland, 97072 Wurzburg, Germany 3 Centre for Quantum Photonics, H. H. Wills Physics Laboratory and Department of Electrical and nology, 5600 MB Eindhoven, The Netherlands Electronic Engineering, University of Bristol, Woodland Road, Bristol BS8 1UB, United Kingdom 4 SUPA, School of Physics and Astronomy, Univer- sity of St Andrews, North Haugh, St Andrews KY16 9SS, United Kingdom * Corresponding author: e-mail: christof.dietrich@physik.uni-wuerzburg.de Copyright line will be provided by the publisher 2 C.P. Dietrich et al.: GaAs integrated quantum photonics highly efficient integrated single photon sources, single pho- ton detectors and integrated photonic circuits. In particular, linear quantum photonic integrated circuits (QPICs) for two-photon interference, CNOT-gates and en- tanglement manipulation could already be demonstrated on various platforms and with various materials. These include e.g. silica-on-silicon [13, 15, 16], laser direct-writing silica [17], gallium nitride [18], lithium niobate [19], silicon-on- insulator [20, 21] and GaAs [22]. Indium arsenide/gallium arsenide (InAs/GaAs) quantum dots (QDs) are routinely embedded in photonic crystal waveguides/cavities and have been established as robust and efficient single-photon sources [23 -- 27]. Moreover, high-efficiency superconduct- ing nanowire single-photon detectors based on GaAs and Si waveguides have been successfully demonstrated [28 -- 31]. Among all these platforms, GaAs is a well-known, ma- ture material system for classical integrated photonics. It allows for the fabrication of low-loss waveguides and its high refractive index enables tight confinement of light and therefore compact devices and circuits. It has been employed in GHz modulators due to its high χ (2) nonlinearity [32]. In the context of quantum information science applications, the large electro-optic effect in GaAs makes it a very promis- ing candidate for fast routing and manipulation of single- photons. Single-photon emission in GaAs can be realized by either spontaneous parametric down-conversion in waveg- uides [33], taking advantage of the high χ (2), or by integrat- ing single QDs into nanophotonic structures. Single-photon sources based on parametric generation of photon pairs and heralding must be operated at relatively low average photon numbers, resulting in a low single-photon probability, un- less complex multiplexing schemes are employed [34]. The QD approach, facilitated by the direct bandgap of GaAs, enables efficiencies close to 100% [35] and represents a fundamental advantage of GaAs QPICs with respect to Si or LiNbO3 circuits. Besides single-photon emission, also single-photon de- tectors on GaAs waveguides have already been successfully demonstrated [28]. Therefore, GaAs waveguide circuits monolithically integrated with single-photon sources and superconducting single-photon detectors offer a promising approach to large-scale quantum photonic integrated cir- cuits. Here, we review recent developments in InAs/GaAs QD single-photon sources (Sections 2-4), ridge waveguide quantum photonic circuits (Section 5) and GaAs waveguide single-photon detectors (Section 6), and we also discuss the challenge of monolithic integration of individual com- ponents (Section 7) towards the realization of dense and fully-functional quantum photonic integrated circuits. Re- cent promising developments in parametric generation of photon pairs [33] and entangled photons [36], including by electrical injection [37], have been reviewed recently in Ref. [38] and are therefore not discussed here. 2. Semiconductor quantum dots Integrated quantum photonic experiments rely on the gener- ation, manipulation and detection of single photons that are Figure 1 Left: Transmission electron micrograph of an InAs QD (bright area) in a GaAs matrix (dark surrounding) grown by MBE in Stranski-Krastanov growth mode. The white scale bar represents 10 nm. Right: SEM image of uncapped InGaAs QDs. ideally created on demand in a two-level system. Besides natural two-level systems such as single atoms, molecules or defect centers, also nanometer-sized inclusions of a low- gap semiconducting material incorporated into a high-gap matrix delivers a two-state configuration by forming a quasi zero-dimensional potential trap. Those energy traps are called semiconductor QDs (QDs). Single-photon sources based on QDs have the huge advantage that they can be grown monolithically with monolayer precision under con- trolled conditions by well-established growth techniques such as molecular beam epitaxy (MBE) [39 -- 42] or metal- organic vapor phase epitaxy (MOVPE) [43 -- 45]. The unique optical properties of semiconductor QDs have enabled exten- sive research and tremendously increased the scientific in- terest in realizing photonic QD devices. QDs are nowadays widely used in applications such as light-emitting diodes or solar cells. Regarding integrated quantum photonics, numer- ous proof-of-principle experiments were carried out on QDs, including single photon emission [46], two-photon interfer- ence [47, 48], polarization-entanglement [49] and strong coupling [50, 51], underpinning their promising applicabil- ity as qubit source in quantum communication schemes. The most extensively studied QD materials are InAs and InGaAs in GaAs matrices (see Figure 1). The huge difference in band-gap energies between InAs (Eg(2 K) = 0.422 eV) and GaAs (1.522 eV) and the three-dimensional quantum confinement makes it possible to tune the QD emission in a very large spectral window - from almost 850 nm up to 1400 nm - by adjusting the QD dimensions. In the following, we will highlight achievements in the growth of QDs (Section 2.1), their positioning on determined sites (Section 2.2) and their optical properties (Section 2.3). 2.1. Self-assembled quantum dots The lattice mismatch between InAs and GaAs is about 7% giving rise to considerable strain when the two materials are deposited on top of each other. This circumstance is used in the most exploited growth mode for QDs - the Stranski- Krastanov growth mode [52] - which makes use of the fact that coherent, dislocation-free islands form self-assembled as result of strain compensation. The size of the islands is Copyright line will be provided by the publisher 3 Figure 3 Top row: Atomic force micrographs from QDs grown by droplet epitaxy before (left) and after annealing (right) at 400◦C. ©IOP Publishing. Reproduced with permission. All rights reserved. Bottom row: Scanning tunnel micrographs of monolayer fluctu- ations just after growth interruption for bottom and top QW in- terfaces (left and right). Adapted with permission from Ref. [70]. Copyrighted by the American Physical Society. sion properties of QDs are determined by the strength of the quantum confinement effect which can be altered in several different ways. Besides changing the vertical height of QDs by e.g. partial capping, the capping layer composition can be optimized to alter the strain state and heterostructure potential in the QD, or QDs can be annealed after growth leading to an inter-diffusion of gallium and indium at the QD surface. The diffusion process undermines the quan- tum confinement and causes a blueshift [56 -- 58] whereas an InGaAs capping layer reduces the strain in the QD and reduces the inter-diffusion, leading to a red-shift of the QD emission [59, 60]. The control of the density of QDs is crucial for the per- formance of single photon experiments since it requires the ability to optically address individual QDs by conven- tional spectroscopic methods. Several routes were already demonstrated including the change of growth parameters such as growth temperature [61], III/V-ratio [62] or growth interruptions [63]. One of the most reliable ways to control the density of QDs (over several orders of magnitude) is the change of the growth rate, or by varying the amount of deposited material (approaching the 2D to 3D material growth transition range that can be controlled by system- atically applying material flux gradients). In this way, den- sities of self-assembled QDs as low as < 0.1 µm−2 can be achieved [61, 64]. Alternatives to growing self-assembled QDs in the Stranski-Krastanov growth mode are techniques such as droplet epitaxy [65] or the control of monolayer fluctuations in quantum wells [66, 67] (see Figure 3). Droplet epitaxy is based on the affinity of some group-III elements to form Copyright line will be provided by the publisher Figure 2 Schematic capping process of a pyramidal InAs QD (a) overgrown by GaAs layers with increasing thicknesses (b-e). Reprinted with permission from [53]. ©2008, AIP Publishing LLC. thereby extremely sensitive to the amount of material that is deposited. The growth of InAs on a GaAs (100) surface ini- tially results in a thin 2D wetting layer. Due to the lattice- mismatch, the two-dimensional growth mode turns into a three-dimensional growth after deposition of a few mono- layers resulting in the creation of randomly positioned QDs with a pyramidal shape. In order to prevent the dots from oxidation and to separate them from the surface (for the in- tegration into photonic nanostructures, see Section 4), they are commonly overgrown by a capping layer of GaAs com- pleting the three-dimensional quantum confinement. The capping changes the shape of the QDs from purely pyra- midal to truncated pyramidal (see Figure 2) [53]. During the capping process, intermixing between the capping ma- terial and the QDs might occur leading to the formation of In(Ga)As QDs. This intermixing strongly affects the QD composition and potential profile as well as the QD emis- sion. In recent years, several techniques have been developed that facilitate the direct manipulation of QD properties in- cluding their size, their size distribution, their density, their shape and emission wavelength. The QD size can be var- ied by changing the composition of the dots [54]. The size distribution of QDs directly determines the inhomogeneous linewidth of the QD ensembles: large size variations lead to an unwanted broadening of the emission. A way to reduce the size distribution spread would be partial capping and annealing: an almost uniform height of QDs can be achieved by introducing an annealing step during the process of QD capping [56]. This process is usually accompanied by a blueshift of the QD emission due to the height reduction. Besides the control of the uniformity, the growth condi- tions can be used to change the shape of the confinement potential and thereby the emission energy. Indeed, the emis- 4 C.P. Dietrich et al.: GaAs integrated quantum photonics droplets on a semiconductor surface and has the advantage that it does not require a lattice-mismatch between the in- volved materials [68, 69]. For droplet epitaxy, droplets of group-III elements are annealed after growth in an arsenic atmosphere and, under optimum annealing conditions, form defect-free QDs by saturating with As (see Figure 3 top). Compared to the Stranski-Krastanov growth, QDs grown by droplet epitaxy are relatively large. Alternatively, strain-free QDs can be achieved by interrupting the growth of thin quantum wells. These growth interruptions intentionally introduce monolayer fluctuations (Figure 3 bottom) at the interface effectively creating an additional quantum confine- ment within the quantum well plane. This type of QD has a larger lateral extension in the layer plane than QDs grown by the Stranski-Krastanov mode or droplet epitaxy [70], lead- ing to a larger oscillator strength. They further benefit from the fact that they are not subject to material intermixing. 2.2. Site-controlled quantum dots The implementation of semiconductor QDs as sources of flying qubits into quantum integrated circuits requires a precise control over the relative position of the QD with respect to the circuit. Small misalignment (e.g. of a QD in the center of a defined PhC cavity) can lead to severe reductions in device efficiency. Adequate control can be achieved by either placing the circuit around the random position of a self-assembled QD [71, 72] or by predetermination of both the QD and circuit location. The first approach makes a pre-characterization of the QD distribution necessary, followed by a top-down etching around a chosen dot. However, this process is tech- nologically very challenging and seems incompatible with the upscaling to large quantum photonic circuits, as the de- vice layout must be adapted to the QD spatial positions. The growth of QDs on predetermined sites provides ac- curate alignment (see Figure 4) of single photon sources with respect to photonic circuits [73] and facilitates high yield device fabrication. Several strategies were developed in the past to precisely position QDs on a semiconductor platform [74]. This includes e.g. the deposition of optically active QDs on an optically inactive stress layer [75]. A more common strategy is the pre-patterning of the semiconductor surface by e.g. drilling holes into it [76 -- 83]. This can be achieved by a combination of electron beam lithography and ion etching, by atomic force nano-lithography [84], by nanoimprint lithography [85] or by local oxidation nano- lithography [86]. During subsequent regrowth the adatoms preferentially nucleate at the patterned sites and form QDs. This process can provide an accuracy of the QD position of about ±50 nm which in many cases is sufficient for the application in photonic circuits [87]. Similar results can be obtained by growing site-controlled QDs using masked surfaces [88]. In experiments and applications exploiting multipho- ton interference (see section 3.3), QD emission linewidths are an important figure of merit and should ideally reach transform-limited values. In this regard, the linewidth has Figure 4 Atomic force micrograph of site-controlled QDs with 1 µm lattice period in a square lattice grown on a mesa structure. The scale bar represents 1 µm. Reprinted with permission from [93]. ©2008, AIP Publishing LLC. been found to strongly depend on the proximity to heteroin- terfaces and free surfaces, due to the effect of fluctuating charge states [89]. For optimized growth conditions and advanced sample designs, record linewidth values as low as 7 µeV were reported for randomly occupied QD sites [90] and around 20 µeV were demonstrated for QD patterns with only one QD at each site [79, 91] (p-shell excitation, see Section 3.1). In this regard, site-controlled QDs are compa- rable to self-assembled QDs in terms of emission linewidths. Another crucial feature for photonic applications is the spec- tral width of the QD ensemble, which is mainly determined by size and shape fluctuations of the grown dots. Whereas self-assembled QDs exhibit inhomogeneous broadening in the range of several tens of meV, the nucleation on predeter- mined sites can reduce this value to only a few meV [92]. 2.3. Optical properties of quantum dots Most common barrier and QD material combinations, in- cluding the extensively investigated combination of InAs QDs in GaAs barriers, form a type-I heterostructure result- ing in strong quantum confinement for both electrons and holes. In good approximation, it can be assumed that the QD behaves as a two-level system, where the lowest-energy electronic excitation (heavy-hole exciton, X) involves one electron in the conduction band and one hole in the valence band with strong heavy-hole character (note that by apply- ing elastic stress to initially unstrained QDs the excitonic ground state can also have light-hole character [94]). Higher occupied states such as biexcitons or trions are detuned in energy due to Coulomb interactions. The degeneracy of the QD valence band is usually lifted by the asymmetric shape of the confinement potential and by strain causing a non-vanishing energy splitting between heavy and light hole levels on the order of several tens of meV. For typical self-assembled QDs the splitting is so large that the transi- Copyright line will be provided by the publisher excitons (X), biexcitons (XX) as well as trions (X−, X+) can be seen in Fig.5. 5 XX → σ− X and σ− Biexcitons are particularly interesting since they posses a net projection of the angular momentum of 0 and are therefore not subject to exchange interactions. If the fine- structure splitting is suppressed, this leads to the decay into two bright excitons with opposite circular polarization. This decay scheme has extensively been investigated for the creation of polarization-entangled photon pairs in QDs systems [49, 102 -- 104]. For the entanglement, the two decay paths σ + X (with σ being the pho- ton polarization state) need to be indistinguishable. This is typically not the case due to finite fine structure split- tings. Therefore, in QD systems with fine structure splittings smaller than the exciton linewidths (as for QDs grown on GaAs(111) layers [105]), entangled photon pairs can deter- ministically be created [106]. Other techniques such as ther- mal annealing [107], the application of electric fields [108], strain [109], as well as the shaping of the confinement po- tential [104] are also able to reduce the FSS and produce entangled photon pairs. XX → σ + 3. Quantum dots as single photon sources On-chip single-photon sources can be realized by exploit- ing the radiative recombination from an excitonic state of a single QD [46]. Single QDs have been widely investigated as single-photon and entangled-photon sources (see [110] for a review). As compared to single-photon sources based on parametric down-conversion and heralding, QD-based sources present the advantage of deterministic, on-demand single photon emission, much easier filtering of the pump due to the lower powers needed (tens of nW level) and the possibility of pumping from the top - this is particularly rel- evant in the context of quantum photonic integrated circuits as it reduces the coupling of pump light into the chip. 3.1. First- and second-order coherence In general, quantum emitters can be classified regarding their coherence properties and photon statistics. The de- gree of first-order coherence is quantified by the field-field correlation function g(1) and the degree of second-order co- herence is described by the intensity-intensity correlation function g(2). g(1) of a quantum emitter is usually probed in optical interferometers where the optical beam is split into two arms and joined again after introducing a variable time delay τ into one of the arms. The light produced by an ideal single-quantum emitter with a lifetime T1 is de- scribed by constant T2 = 2T1 - corresponding to a single-photon pulse with no phase jumps. However, real quantum emitters such as excitons in QDs are imperfect systems and show faster decays of as phonon interactions. Dephasing reduces the coherence (cid:12)(cid:12)(cid:12) which decays exponentially with a time (cid:12)(cid:12)(cid:12) caused by dephasing mechanisms such (cid:12)(cid:12)(cid:12)g(1)(τ) (cid:12)(cid:12)(cid:12)g(1)(τ) Copyright line will be provided by the publisher Figure 5 Non-resonant photoluminescence spectrum of a GaAs QD showing exciton (X), biexciton (XX) and trion (X−, X+) tran- sitions. The upper panel schematically depicts the QD charge configuration for each transition. Adapted with permission from Ref. [95]. Copyrighted by the American Physical Society. tion between conduction band and light-hole valence band can easily be neglected making the system a true two-level system. The total angular momentum J of heavy-hole excitons is a sum of heavy-hole angular momentum and electron spin allowing four different transition configurations having an angular momentum along the growth direction: Jz,1 = ±1 and Jz,2 = ±2. The Jz,1 configurations are termed bright ex- citon states as they couple to the optical field and therefore preferentially decay through radiative recombination chan- nels. In contrast, configurations with total angular momen- tum Jz,2 are termed dark exciton states and decay through non-radiative recombination channels. Due to the close prox- imity of both electrons and holes in semiconductor QDs, exchange interactions of electron-hole pairs are enhanced and cause an energetic splitting between dark and bright states as well as the formation of mixed dark and mixed bright states. Bright and dark exciton decay bi-exponentially caused by radiative and non-radiative recombination as well as spin flip processes (turning bright states into dark states, and vice versa, under creation or annihilation of LA phonons) [96, 97]. While the pure states are circularly polarized, they mix as a consequence of exchange interactions and of the pref- erential QD elongation along the in-plane [110]-direction for growth on GaAs(100) substrates, giving rise to mixed states with linear polarization planes parallel to the [110]- and [110]-crystal direction [98,99]. Those linearly polarized transitions can directly be recorded in luminescence experi- ments [100]. Their energetic separation is the so-called fine structure splitting (FSS). Dark states have typical fine struc- ture splittings in the order of 1µeV [101] whereas bright states can have FSS up to several hundreds of µeV [98] (few to few tens of µeV are more commonly observed). Besides bright and dark exciton states, also states with higher occupation such as trions (i.e. charged excitons, X− or X+) or biexcitons (XX) occur. Owing to Coulomb interac- tions between confined carriers, these states have recombina- tion energies different from the exciton transitions allowing the spectral filtering of individual excitonic transitions. A typical QD spectrum showing the radiative transitions of 6 C.P. Dietrich et al.: GaAs integrated quantum photonics Figure 6 Non-resonant, quasi-resonant and resonant excita- tion scheme: high-energy particles (dotted circles) decay through phonon-interactions into the QD ground state (solid circles). The vertical arrows indicate the energy necessary for each process. 2 (with T∗ time of the QD resulting in 1/T2 = 1/2T1 + 1/T∗ 2 being the pure dephasing time) and increases its linewidth. Dephasing processes are detrimental in quantum photonic applications as they make photons emitted from different sources distinguishable and reduce the visibility in two- photon interference experiments. The occurrence of dephasing is strongly related to the excitation scheme used for the experiment. Nowadays, three main different schemes are distinguished: off-resonant, quasi-resonant (p-shell excitation) and resonant pumping (s-shell excitation), see Figure 6 for the different energy configurations (less common excitation schemes omitted here include wetting layer excitation or excitation above shell resonant to an LO phonon replica). The excitation of a QD well above its ground state or resonant with an excited state (e.g. p-shell) results in the relaxation of carriers into the ground state by generation of phonons [111]. In gen- eral, QDs always emit and absorb phonons through inelastic processes. Especially, LA phonons are well known sources for dephasing and cause an asymmetric broadening of the QD transition (as visible in the calculated absorption spec- trum shown in Figure 7) [112 -- 116]. The interaction of QD excitons with phonons can either be suppressed by perform- ing the experiment at very low temperatures or by adapting resonant s-shell excitation (see Section 3.2). We note that off-resonant and quasi-resonant excitation also produce a certain time jitter in the generation of single photon events, reducing the indistinguishability [117, 118, 121]. Another dephasing process is induced by the random distribution of charge carriers inside a QD and changes in the electronic configuration around the QD. Charge and spin fluctuations result in a persistent change of the excitonic ground state inducing spectral wandering of the QD emission (see Figure 8). This leads to an effective increase of the QD linewidth and a decrease of the QD coherence time [122]. Charge fluctuations can be suppressed drastically by employing resonant pumping and adding a weak auxiliary continuous wave reference beam to the excitation beam of the QD [123]. Spectral wandering is typically much slower than the radia- tive decay of excitons and its effect can in some cases be circumvented by pumping the QD with a short pulse twice at short time intervals to obtain two photons with very sim- ilar energies [124]. However, when interference between photons from different QDs is required, spectral wandering Figure 7 Calculated absorption spectra of an InAs QD at differ- ent temperatures. At low temperatures, a clear asymmetric broad- ening of the zero-phonon line (ZPL) caused by exciton-phonon- interactions with longitudinal acoustic phonons is visible. Inset: Calculated broadening of the ZPL compared with experimental results from [111]. Reprinted with permission from Ref. [112]. Copyrighted by the American Physical Society. Figure 8 (a) Typical spectrum of InAs/GaAs QDs measured through an aperture with 100 nm width. (b) Temporal evolution of (a) showing spectral wandering of almost all QD transitions. Reprinted with permission from Ref. [127]. Copyrighted by the American Physical Society. produces a loss of indistinguishability in the same way as pure dephasing, and must be suppressed [125, 126]. Photon statistics can be probed by measuring the second- order correlation function g(2)(τ) that gives the probability of detecting two photon events with time delay τ between them. In this regard, the g(2)(0) value (at zero time delay) is particularly important since it measures the probability that two photons are detected at the same time. Experimentally, g(2)(τ) is routinely determined in a Hanbury Brown and Copyright line will be provided by the publisher Twiss (HBT) setup that consists of a 50:50 beam splitter and two nominally equal single-photon detectors with a tempo- ral resolution much better than τ. Whereas for a coherent and thermal light source g(2)(0) approaches 1 and 2, respec- tively, a perfect single photon source gives g(2)(0) = 0 since the probability of detecting two photons events at the same time is then zero. The photon statistics of real single-photon sources are strongly affected by non-idealities such as multi-photon emission from multi-exciton states, collection of the emis- sion of other QDs, re-pumping of the emitter after the emis- sion of the first photon, etc., and therefore produce non-zero coincidences [115, 116]. Resonant, pulsed excitation can lead to g(2)(0) values very close to zero [124, 128, 129]. Pumping by adiabatic rapid passage using chirped laser pulses can further suppress multi-photon emission [130]. Even with these highly sophisticated excitation and detec- tion schemes, multi-photon emission in QD systems can never be turned off completely. In this regard, the measured g(2)(0) value can be seen as a measure of the quality of the single-photon source [128, 129]. 3.2. Resonance fluorescence As mentioned above, resonant s-shell excitation circumvents several dephasing processes, such as the creation of high- energy carriers or the interaction with phonons, leading to a considerable increase of the coherence time of QDs [131]. However, the monochromatic pumping of a two-level system at resonance offers many more advantages including the coherent generation [117, 132 -- 135], manipulation [134, 136 -- 138] and characterization [139] of the excitonic states. In general, the spectral and temporal properties of pho- tons scattered by a two-level system are determined by the laser detuning from the excitonic resonance. The electric- dipole interaction is responsible for a coupling of the exci- tonic states to the driving field and turns the two levels of the exciton into "dressed states" [140]. The dressed-state picture allows four radiative transitions of which two are de- generate and two are spectrally displaced by the Rabi energy EΩ (see Figure 9 top right). This results in the emergence of three emission lines, the so-called Mollow triplet [141], with the central line having the highest intensity (Figure 9 left). The Rabi energy scales with the excitation density of the driving field (as shown in Figure 9 bottom right). This enables a direct control of the side band transitions. By in- troducing a second laser beam in resonance with one of the sideband transitions (preferably the lower energy sideband), the Mollow triplet turns into a multitude of peaks under suppression of the main peak (due to deconstructive interfer- ence). In these multiply dressed QD states also phenomena like the multi-photon AC Stark effect can be present that have previously only been observed in atomic physics [142]. Whereas the above mentioned phenomena can be ob- served by using continuous laser irradiation, pulsed resonant driving of a two-level system (with ultra-short pulses) fur- ther facilitates the external control of the QD population as a consequence of the oscillatory behavior of the QD 7 Figure 9 Left: Excitation-dependent PL spectra of a resonantly- pumped single QD. EΩ = ¯hΩ denotes the energetic splitting be- tween zero-phonon line and its first Mollow side band. Top right: Evolution of QD transitions in the "dressed" picture. Bottom right: The determined sideband splitting ¯hΩ vs. square root of the excita- tion power showing a linear dependence. Adapted with permission from Ref. [117]. Copyrighted by the American Physical Society. population versus excitation pulse area (so-called Rabi os- cillations [143, 144]). In this way, the creation of single photons can drastically be increased by applying pulse areas matching the maximum QD population (π-pulses). In the Mollow triplet regime, most of the light is scat- tered incoherently owing to the saturation of the QDs. Re- ducing the strength of the driving field changes the scattering properties of QDs [123]. In the so-called Heitler regime (at very low excitation powers) the QD behaves like a passive scatterer and incident photons are scattered almost fully coherently [141]. As a consequence, the spectrum of coher- ently scattered photons is only determined by the spectral characteristics of the excitation source [145]. In this way, QDs linewidths as low as 0.03 µeV have been observed already. Those QDs are in the "subnatural linewidth" [146] or "ultracoherence" regime [147]. The application of pulsed excitation schemes in combination with the weak coherent scattering regimes enables pure QD spectroscopy in the ab- sence of almost all dephasing mechanism and, with that, facilitates the deterministic generation of single photons. However, the low photon generation rates are a disadvan- tage. At moderate excitation powers, coherent scattering and the Mollow triplet can also coexist when the resonantly driven dot is thermalized with the phonon bath [148]. Experimentally, the observation of resonance fluores- cence is challenging. Sophisticated resonant electrical injec- tion is possible [149] but technologically very challenging. More common optical pumping schemes suffer from prob- lems related to the separation of the excitonic emission from the scattered laser light requiring elaborated sample designs or advanced excitation and detection schemes. A simple way of performing optically pumped resonance fluorescence ex- periments is the use of planar waveguides [132, 150, 151]. The laser light is coupled to the waveguide mode exciting the QDs which emit perpendicular to the sample plane. Another approach to the pump suppression is the cross-polarization Copyright line will be provided by the publisher 8 C.P. Dietrich et al.: GaAs integrated quantum photonics pulses temporally separated by a 2 ns delay [47]. A variable mirror able to compensate for the delay time was placed at one of the beam splitter outputs, a fixed mirror at another output. In this way, five different scenarios were created of photons impinging on the detectors of which only one creates a temporal overlap of both photons showing a dip in the correlation function. So far, two-photon interference could be demonstrated for very different systems including dissimilar (between a QD and a Poissonian laser [160], a parametric down-conversion source [161] or a frequency comb [162]) and similar single photon sources (between different QDs) [137, 163]. In most cases, indistinguishabil- ity has been demonstrated between QDs of two different and spatially separate samples [48, 125, 126], but in the perspective of quantum photonic integrated circuits it is of special importance to demonstrate two-photon interference between single photon sources on the same semiconduc- tor chip (Figure 10). This has been achieved for two QDs on the same substrate with only 40 µm distance between them [156]. The experimental difficulty here is to find two QDs within the field of view that have almost identical emis- sion energies, linewidths and polarization. A much more practical approach is the tuning of the excitonic QD transi- tions by changing the properties externally, e.g. by applying an electric field. See section 4.3 for a review about tuning mechanisms. 4. Purcell-enhanced single-photon emission Intrinsic dephasing processes (e.g. phonon scattering) and extrinsic spectral fluctuations, related to varying charge en- vironment in the vicinity of the QD, typically limit the coherence time of single QDs to T2 ≈ 100− 200 ps, while the natural exciton lifetimes are limited to T1 > 400 ps. The exploitation of two-photon interference in photonic quan- tum information processing (QIP) requires the reduction of the lifetimes in order to achieve T2 ≈ 2T1. This can be achieved employing cavity quantum electrodynamic effects in nanophotonic structures that allow the precise tailoring of the electromagnetic field surrounding of the dot. In this way, the rate of radiative recombination from the embed- ded emitter can be controlled via the Purcell effect [164] in e.g. a cavity. The enhancement factor of the spontaneous emission rate is hereby determined by the spatial and spec- tral matching between emitter and cavity mode, the quality factor Q and the mode volume V of the cavity mode in the photonic nanostructure [165]. In particular, for a spectrally narrow source in resonance with a broader optical mode, the Purcell-enhancement is proportional to the Q/V ratio (bad cavity regime). In order to approach an almost perfect cavity-emitter interface by increasing the light-matter inter- action, photonic nanostructures need to provide very high quality factors and low mode volumes. 4.1. Cavities Typical examples of photonic nanostructures include mi- crodisks, micropillars, nanobeams, photonic crystal slabs Copyright line will be provided by the publisher Figure 10 (a) Second-order correlation function g(2)(τ) for two QDs on the same chip separated by 40 µm. (b) Two-photon inter- ference of the two QDs from (a). The grey line (IRF) in (a) displays the instrument response function. Adapted with permission from Ref. [156]. Copyrighted by the American Physical Society. technique that requires a high extinction rate between both polarizations [134]. In general, the detection of resonance fluorescence is always aggravated by scattering at imper- fections. This circumstance becomes even more delicate in photonic nanostructures such as ridge waveguides [152,153] and photonic crystals [154,155]. (Pulsed) resonance fluores- cence has so far not been observed in waveguide-coupled photonic crystal cavities. 3.3. Two-photon interference When two single photons with the same Fourier-transform limited spectrum, temporal profile and polarization from two spatially separate sources propagate with a linear network, one cannot distinguish which of the two photons stems from one or the other source, i.e. they are indistinguishable [157]. Due to the interference of the probability amplitudes, two indistinguishable photons impinging on the two inputs of a balanced beam splitter (50:50) will always exit together (so-called photon-bunching). The experimental demonstra- tion of photon-bunching was first achieved by Hong, Ou and Mandel in 1987 [158] and is one of the main prerequi- sites for quantum information science being the basis for e.g. KLM quantum gates [15] and boson sampling [6]. This so-called two-photon interference is usually measured using a Michelson interferometer consisting of a beam splitter, two distinct optical paths and two single-photon detectors at the two outputs of the splitter (an equivalent configura- tion based on a Mach-Zehnder interferometer can also be used). A temporal displacement between both inputs can be introduced either by changing the position of the beam split- ter [158] or by introducing a variable delay line into one of the photon paths [159] for making the photons gradually dis- tinguishable. For single photons from semiconductor QDs, two-photon interference was first demonstrated for photons from the same source by exciting the QD with two short 9 Figure 11 Overview over different GaAs-based cavity types providing three-dimensional photonic confinement: (a) microdisks [166], (b) micropillars, (c) nanobeams [166], (d) PhC slabs and (e) three-dimensional photonic crystals. The respective record optical Q-factors for GaAs-based photonic nanocavities embedding QDs are given at the bottom of each image together with estimated mode volumes. (b) Adapted with permission from Ref. [50]. Copyrighted by the American Physical Society. (d,e) Adapted by permission from Macmillan Publishers Ltd: Sci. Rep. [167], ©2013, and Nat. Photon. [168], ©2011. and 3D photonic crystals (see Figure 11 for an overview). All structures provide an effective three-dimensional pho- tonic confinement based on the possibility to manipulate and control light propagation by introducing interfaces with very high refractive index contrast. Obviously, the best index contrast for GaAs-based devices is given for the interface between GaAs (with n = 3.5) and air (n = 1). In microdisks, light is confined by total internal reflec- tion (TIR) at the GaAs/air boundary and propagates within the disk cross section in the vicinity of the disk rim. The so-called whispering-gallery modes possess very high qual- ity factors of up to Q = 100,000 [169] but suffer from the relatively large mode volumes. Even higher Q-factors can be observed in micropillar resonators (Q can be as high as 250,000 [170]) where Bragg mirrors (alternating pairs of dielectric materials with high index contrast and λ /4- thickness for constructive interference) below and on top of the cavity region confine photons in the pillar effectively. Depending on the pillar diameters mode volumes as low as 2.3(λ /n)3 [171] can be achieved. However, while mi- cropillars with embedded QDs are ideal vertical-emitting single-photon sources, they do not match the requirements of quantum photonic integrated circuits that involve in-plane photon routing and processing. In this regard, photonic crys- tals slabs embody the perfect compromise by facilitating high quality factors, low mode volumes and in-plane con- trol. They are based on Bragg scattering in periodic photonic structures which are most commonly realized by drilling air holes in a lattice-geometry into a membrane of semicon- ductor material. In-plane confinement is then achieved by the photonic band gap introduced by the holes etched into the membrane, while out-of-plane confinement fully relies on total internal reflection at the membrane-air interface. Photonic crystals can be fabricated in square or triangular lattice whereas the latter is most commonly preferred as it provides a larger photonic band gap [172]. PhC cavities are formed by displacing neighboring holes (H0 cavity [173]) or leaving out a single (H1 cav- ity [174, 175]) or several air holes, thereby forming a defect in the photonic crystal bandgap. In the prospect of single- photon emission, L3 cavities (three missing holes in a line) are most promising as they support polarized single modes with a wide spectral margin [176]. The highest Q-factors for GaAs were so far reported for passive L3 cavities in a triangular lattice with reduced hole radius [177] and shifted next neighbor positions [178] with experimental values up to 700,000 [179, 180] (calculated Q-values exceed 1× 106 for optimised L3 designs [181]). Active structures (with embedded QDs) show quality factors up to Q = 35,000 (25,000) and mode volumes V /(λ /n)3 = 0.9(0.4) for L3 (H1) cavities [182]. Note that the difference between Q- values with and without QDs may be related to the different wavelength range of operation (typically around 900 nm with QDs, therefore in the spectral region of band-tail ab- sorption in GaAs) or to residual non-resonant absorption by the QDs. Comparable results have been obtained using one-dimensional PhCs in nanobeams [166, 183]. Three-dimensional PhCs, for example based on the woodpile structure [168] (see Figure 11), may, in princi- ple, offer higher quality factors due to suppressed leakage in the vertical direction and additionally provide better con- trol of the QD spontaneous emission. However, they are technologically much more challenging to fabricate. 4.2 Photon collection and routing in waveguides Quantum photonic integrated circuits are two-dimensional arrangements of different functionalities on one semicon- Copyright line will be provided by the publisher 10 C.P. Dietrich et al.: GaAs integrated quantum photonics Figure 12 Left: Calculated band structure (inside first Brillouin zone) for TE-modes of an infinite PhC waveguide. The black lines indicate y-polarized modes, the gray lines indicate x-polarized modes. Right: Scanning electron microscope image of the PhC waveguide. The scale bar represents 1 µm. Reprinted with per- mission from [25]. ©2011, AIP Publishing LLC. ductor chip that allow the generation, manipulation and detection of single photons. This requires efficient photon collection from single photon sources, interconnections be- tween the individual functionalities as well as low-loss trans- port of light within the circuit from the stationary qubit to the manipulation and detection sites. Waveguides (such as PhC waveguides [23, 25 -- 27, 184] or ridge waveguides, see Section 5) perfectly meet these requirements as they effi- ciently collect photons emitted by the QDs (even in the absence of a cavity) and transport them via propagating modes. PhC waveguides have the huge advantage of a tight mode confinement paving the way for a high density of components on one chip. On the downside, they suffer from optical losses and mechanical instabilities. One way of efficiently collecting photons in a single guided mode is the use of a PhC waveguide [185 -- 187] (e.g. "W1" waveguide, based on a missing row of holes in a tri- angular lattice of holes patterned in a semiconductor slab). The radiative bandgap created by the PhC suppresses the emission into in-plane modes other than the intended mode (see Figure 12), resulting in a large (β = 80− 90%) sponta- neous emission coupling factor (defined as the fraction of spontaneous emission coupled to the desired mode). This is true even in the absence of a strong Purcell effect. The β -factor can be even higher in case the QD is spectrally close to the bottom of the dispersion curve. Then the emis- sion into the guided mode is enhanced due to a decreased group velocity, further increasing β . Efficient funneling of the emission of single QDs into a PhC waveguide mode has been observed experimentally for QDs emitting in the 900 nm [26] range but also for the telecommunication range around 1300 nm [27]. In this context, β -factors as high as 0.98 [24, 35] and Purcell enhancements of up to Fp = 2.7 [188] were already reported. Figure 13 Left: Band structure of a Stark-tunable device consist- ing of InAs QDs grown within GaAs quantum wells and AlGaAs superlattices. Right: Field-dependence of two QDs (Dot1 and DotA) as well as the cavity mode. Reprinted with permission from Macmillan Publishers Ltd: Nat. Photon. [48], ©2010. 4.2. Tuning Quantum Dot and Cavity Mode In a perfect cavity-emitter interface, both QD transition and cavity mode have the same energy for maximum Purcell en- hancement. However, in real photonic nanostructures with embedded QDs it is most likely that both resonances are slightly detuned from each other as result of growth and processing imperfections. This circumstance makes exter- nal tuning mechanisms indispensable. The simplest way of tuning the QD excitonic emission with respect to the cavity mode is by changing the sample temperature in a temperature-variable cryostat [23, 51, 189], by laser irradi- ation or by implementing heat pads close to the photonic crystal [190,191]. All these mechanisms will affect the elec- tronic band structure, the refractive index of the device and the lateral expansion of the cavity. The two latter changes only slightly affect the spectral position of the cavity mode, while the bandgap change strongly tunes the QD energy. Ex- perimentally more challenging are other post-growth tuning mechanisms such as the application of external magnetic fields [192] or strain [193,194]. A more integrated approach is the implementation of electrical contacts into the pho- tonic nanostructures and shifting the QD transitions via Stark tuning by applying an electric field across the QD layer [27, 48, 195 -- 200]. Additional heterostructure barriers (see Figure 13) around the QD suppress carrier tunneling out of the dot enabling a tuning range of several tens of meV [48]. Whereas a single exciton can easily be tuned into reso- nance with a cavity mode using the Stark effect, more elab- orated integration schemes such as the interference of two indistinguishable single photons from two QDs in two sepa- rate photonic surroundings require the ability to control both the QD and cavity frequencies independent from each other so that different QD and cavity lines can be all brought to resonance [201]. In this regard, Stark tuning only affects the QD exciton energy and does not shift the cavity resonance. The separate tuning of PhC cavity resonances is a challeng- ing task, especially at the low temperatures of interest. Sev- eral methods for cavity tuning have been proposed, includ- ing the use of photosensitive materials [202], wet-chemical etching [203], nano-oxidation techniques [204 -- 207], the Copyright line will be provided by the publisher 11 Figure 14 Left: Schematic representation of an electrostatically tunable double-membrane PhC cavity showing the field distribu- tions of symmetric (s) and antisymmetric (as) mode. Right: PL of the antisymmetric L3 mode of a double-membrane PhC cavity as function of the DC bias. The L3 mode gradually shifts in res- onance with the QD lines (by sweeping the bias from 0 to -4.8 V. Reprinted with permission from [216]. ©2012, AIP Publishing LLC. insertion of gases [208] or liquids [209 -- 211] as well as near- field probes [212, 213]. However, all these techniques are not very convenient in the prospect of realizing fully inte- grated quantum photonic circuits. None of these approaches allows the real-time tuning of each cavity in an array, as needed for QPICs. The latter functionality can be achieved using nano-electro-mechanical actuation, for example us- ing a double-membrane cavity [214] (see Figure 14). This was recently demonstrated in [215]. In this structure, the PhC mode extends over two closely-spaced membranes, and an electromechanical control of their distance results in a change of the mode wavelength, potentially over several tens of nm. With this structure, electrical tuning of the PhC resonance over >10 nm at 10 K was obtained [216] (see Fig- ure 14), without affecting the QD emission lines. Combining this cavity tuning with the Stark tuning of the QD excitons results in widely tunable single-photon sources to be used as key building blocks for scalable integrated photonic quan- tum processing [201]. Note that those electromechanical PhC cavities can also be used as tunable filters - a much needed component in quantum photonic integrated circuits with detectors. 5. Photon Routing and Manipulation 5.1. Photon transport in ridge waveguides While PhC WGs are attractive for the modification of the LDOS and the optimization of the efficiency, a photonic circuit fully based on the photonic crystal geometry typi- cally suffers from high optical losses (although losses down to 8 dB/cm have been demonstrated on GaAs [217]) and substantial structural instability. A more efficient way of transporting single photons through photonic circuits is the use of ridge waveguides (RWGs) as they exhibit good mode confinement and low losses (< 1 dB/cm [217]). This re- quires the efficient coupling of single photons from PhC waveguides to low-loss, supported ridge waveguides. Cou- pling from the PhC waveguide into the RWG using a simple, Figure 15 Top: sketch of the interface between PhC waveguide, suspended nanobeam and ridge waveguide. Bottom: scanning electron microscopic image of the fabricated waveguide devices at the interface of suspended and ridge waveguide. Reprinted with permission from [218]. ©2013, AIP Publishing LLC. single-step lithographic process was recently demonstrated showing coupling efficiencies of up to 70% [218]. It is based on tapering the mode in both the lateral and the vertical di- rection by gradually changing the width of the waveguide (see Figure 15). By using this structure, single photons were coupled to a ridge waveguide at a rate as high as 3.5 MHz, which exceeds the typical rates obtained by coupling single- photon emission from QDs into fibers, showing the huge advantage of integration. 5.2. Optical phase shifters Phase tuning and/or photon switching is needed in many im- plementations of photonic QIP, such as boson sampling [13]. Additionally, fine tuning of the splitting ratio of integrated beamsplitters is often required to correct fabrication im- perfections. These circumstances require a robust way to manipulate the phase of photons in a given waveguide of an individual input arm. Typically, this is achieved by using electro- or acousto-optical modulators which are based on external modulations of the refractive index of the medium in which the single photons propagate. Nowadays, electro-optic phase modulators (EOPMs) are routinely fabricated and operated by applying a voltage across a non-centrosymmetric material (e.g. GaAs) vertical to the propagation direction through metallic electrodes. The voltage causes a change of the refractive index as function of the applied electric field. In general, the quality of an EOPM is related to VπL, the voltage length product required for a phase shift of π, and should ideally be (cid:28)1 Vcm to enable dense packaging of photonic components. GaAs presents an enhancement of the electro-optic effect at energies very close to the bandgap [219]. In this wavelength region it has a very large electro-optic coefficient (γ = 2.4× 10−11 m/V [220]) and relatively low values of VπL = 0.21 Vcm [221]. Suspended GaAs waveguides can confine the optical mode very tightly due to the high index contrast between Copyright line will be provided by the publisher 12 C.P. Dietrich et al.: GaAs integrated quantum photonics 5.3. Beam splitters, directional couplers and interferometers Quantum integrated photonic circuits not only provide a platform to monolithically integrate single-photon sources, large-scale quantum circuits and detectors, but also offer high-visibility classical and quantum interference due to the inherent stability of the interferometers and perfect mode- overlap at the beam splitters. Using dual-rail (or path) en- coding, an arbitrary single path-encoded qubit can be rep- resented as a superposition of two states: α 10(cid:105) + β 01(cid:105) (in this notation the state nm(cid:105) indicates the presence of n photons in one waveguide and m in the other). An integrated optical beam splitter, usually implemented with a 50:50 directional coupler or a multimode interference (MMI) cou- √ pler, can easily perform a Hadamard operation and produce a superposition state such as (10(cid:105) +01(cid:105))/ 2 from a single photon propagating in one waveguide. Moreover, when two indistinguishable photons meet at the Hadamard gate, quan- tum interference produces photon bunching and a maximally entangled state (both photons are either reflected or transmit- ted). The initial input state 11(cid:105) is then transformed into the √ entangled state (20(cid:105) +02(cid:105))/ 2 after the Hadamard opera- tion. The same quantum interference, in combination with detection, is also at the origin of the photon non-linearity exploited in linear optics quantum computing. This key effect was already demonstrated in GaAs waveguides [22, 224] where directional couplers were de- signed and fabricated using GaAs/AlGaAs ridge waveguides with widths and heights of 2-4 µm. Figure 16 shows the cal- culated field distribution for the quasi-TE fundamental mode and schematic image of the fabricated GaAs/AlGaAs direc- tional couplers. A propagation loss of 2.5-3.5 dB/cm and a coupling loss of 2-2.5 dB/facet were measured using the Fabry-P´erot method. The coupling ratio ε of the directional coupler is defined as the ratio of reflectivity and transmis- sion. By varying the length of the coupler, an arbitrary value for ε can be set. As can be seen in Figure 16, a near 50:50 directional coupler with two bent input/output waveguides can be realized when the coupling length is about 140 µm for a 2.5 µm gap. Single photons pairs were launched into the 50:50 directional coupler and the coincidence counts af- ter the coupler were measured using a counting module and two single-photon detectors. By changing the time arrival of the two photons, a Hong-Ou-Mandel (HOM) dip with a high visibility of almost 95% was observed as shown in Figure 16. On-chip beamsplitters were recently integrated with single-photon sources based on QDs [224 -- 226]. Besides the creation of entangled states by using direc- tional couplers, also arbitrary unitary operations of quan- tum states are required to implement quantum communica- tion and universal quantum computing. In order to prepare and measure arbitrary path-encoded qubits, two Hadamard gates and three phase shifters are required (representing a Mach-Zehnder interferometer), one of which induces a relative phase and amplitude between the two waveg- uides (see Figure 17). The Mach-Zehnder interferometer then transforms a single-photon quantum state 10(cid:105) into Copyright line will be provided by the publisher Figure 16 Top left: Cross section of a GaAs/AlGaAs ridge waveg- uide and the simulated field distribution of the fundamental TE- mode. Top right: Schematic image of a GaAs directional coupler. Bottom left: Coupling ratio of the couplers for different gaps as function of coupling length. Bottom right: Two-photon quantum in- terference in the directional couplers with a coupling ratio ε ≈ 0.5 showing high visibility of almost 95%. Reprinted from [22] , ©2014, with permission from Elsevier. GaAs and air. By interconnecting the waveguide with bridges for electrical connection, only very small voltages are necessary to achieve large electric field strengths as re- sult of the short inter-electrode distance. An effective doping scheme with gradually doped p- and n-layers can further reduce Vπ [221]. A different way of changing the optical phase of a pho- ton wave is the structural modification of the waveguide. This can be achieved for example through the application of a surface acoustic wave via a piezoelectric transducer perpendicular to the propagation direction of photons. The acoustic wave generates regions of compression and extrac- tion. This changes the refractive index of the waveguide material via the elasto- and electro-optic effect due to the applied strain and the inherent piezoelectricity of GaAs, re- spectively. These acousto-optic modulations can operate up to GHz frequencies [222]. However, those transducers are hard to implement, require substantial power and cannot be used to fix the phase over long time scales. A very promising alternative for the phase tuning and re- configuration of QPICs is given by the use of low-frequency micro- and nano-mechanical structures, where a physi- cal displacement is produced via electrical actuation (e.g. through capacitive forces). Nanomechanical phase modula- tors have been demonstrated, for example in silicon [223]. The double-membrane structure used for the tuning of PhC cavities [215, 216] lends itself to this type of phase modula- tors, potentially featuring VπL products in the order of 0.005 Vcm and maximum frequencies in the MHz range. 13 Figure 17 Top left: Schematic image of the Mach-Zehnder interferometer (MZI) with two directional couplers and two electro-optical phase shifters. Top right: Optical image of the fabricated MZIs. Bottom left: Normalized intensities of the two outputs as function of relative phase shift for coherent bright-light input (for a coupling ratio of ε = 0.3). Bottom right: Quantum interference fringes showing manipulation of the two-photon state. Reprinted from [22], ©2014, with permission from Elsevier. [(1− 2ε)cos(θ /2) + ısin(θ /2)]10(cid:105)+2ı(cid:112)ε(1− ε)cos(θ /2)01(cid:105) where θ is the relative phase between the two arms [22]. Thus, a simple integrated Mach-Zehnder interferometer (MZI) with two 50:50 directional couplers and three phase shifters can prepare and manipulate an arbitrary quantum state. GaAs/AlGaAs MZIs consisting of two identical 50:50 directional couplers and two electro-optical phase shifters on the top of two arms were demonstrated in [22], a schematic representation and an optical microscopic image can be seen in Figure 17. The working principle was demon- strated by using both a classical light source and a single photon source. Classical light showed sinusoidal oscillations at both outputs. Their obvious unbalance was ascribed to the non-perfect coupling ratio of ε = 0.3. When two single pho- tons are separately launched into the two input ports of the Mach-Zehnder interferometer, the first coupler creates the For ε = 0.5, the two photons are maximally path-entangled. The phase shifters then perform a rotation and the second coupler transforms the state to two-photon state(cid:112)2ε(1− ε)ı(20(cid:105) +02(cid:105)) + (1− 2ε)11(cid:105). (cid:104)−εe−2ıθ + (1− 2ε)e−ıθ + 1− ε (cid:105)20(cid:105) ı(cid:112)2ε(1− ε) (cid:105)02(cid:105) (cid:104) +ı(cid:112)2ε(1− ε) (cid:104)−2ε(1− ε)e−2ıθ + (1− 2ε)2e−ıθ − 2ε(1− ε) (cid:105)11(cid:105) (1− ε)e−2ıθ + (1− 2ε)e−ıθ − ε + The manipulation of the two-photon entanglement state can be seen in Figure 17 showing two-photon quantum interfer- ence with a doubled frequency compared to the classical light experiment. 5.4. Switching and routing There are several ways of controlling the propagation of single photons by using the inherent optical properties of quantum dots. The two most prominent examples herein are single-photon switching and spin-photon routing. Single-photon switches are systems based on nonlin- ear optical interactions [227, 228] in which the injection of control photons strongly affects the propagating of sig- nal photons. This effect was already observed in strongly coupled QD-cavity systems (where strong coupling sets in when the coherent coupling constant of the QD-cavity sys- tems becomes larger than the photon decay rate) [229 -- 231]. In this case, the energy structure, described by the Jaynes- Cummings ladder, is anharmonic, which means that, by exciting the system at the frequency of the transition to the upper polariton state in the first manifold with a control pho- ton, the scattering rate of photons in the second manifold is changed drastically [229]. Typical switching times lie in the ps-range. Single-photon switches thus may represent a key element in future high-bandwidth QPICs. Another way of externally manipulating the propagation of single photons is given by the exploitation of the QD spin degree of freedom. Spin states such as those in a semicon- ductor QD are considered to be very good qubits [232] as they possess coherence times in the tens of nanoseconds range, which can be extended to the microseconds range by spin-echo techniques [233]. The spin of resident electrons in QDs can be mapped onto the circular polarization state of photons emitted by the related charged exciton (trion) [234], when an additional exciton is pumped in the QD [235]. However, this in-plane circular polarization state is difficult to map into polarization- or path-encoded photons propa- gating along the layers. Additionally, the strong birefrin- gence of nanophotonic waveguides makes the transmission of polarization-encoded qubits difficult. This difficulty can be circumvented by a proper design of the electromagnetic environment around the QD. For example, by placing a QD at the crossing point of two or- thogonally arranged photonic waveguides (Figure 18, left) and detecting the output at the waveguide ends, the state of a spin-polarized exciton can be mapped onto the propaga- Copyright line will be provided by the publisher 14 C.P. Dietrich et al.: GaAs integrated quantum photonics Figure 18 Left: SEM image of perpendicular waveguides with grating outcouplers at each end. The QD is located at an off-center position (90 nm away from the center) as indicated in the inset. Right: PL spectra from the off-center QD excited with circularly polarised light and recorded at opposite grating outcouplers for magnetic fields between ±4 T. Reprinted with permission from [237]. ©2011, AIP Publishing LLC. tion direction of the generated photons [236]. Although this experiment is hardly compatible with the KLM scheme in terms of scalability, its outcome has triggered the idea of using QD spin states to transfer vertically coupled photon polarization degrees into in-plane path qubits. It was found that, by placing the QD off-center (see Figure 18), the indi- vidual spin polarization can deterministically be addressed (and directed within the photonic circuit) by using circularly polarized light [237]. In the same way, when placing a QD into the so-called C-point of a W1 PhC waveguide, in prin- ciple, unidirectional emission and even an entangled photon source can be achieved [238]. Recently, this scheme was experimentally demonstrated in a glide-plane waveguide whose mirror-symmetry is broken by gradually shifting the inner hole row away from the W1 defect (towards one end of the waveguide) leading to a preferential emission of pho- tons from a circularly polarized dipole into one of the two directions (depending on the helicity) [239]. 6. Single-photon detectors 6.1. Superconducting nanowire detectors The integration of single-photon detectors (SPDs) with quan- tum photonic integrated circuits is particularly challenging, as the complex device structures associated with conven- tional single-photon detectors, such as avalanche photodi- odes, are not easily compatible with the integration with low-loss waveguides and even less with sources. Transition- edge sensors (TES) may be suited for integration [29], but they are plagued by very slow response times (typically lead- ing to maximum counting rates in the tens of kHz range) and require cooling down to < 100 mK temperatures. In- stead, the "hot-spot" detection mechanism in superconduct- ing nanowires [240 -- 242] is much more promising for fast single-photon detection. This detection mode can be imple- mented in a GaAs photonic circuit by sputtering an ultrathin (4-5 nm) NbN film on top of GaAs/AlGaAs waveguide het- erostructures and by its subsequent patterning into narrow wires [28, 243]. The evanescent field of the guided mode in- teracts with the wires (see Figure 19a), resulting in a modal absorption coefficient of several hundreds of cm−1, ensuring nearly 100% absorptance in a propagation distance of a few tens of µm. The devices need to be cooled down to cryogenic tem- peratures and biased with a current Ib close to the critical current Ic of the superconducting wire. When the photon is absorbed, a region with a high concentration of quasi- particles is formed ("hot-spot"), leading to a local disruption of the superconductivity and resulting in a resistance appear- ing in the wire through a process involving the creation and crossing of vortices [244, 245]. The bias current is corre- spondingly expelled from the wire to the load formed by the input resistance of an amplifier producing a voltage pulse. The typical shape of such an output pulse is shown in Figure 19, revealing a time constant being related to the recovery of the bias current in the wire after superconductivity has been re-established. This time constant (typically in the few ns range) is limited by the wire's kinetic inductance [246]. Maximum counting rates in the 100 MHz range can be ob- tained, orders of magnitude higher than with TES detectors. The jitter in the output pulse, measured to be 60 ps in waveguide SPDs on GaAs [28] and even shorter in SPDs on Si [30], is also much better than in TES and avalanche photodiodes, allowing much higher temporal resolution in single-photon measurements. The quantum efficiency (QE) of superconducting SPDs is observed to strongly depend on the bias current, with a maximum close to the critical current, since the proba- bility of switching the wire from the superconducting to the resistive state is maximum there. Besides that, also the nanowire geometry [247] as well as fabrication imperfec- tions such as nanowire width and film thickness fluctua- tions [248] strongly affect the quantum efficiency. The QE is further a function of the position across the nanowire width at which the incident photon is absorped [245]. In GaAs-based waveguide SPDs, maximum device quantum ef- ficiencies of about 20% (defined as the photocounts normal- ized to the number of photons coupled into the WG for the transverse-electric (TE) polarization) were measured [28], which are promising for applications in quantum photonic integrated circuits. Note that even higher efficiencies were obtained on Si/SiO2 [30, 249, 250] or SiNx [251, 252] due to the possibility to deposit NbN at higher temperatures. Improvement of the quality of NbN films on GaAs, the use of lower-gap superconductors such as WSi [253, 254] and the use of suspended or PhC GaAs waveguides [255] should result in QEs approaching 100%, as needed in large-scale quantum photonic integrated circuits. The yield of SPDs on a chip is also a key issue for the scalability of integrated circuits. NbN-based SPDs are typ- ically plagued by the low yield of efficient devices, which has been attributed to the presence of localized defects (constrictions) [256] and continuously distributed inhomo- geneities [248]. These regions with lower local critical cur- rent prevent a uniform biasing of the detector and thereby limit the efficiency. Their effect is expected to be less critical Copyright line will be provided by the publisher 15 vertical direction and temporally filtering the QD emission from the short pump pulse [153]. 6.3. Photon-number resolving detectors Apart from their use as autocorrelators, multi-wire struc- tures can also be configured to work as photon-number- resolving (PNR) detectors [261]. In this case, instead of reading out each wire separately, the wires are connected in parallel [261] or in series [262,263] in order to provide a sin- gle voltage output proportional to the number of switching wires. In particular, the series configuration has been shown to be attractive in view of the potential for high efficiency and scalability to large photon numbers [262, 264]. For ex- ample, as shown in Figure 20, four wires can be patterned on top of the same waveguide, and connected in series, with an integrated resistor in parallel to each wire [265]. When a wire switches to the resistive state after absorption of a guided photon, the bias current flowing through it is diverted to the parallel resistor, producing a voltage pulse. When two or more wires switch, the sum of the corresponding voltages is read out in the external circuit [265]. Such waveguide PNR detectors will perform a fundamental function in fu- ture fully-integrated quantum photonic circuits, enabling photon number measurements with high efficiency in an extremely compact footprint. 7. Integration perspectives and outlook Figure 21 shows an illustration of a simple, possible quan- tum photonic integrated circuit, which enables to perform a two-photon interference measurement. In this circuit, single photons are generated by the radiative recombination of QD excitons placed in a PhC waveguide. The QD emission is tunable via the Stark effect created by the introduction of p- and n-doped regions into the PhC. The PhC waveguide transports the single photons towards ridge waveguides after being spectrally filtered by a PhC cavity in order to isolate photons from individual QD transition lines. The spectral filter itself can be tuned via a second underlying membrane (not shown). The single photons are then transferred into the beam splitter (directional couplers) where they exit in bunched pairs as result of their indistinguishability. The superconducting nanowire single photon detectors subse- quently measure the second-order correlation function of the beam splitter output to prove the interference of single photons from two remote sources. This scheme can easily be extended to perform measurements of higher complexity by simply increasing the amount of individual units (sources, splitters and detectors) if a high efficiency is obtained in all parts of the circuit. In this way, multi-photon experiments such as boson sampling or advanced quantum photonic gates (e.g. reconfigurable chips [266]) can be integrated and scaled to large photon numbers. All the building blocks for a multi-functional QPIC (such as illustrated in Fig.21), including single-photon Copyright line will be provided by the publisher Figure 19 Left: Sketch of a waveguide superconducting single- photon detector. Right: Device quantum efficiency (red dots, left axis) and dark count rate (blue dots, right axis) as a function of the normalized bias current. Inset: Output electrical pulse, showing a 1/e time constant of 3.6 ns. Reprinted with permission from [28]. ©2011, AIP Publishing LLC. in WSi films due to the larger hot-spot diameter [253] and the correspondingly lower dependence on the bias current. 6.2. Advanced functionalities with integrated detectors The waveguide configuration provides an opportunity to integrate additional functionality with the detector. For ex- ample, in [30], an optical delay line constituted by a micror- ing resonator was integrated with a nanowire detector on a Si waveguide. Nanophotonic cavities were integrated with superconducting nanowires, enabling spectral selectivity and near-unity efficiency [257]. Superconducting nanowire detectors were also integrated with Si-based small-scale quantum photonic circuits [252]. In the GaAs platform, mul- timode interference (MMI) couplers, Mach-Zehnder inter- ferometers and modulators can easily be integrated with detectors. The detector structure itself can also be tailored to perform advanced measurements such as measuring the second-order auto-correlation function by integrating sev- eral wires on a waveguide [258]. In this case, two NbN nanowires sense the electric field of the same guided mode. The nanowires are separately connected to two different bias and amplification circuits, providing two distinct read- outs of the photo-response signals, similar to the free-space configuration reported in [259]. By combining the detector outputs in a correlation card, the second-order correlation function of light propagating in the waveguide can be mea- sured directly [258]. This integrated auto-correlator provides the advantage of a very small footprint and a reduction in the total length of NbN wires needed to achieve a given absorp- tion probability. Despite the very close physical proximity (wire-to-wire distance of 150 nm), no crosstalk (spurious switching of one wire after detection of a photon in the other wire) was observed, within the experimental accuracy [258]. The on-chip detection of QD emission embedded in a ridge GaAs waveguide was recently demonstrated [153,260]. A major challenge in this regard is the spectral filtering of single QD emission lines and the discrimination of scattered laser light which can be achieved by pumping from the 16 C.P. Dietrich et al.: GaAs integrated quantum photonics Figure 20 Left: Oscilloscope persistence map of a photon flux detected with a waveguide photon-number resolving detector (consisting of four superconducting nanowires) showing 0-4 photon events. The left axis shows a corresponding histogram collected in a time frame indicated by the rectangle. Right: Count rates of the waveguide photon-number resolving detector corresponding to different photon counting levels: 1-photon (red), 2-photon (blue), 3-photon (green), and 4-photon (purple). Inset: the signal amplitude as a function of the detected photon number. Reprinted with permission from [265]. ©2011, AIP Publishing LLC. devices, a photon generation probability of 6× 10−2 per pulse [218] and a detection quantum efficiency of 0.2 [28] were reported, corresponding to a potential overall efficiency in the 10−2 range if the sources and detectors can be coupled efficiently on the same chip. We note that nearly the same waveguide epitaxial de- sign was used for both the source and the detector, making their efficient coupling possible. Significant improvements in both integrated sources and detectors are possible, for example using nearly or strictly resonant excitation of the QD [124, 268], optimized PhC cavity design and process- ing, and improved superconducting nanowire technology, potentially bringing the overall efficiency in the tens of percent range. Additionally, integrating a large number of such devices on a single chip is in principle straightforward. Passive waveguide circuits are readily implemented with GaAs ridge waveguides. While the quantum interference experiments reported in Section 5.2 were carried out using waveguides with larger cores to optimize coupling to fibers, similar structures can easily be designed and realized based on the tightly confined ridge waveguide design used for sources and detectors. Waveguide losses were estimated to be in the few cm−1 range for the relatively unoptimized technology used in [218] and could certainly be improved. Due to the compact size of the needed components, waveg- uide loss is not expected to represent a major limitation for experiments involving a few photons. A few challenges still need to be addressed before GaAs QPICs can be used to perform multi-photon experiments beyond the state-of-the-art: (i) Integrated filtering and buffering: Tunable filters are always part of table-top quantum photonic experiments based on QDs, as they are needed to discriminate the emis- sion from single excitonic lines, particularly in the case of non-resonant or near-resonant excitation. Filters can be re- alised using PhC cavities, and can be integrated with either Copyright line will be provided by the publisher Figure 21 Schematic illustration of a simple quantum photonic integrated circuit able to perform a second-order auto-correlation measurement and to demonstrate two-photon interference of single photons that are generated (by Purcell-enhanced emission from a QD in a PhC waveguide), filtered (by the PhC cavity), processed (by the two directional couplers) and detected (by the two superconducting nanowire detectors) on the same chip. sources, detectors and passive circuits, have been demon- strated on the GaAs platform. The corresponding efficien- cies, while still below the values (>99%) needed for scal- ing to many qubits, are already attractive as compared to the combination of coupling loss and efficiency typically affecting free-space or fiber implementations. Taking the case of a QD-based single-photon source at telecom wave- length as an example, the overall source-detector efficiency is in the 5× 10−4 range when using conventional micro- photoluminescence systems and commercial fiber-coupled SSPDs (resulting from a 5× 10−3 probability per pulse of generating a single photon into a single-mode fiber and 10−1 detector quantum efficiency) [267], although higher ef- ficiencies can be achieved for both the fiber-coupled source (up to 6× 10−2 [268]) and the detector (up to 0.93 [254]). In the very first demonstrations of ridge-waveguide coupled the source or the detector. They can be made tunable by us- ing the same electromechanical actuation needed for tuning the source [216]. Their integration in QPICs is therefore in principle straightforward. Another much needed component for quantum photonic integrated circuits is a single-photon buffer that is able to store single photons without destroying their non-classical nature and entanglement. Single-photon buffers can be realized by utilizing the control of the group index in PhC waveguides [269], although it is challenging to combine a buffering time larger than the photon temporal length and low loss. (ii) Improving the yield of sources and detectors and reducing losses of photonic components: Scaling QPICs for multi-photon experiments requires arrays of sources and detectors with high yield and photonic components with vanishing optical losses. Increasing the yield in detector fabrication relies on improvements in superconducting thin film technology, as discussed above, with good outlook for the future. Achieving a high yield in sources, besides a relatively straightforward integration of the tuning tech- niques discussed in Section 4.3, further requires the active control of the QD spatial position, in order to achieve ef- ficient coupling with the optical mode of the PhC cavity. Promising results have been shown in the position control of both strained InAs/GaAs QDs [90,91,270,271] and pyra- midal GaAs/AlGaAs QDs [272] (as discussed in Section 2.2), making the perspective of their integration in tunable PhC structures realistic. In this context, photonic compo- nents (such as waveguides, couplers etc.) still suffer from high optical losses that currently make possible up-scaling schemes unfeasible. Advanced lithographic steps need to be implemented in order to reduce losses and be able to implement true functionalities. (iii) Improving the coherence and photon indistin- guishability: A major limitation of solid-state single-photon sources is the dephasing introduced by the environment as described in Section 3.1 and can be classified into pure dephasing (phase fluctuations on timescales shorter than the exciton lifetime) and spectral wandering (variations of the QD energy levels induced by a fluctuating charge en- vironment on timescales longer than the exciton lifetime but shorter than the experiment duration). Both processes affect the photon indistinguishability and therefore the in- terference visibility observed in photon bunching experi- ments [47]. Recent results on near-unity indistinguishability of photons generated by a single QD using pulsed resonant fluorescence [124] and quantum interference of photons generated by two distinct QDs [48, 133, 156] suggest that these limitations can be surmounted by resonant pumping techniques and a careful control of the QD environment. Another very interesting route to the suppression of spectral wandering is the real-time control of the QD energy through a feedback loop [273]. In any case, coherence remains the most challenging issue for the application of QD sources to multi-photon experiments. Further down the line, high-density integration on GaAs will require increasing the refractive index contrast to achieve smaller radii of curvature in waveguide bends. This can be done by replacing the conventional GaAs/AlGaAs 17 waveguide structure by a GaAs/SiO2 structure, presenting a similar index contrast as Si/SiO2. The bonding of GaAs membranes on SiO2-coated Si substrates can be used for that purpose and has been demonstrated [274]. In this per- spective of large-scales QPICs, residual waveguide loss will represent an important parameter and will have to be inves- tigated carefully. We also note that other III-V active lay- ers and membranes (and particularly those lattice-matched with InP) may be used instead of GaAs, leveraging on the vast developments in InP classical photonics. However, the progress on QD single-photon sources on InP has so far been limited [275], and it is unclear what the limits will be in terms of photon indistinguishability. A comparison between GaAs and other quantum pho- tonic technologies is useful. The LiNbO3, Si and SiN plat- forms can also in principle provide the key functionalities of photon production, passive control and detection, while the integration of efficient sources on silica-based QPICs appears very challenging. In terms of integration density, the small index contrast in LiNbO3 represents a major funda- mental limitation, making its application to large-scale cir- cuits involving tens of photons and thousands of components doubtful. Si and SiN therefore represent the strongest com- petitors to GaAs for integrated quantum photonics. These materials benefit from decades of investments and technol- ogy developments in the electronics industry, allow fabri- cating devices with high quality and reproducibility, and are in principle compatible with production in CMOS fabs. The index contrast in both systems is high, enabling high integration levels. Indeed, fast progress, particularly in quan- tum silicon photonics, has recently led to the demonstration of QPICs with tens to hundreds of components [21, 276]. However, fundamental challenges remain in the large scale integration of sources, detectors and low-power reconfig- urable circuits on the Si and SiN platforms. On the one hand, single-photon sources in these materials can only be based on two-photon production via spontaneous four-wave- mixing from a pump laser and heralding of a non-vacuum state by detection of one of the photons. This approach is intrinsically limited in terms of efficiency (probability of generating a single photon) since the average photon num- ber must be kept low to avoid multi-photon events. A po- tential solution is the multiplexing of several sources [277], which implies a very significant increase in the number of components and puts stringent requirements on the loss of switches [20, 278]. On the other hand, the integration of superconducting nanowire detectors implies additional tech- nological challenges on the Si and SiN platform, related to the extreme filter performance required to suppress the pump used for photon production, the impossibility to use thermal phase tuning at low temperature, and the incom- patibility of most superconducting materials with CMOS fabs. A solution to these problems might lie in the use of hybrid platforms such as III-V materials on silicon by direct heteroepitaxial growth on top of each other or by wafer bonding [279]. While it is unclear at this point which technology will be suited for large-scale QPICs, the physical properties of GaAs and other III-Vs, including the direct bandgap Copyright line will be provided by the publisher 18 C.P. Dietrich et al.: GaAs integrated quantum photonics and the non-centrosymmetric crystal structure (enabling a linear electro-optic effect), provide them a very fundamental advantage which could become crucial in the long term. 8. Conclusion We have provided an overview of the recent progress in quantum integrated photonic components and circuits based on the GaAs technology platform. All key functionali- ties, including single-photon sources and detectors, photon- number-resolving detectors, integrated autocorrelators and tunable Mach-Zehnder interferometers have been realized and tested, and on-chip photon-photon interference has been demonstrated. The remaining challenges to be addressed for scaling GaAs quantum photonic integrated circuits to the level of few tens of photons have also been discussed. These results lay the foundation of a fully-integrated quantum pho- tonic technology, with potential applications in quantum simulation and computing. Acknowledgements. We gratefully acknowledge many useful discussions with S. Fattah poor, T.B. Hoang, A. Gaggero, R. Leoni, F. Mattioli, L. Midolo, M. Petruzzella, D. Sahin, J.P. Sprengers, T. Xia and C. Schneider. We further acknowledge financial support from the European Commission within the FP7 project QUANTIP (Project No. 244026) and the Dutch Technology Foundation STW (Project No. 10380). Key words: GaAs, photonic qubits, integrated circuits Christof P. Dietrich received his PhD degree in physics from the Graduate School BuildMona of University of Leipzig, Ger- many. After postdoc positions in The Netherlands (TU Eind- hoven) and the United Kingdom (University of St Andrews), he is now group leader in the Technical Physics Department of the University of Wurzburg, Germany, where he conducts projects on integrated GaAs quantum photonic structures. His current research interests also include polariton physics and organic photonics. Andrea Fiore graduated in electronic engineering and physics from the University of Rome "La Sapienza" (Italy) and obtained the PhD degree in 1997 from Univ. of Paris XI (France). He has held appointments at the University of California at Santa Barbara (USA), the Institute of Photonics and Nanotechnol- ogy of the Italian National Research Council (Italy), the Ecole Polytechnique F´ed´erale de Lausanne (Switzerland) and is now Professor of Nanophotonics at the Eindhoven University of Technology (The Netherlands). His research interests are in the field of quantum photonics and semiconductor nanopho- tonics. Mark Thompson is Professor of Quantum Photonics at the University of Bristol, UK. He received his M.Phys. degree in Physics from the University of Sheffield UK, and Ph.D. degree in Electrical Engineering from the University of Cambridge UK, in 2000 and 2007 respectively. Previous appointments include Research Scientist with Bookham Technology Ltd. UK, a Research Fellow at the University of Cambridge and a Re- search Fellow at the Toshiba Corporate R&D Centre, Japan. He joined the University of Bristol in 2008, and currently leads a team of researchers developing integrated quantum photonic technologies for applications in quantum communications and computation. Martin Kamp studied physics at the University of Wurzburg, Germany, and obtained a Master of Arts degree from Stony Brook University, USA. He received a PhD degree from the University of Wurzburg in 2003 for his research on laterally cou- pled distributed feedback lasers. Until 2014, he was interims chair for the Technische Physik for several years and head of the Gottfried-Landwehr Laboratory for Nanotechnology. Since 2014, he is head of the Wilhelm Conrad Rontgen Center for Complex Material Systems. His research interests include electronic and photonic semiconductor structures, quantum information processing and interband cascade lasers. Sven H ofling obtained the diploma degree in Applied Physics from the University of Applied Science in Coburg and his PhD degree from the University of Wurzburg. He was with the Fraun- hofer Institute of Applied Solid-State Physics (Freiburg, Ger- many). In 2013, he became Professor in Physics at the Univer- sity of St Andrews (United Kingdom). Since 2015, he holds the Chair for Technical Physics at the University of Wurzburg and is head of the Gottfried-Landwehr Laboratory for Nanotech- nology. His research interests include the design, fabrication, and characterization of low-dimensional electronic and pho- tonic nanostructures, including quantum wells and quantum dots, organic semiconductors, high-quality factor microcavities, photonic crystal devices and semiconductor lasers. References [1] J.L. O'Brien, A Furusawa and J. Vuckovi´c, Nat. Photon. 3, 687 (2009). [2] A. Faraon, A. Majumdar, D. Englund, E. Kim, M. Bajcsy, and J. Vuckovi´c, New J. Phys. 13, 055025 (2011). [3] P. Lodahl, S. Mahmoodian, and S. Stobbe, Rev. Mod. Phys. 87, 347 (2015). (2001). [4] E. Knill, R. Laflamme, and G.J. Milburn, Nature 409, 46 [5] R. Raussendorf, D.E. Browne, and H.J. Briegel, Phys. Rev. A 68, 022312 (2003). [6] S. Aaronson and A. Arkhipov, arXiv:1011.3245 (2010). [7] J.L. O'Brien, G.J. Pryde, A.G. White, T.C. Ralph, D. and Branning, Nature 426, 264-267 (2003). [8] B.P. Lanyon, M. Barbieri, M.P. Almeida, T. Jennewein, T.C. Ralph, K.J. Resch, G.J. Pryde, J.L. O'Brien, A. Gilchrist, A.G. White, Nat. Phys. 5, 134 (2009). [9] B.P. Lanyon, T.J. Weinhold, N.K. Langford, M. Barbieri, D.F.V. James, A. Gilchrist, and A.G. White, Phys. Rev. Lett. 99, 250505 (2007). [10] C.-Y. Lu, D.E. Browne, T. Yang, and J.-W. Pan, Phys. Rev. Lett. 99, 250504 (2007). [11] M. Tillmann, B. Daki´c, R. Heilmann, S. Nolte, A. Szameit, and P. Walther, Nat. Photon. 7, 540 (2013). Copyright line will be provided by the publisher [12] A. Crespi, R. Osellame, R. Ramponi, V. Giovannetti, R. Fazio, L. Sansoni, F. De Nicola, F. Sciarrino, and P. Mataloni, Nat. Photon. 7, 322 (2013). [13] J.B. Spring, B.J. Metcalf, P.C. Humphreys, W.S. Kolthammer, X.-M. Jin, M. Barbieri, A. Datta, N. Thomas-Peter, N.K. Langford, D. Kundys, J.C. Gates, B.J. Smith, P.G.R. Smith, I.A. Walmsley, Science 339, 798 (2013). [14] M.A. Broome, A. Fedrizzi, S. Rahimi-Keshari, J. Dove, S. Aaronson, T.C. Ralph, and A.G. White, Science 339, 794 (2013). [15] A. Politi, M.J. Cryan, J.G. Rarity, S. Yu, J.L. O'Brien, Science 320, 646 (2008). [16] J. Carolan, C. Harrold, C. Sparrow, E. Mart´ın-L´opez, N.J. Russell, J.W. Silverstone, P.J. Shadbolt, N. Matsuda, M. Oguma, M. Itoh, G.D. Marshall, M.G. Thompson, J.C.F. Matthews, T. Hashimoto, J.L. O'Brien, and A. Laing, Science 349, 711 (2015). [17] G.D. Marshall, A. Politi, J.C.F. Matthews, P. Dekker, M. Ams, M.J. Withford, and J.L. O'Brien, Opt. Expr. 17, 12546 (2009). [18] C. Xiong, W. Pernice, K.K. Ryu, C. Schuck, K.Y. Fong, T. Palacios, and H.X. Tang, Opt. Expr. 19, 10462 (2011). [19] H. Jin, F.M. Liu, P. Xu, J.L. Xia, M.L. Zhong, Y. Yuan, J.W. Zhou, Y.X. Gong, W. Wang, and S.N. Zhu, Phys. Rev. Lett. 113, 103601 (2014). [20] D. Bonneau, G.J. Mendoza, J.L. O'Brien, and M.G. Thomp- son, New J. Phys. 17, 043057 (2015). [21] N.C. Harris, G.R. Steinbrecher, J. Mower, Y. Lahini, M. Prabhu, T. Baehr-Jones, M. Hochberg, S. Lloyd, and D. En- glund, arXiv:1507.03406 (2015). [22] J.W. Wang, A. Santamato, P. Jiang, D. Bonneau, E. Engin, J.W. Silverstone, M. Lermer, J. Beetz, M. Kamp, S. Hofling, M.G.Tanner, C.M. Natarajan, R.H. Hadfield, S.N. Dorenbos, V. Zwiller, J.L. O'Brien, M.G. Thompson, Opt. Commun. 327, 49 (2014). [23] D. Englund, A. Faraon, I. Fushman, N. Stoltz, P. Petroff, and J. Vuckovi´c, Nature 450, 857 (2007). [24] T. Lund-Hansen, S. Stobbe, B. Julsgaard, H. Thyrrestrup, T. Sunner, M. Kamp, A. Forchel, and P. Lodahl, Phys. Rev. Lett. 101, 113903 (2008). [25] A. Schwagmann, S. Kalliakos, I. Farrer, J.P. Griffiths, G.A.C. Jones, D.A. Ritchie, and A.J. Shields, Appl. Phys. Lett. 99, 261108 (2011). [26] A. Laucht, S. Putz, T. Gunthner, N. Hauke, R. Saive, S. Frederick, M. Bichler, M.-C. Amann, A.W. Holleitner, M. Kaniber, and J.J. Finley, Phys. Rev. X 2, 011014 (2012). [27] T.B. Hoang, J. Beetz, L. Midolo, M. Skacel, M. Lermer, M. Kamp, S. Hofling, L. Balet, N. Chauvin, and A. Fiore, Appl. Phys. Lett. 100, 061122 (2012). [28] J.P. Sprengers, A. Gaggero, D. Sahin, S. Jahanmirinejad, G. Frucci, F. Mattioli, R. Leoni, J. Beetz, M. Lermer, M. Kamp, S. Hofling, R. Sanjines, and A. Fiore, Appl. Phys. Lett. 99, 181110 (2011). [29] T. Gerrits, N. Thomas-Peter, J.C. Gates, A.E. Lita, B.J. Met- calf, B. Calkins, N.A. Tomlin, A.E. Fox, A.L. Linares, J.B. Spring, N.K. Langford, R.P. Mirin, J.B. Smith, I.A. Walmsley, and S.W. Nam, Phys. Rev. A 84, 060301 (2011). [30] W.H.P. Pernice, C. Schuck, O. Minaeva, M. Li, G.N. Gol'tsman, A.V. Sergienko, and H.X. Tang, Nat. Comm. 3, 1325 (2012). [31] G. Reithmaier, J. Senf, S. Lichtmannecker, T. Reichert, F. Flassig, A. Voss, R. Gross, and J.J. Finley, J. Appl. Phys. 113, 143507 (2013). 19 [32] R.G. Walker, Appl. Phys. Lett. 54, 1613 (1989). [33] G. Leo, V. Berger, C. Ow Yang, and J. Nagle, J. Opt. Soc. Am. B 16, 1597 (1999). [34] M.J. Collins, C. Xiong, I.H. Rey, T.D. Vo, J. He, S. Shahnia, C. Reardon, T.F. Krauss, M.J. Steel, A.S. Clark, and B.J. Eggleton, Nat. Commun. 4, 2582 (2013). [35] M. Arcari, I. Sollner, A. Javadi, S. Lindskov Hansen, S. Mah- moodian, J. Liu, H. Thyrrestrup, E.H. Lee, J.D. Song, S. Stobbe, and P. Lodahl, Phys. Rev. Lett. 113, 093603 (2014). [36] A. Orieux, A. Eckstein, A. Lematre, P. Filloux, I. Favero, G. Leo, T. Coudreau, A. Keller, P. Milman, and S. Ducci, Phys. Rev. Lett. 110, 160502 (2013). [37] F. Boitier, A. Orieux, C. Autebert, A. Lemaıtre, E. Galopin, C. Manquest, C. Sirtori, I. Favero, G. Leo, and S. Ducci, Phys. Rev. Lett. 112, 183901 (2014). [38] C. Autebert, G. Boucher, F. Boitier, A. Eckstein, I. Favero, G. Leo, and S. Duccia, J. Mod. Opt. 62, 1739 (2015). [39] C. W. Snyder, B. G. Orr, D. Kessler, and L. M. Sander, Phys. Rev. Lett. 66, 3032 (1991). [40] D. Leonard, M. Krishnamurthy, C.M. Reaves, S.P. Denbaars, and P.M. Petroff, Appl. Phys. Lett. 63, 3203 (1993). [41] J.M. Moison, F. Houzay, F. Barthe, L. Leprince, E. Andr´e, and O. Vatel, Appl. Phys. Lett. 64, 196 (1994). [42] M. Grundmann, N.N. Ledentsov, R. Heitz, L. Eckey, J. Chris- ten, J. Bohrer, D. Bimberg, S.S. Ruvimov, P. Werner, U. Richter, J. Heydenreich, V.M. Ustinov, A.Yu. Egorov, A.E. Zhukov, P.S. Kopev, and Zh.I. Alferov, Phys. Stat. Sol. B 188, 249 (1995). [43] J. Oshinowo, M. Nishioka, S. Ishida, and Y. Arakawa, Appl. Phys. Lett. 65, 1421 (1994). [44] P.M. Petroff, and S.P. DenBaars, Superlattices Microstruct. 15, 15 (1994). [45] F. Heinrichsdorff, A. Krost, M. Grundmann, D. Bimberg, A. Kosogov, and P. Werner, Appl. Phys. Lett. 68, 3284 (1996). [46] P. Michler, A. Kiraz, C. Becher, W.V. Schoenfeld, P.M. Petroff, L. Zhang, E. Hu, and A. Imamoglu, Science 290, 2282 (2000). [47] C. Santori, D. Fattal, J. Vuckovi´c, G.S. Solomon, and Y. Yamamoto, Nature 419, 594 (2002). [48] R.B. Patel, A.J. Bennett, I. Farrer, C.A. Nicoll, D.A. Ritchie, and A.J. Shields, Nat. Photon. 4, 632 (2010). [49] N. Akopian, N.H. Lindner, E. Poem, Y. Berlatzky, J. Avron, D. Gershoni, B.D. Gerardot, and P.M. Petroff, Phys. Rev. Lett. 96, 130501 (2006). [50] J.P. Reithmaier, G. Sek, A. Loffler, C. Hofmann, S. Kuhn, S. Reitzenstein, L.V. Keldysh, V.D. Kulakovskii, T. L. Reinecke, and A. Forchel, Nature 432, 197 (2004). [51] T. Yoshie, A. Scherer, J. Hendrickson, G. Khitrova, H.M. Gibbs, G. Rupper, C. Ell, O.B. Shchekin, and D.G. Deppe, Nature 432, 200 (2004). [52] I.N. Stranski and L. Krastanov, Mathematische Naturwis- senschaftliche Klasse 2B 146, 79 (1938). [53] H. Eisele, A. Lenz, R. Heitz, R. Timm, M. Dahne, Y. Temko, T. Suzuki, and K. Jacobi, J. Appl. Phys. 104, 124301 (2008). [54] A. Loffler, J.-P. Reithmaier, A. Forchel, A. Sauerwald, D. Peskes, T. Kummell, G. Bacher, J. Cryst. Growth 286, 6 (2006). [55] L. Wang, A. Rastelli, and O. G. Schmidt, J. Appl. Phys. 100, 064313 (2006). [56] J.M. Garcia, T. Mankad, P.O. Holtz, P.J. Wellman, and P.M. Petroff, Appl. Phys. Lett. 72, 3172 (1998). [57] T. Yang, J. Tatebayashi, K. Aoki, M. Nishioka, and Y. Arakawa, Appl. Phys. Lett. 90, 111912 (2007). Copyright line will be provided by the publisher 20 C.P. Dietrich et al.: GaAs integrated quantum photonics [58] D.J.P. Ellis, A.J. Bennett, A.J. Shields, P. Atkinson, and D.A. Ritchie, Appl. Phys. Lett. 90, 233514 (2007). [59] K. Nishi, H. Saito, S. Sugou, and J.-S. Lee, Appl. Phys. Lett. 74, 1111 (1999). [60] N.-T. Yeh, T.-E. Nee, J.-I. Chyi, T.M. Hsu, and C.C. Huang, Appl. Phys. Lett. 76, 1567 (2000). [61] B. Alloing, C. Zinoni, L.H. Li, A. Fiore, and G. Patriarche, J. Appl. Phys. 101, 024918 (2007). [62] S.L. Li, Q.Q. Chen, S.C. Sun, Y.L. Li, Q.Z. Zhu, J.T. Li, X.H. Wang, J.B. Han, J.P. Zhang, C.Q. Chen, and Y.Y. Fang, Nanoscale Res. Lett. 8, 367 (2013). [63] A. Convertino, L. Cerri, G. Leo, S. Viticoli, J. Cryst. Growth 261, 458 (2004). [64] S. Huang, Z. Niu, H.Q. Ni, Y.H. Xiong, F. Zhan, Z. Fang, J.B. Xia, J. Cryst. Growth 301, 751 (2007). [65] N. Koguchi and K. Ishige, Jpn. J. Appl. Phys. 32, 2052 (1993). [66] D. Gammon, E.S. Snow, B.V. Shanabrook, D.S. Katzer, and D. Park, Science 273, 87 (1996). [67] J. Hours, P. Senellart, E. Peter, A. Cavanna, and J. Bloch, Phys. Rev. B 71, 161306(R) (2005). [68] T. Mano, M. Abbarchi, T. Kuroda, C.A. Mastrandrea, A. Vinattieri, S. Sanguinetti, K. Sakoda, and M. Gurioli, Nan- otechnology 20, 395601 (2009). [69] P. Tighineanu, R. Daveau, E.H. Lee, J.D. Song, S. Stobbe, and P. Lodahl, Phys. Rev. B 88, 155320 (2013). [70] E. Peter, P. Senellart, D. Martrou, A. Lemaıtre, J. Hours, J.M. G´erard, and J. Bloch, Phys. Rev. Lett. 95, 067401 (2005). [71] A. Badolato, K. Hennessy, M. Atature, J. Dreiser, E. Hu, P.M. Petroff, A. Imamoglu, Science 308, 1158 (2005). [72] K. Hennessy, A. Badolato, M. Winger, D. Gerace, M. Atature, S. Gulde, S. Falt, E.L. Hu, A. Imamoglu, Nature 445, 896 (2007). [73] T. Sunner, C. Schneider, M. Strauss, A. Huggenberger, D. Wiener, S. Hofling, M. Kamp, and A. Forchel, Opt. Lett. 33, 1759 (2008). [74] O.G. Schmidt, Lateral Alignment of Epitaxial Quantum Dots (Springer, Berlin, 2007). [75] H.J. Krenner, M. Sabathil, E.C. Clark, A. Kress, D. Schuh, M. Bichler, G. Abstreiter, and J.J. Finley, Phys. Rev. Lett. 94, 057402 (2005). [76] H. Heidemeyer, C. Muller, O.G. Schmidt, J. Cryst. Growth 261, 444 (2004). [77] S. Kiravittaya, M. Benyoucef, R. Zapf-Gottwick, A. Rastelli, and O. G. Schmidt, Appl. Phys. Lett. 89, 233102 (2006). [78] P. Gallo, M. Felici, B. Dwir, K. A. Atlasov, K. F. Karlsson, A. Rudra, A. Mohan, G. Biasiol, L. Sorba, and E. Kapon, Appl. Phys. Lett. 92, 263101 (2008). [79] L.O. Mereni, V. Dimastrodonato, R.J. Young, and E. Pelucchi, Appl. Phys. Lett. 94, 223121 (2009). [80] A. Huggenberger, S. Heckelmann, C. Schneider, S. Hofling, S. Reitzenstein, L. Worschech, M. Kamp, and A. Forchel, Appl. Phys. Lett. 98, 131104 (2011). [81] M. Helfrich, P. Schroth, D. Grigoriev, S. Lazarev, R. Felici, T. Slobodskyy, T. Baumbach, and D.M. Schaadt, Phys. Stat. Sol. A 209, 2387 (2012). [82] S. Maier, K. Berschneider, T. Steinl, A. Forchel, S. Hofling, C. Schneider, and M. Kamp, Semicond. Sci. Technol. 29, 052001 (2014). [83] S. Unsleber, S. Maier, D.P.S. McCutcheon, Y.-M. He, M. Dambach, M. Gschrey, N. Gregersen, J. Mørk, S. Reitzen- stein, S. Hofling, C. Schneider, and M. Kamp, Optica 2, 1072 (2015). [84] J. Martin-Sanchez, G. Munoz-Matutano, J. Herranz, J. Canet- Ferrer, B. Alen, Y. Gonzalez, P. Alonso-Gonzalez, D. Fuster, L. Gonzalez, J. Martinez-Pastor, and F. Briones, ACS Nano 3, 1513 (2009). [85] J. Tommila , C. Strelow, A. Schramm, T.V. Hakkarainen, M. Dumitrescu, T. Kipp, and M. Guina, Nanosc. Res. Lett. 7, 313 (2012). [86] J. Canet-Ferrer, G. Munoz-Matutano, J. Herranz, D. Rivas, B. Alen, Y. Gonzalez, D. Fuster, L. Gonzalez, and J. Martinez- Pastor, Appl. Phys. Lett. 103, 183112 (2013). [87] A. Huggenberger, C.Schneider, C.Drescher, S.Heckelmann, T.Heindel, S. Reitzenstein, M. Kamp, S. Hofling, L. Worschech, A. Forchel, J. Cryst. Growth 323, 194 (2011). [88] J. Tatebayashi, M. Nishioka, T. Someya, and Y. Arakawa, Appl. Phys. Lett. 77, 3382 (2000). [89] J. Houel, A.V. Kuhlmann, L. Greuter, F. Xue, M. Poggio, B.D. Gerardot, P.A. Dalgarno, A. Badolato, P.M. Petroff, A. Ludwig, D. Reuter, A. D. Wieck, and R.J. Warburton, Phys. Rev. Lett. 108, 107401 (2012). [90] K.D. Jons, P. Atkinson, M. Muller, M. Heldmaier, S.M. Ul- rich, O.G. Schmidt, and P. Michler, Nano Lett. 13, 126 (2013). [91] C. Schneider, A. Huggenberger, M. Gschrey, P. Gold, S. Rodt, A. Forchel, S. Reitzenstein, S. Hofling, and M. Kamp, Phys. Stat. Sol. A 209, 2379 (2012). [92] A. Mohan, P. Gallo, M. Felici, B. Dwir, A. Rudra, J. Faist, and E. Kapon, small 6, 1268 (2010). [93] C. Schneider, M. Strauss, T. Sunner, A. Huggenberger, D. Wiener, S. Reitzenstein, M. Kamp, S. Hofling, and A. Forchel, Appl. Phys. Lett. 92, 183101 (2008). [94] Y.H. Huo, B.J. Witek, S. Kumar, J.R. Cardenas, J.X. Zhang, N. Akopian, R. Singh, E. Zallo, R. Grifone, D. Kriegner, R. Trotta, F. Ding, J. Stangl, V. Zwiller, G. Bester, A. Rastelli, and O.G. Schmidt, Nat. Phys. 10, 46 (2014). [95] M. Abbarchi, T. Kuroda, T. Mano, K. Sakoda, C.A. Mastran- drea, A. Vinattieri, M. Gurioli, and T. Tsuchiya, Phys. Rev. B 82, 201301(R) (2010). [96] K. Roszak, V.M. Axt, T. Kuhn, and P. Machnikowski, Phys. Rev. B 76, 195324 (2007). [97] J. Johansen, B. Julsgaard, S. Stobbe, J.M. Hvam, and P. Lo- dahl, Phys. Rev. B 81, 081304(R) (2010). [98] R. Seguin, A. Schliwa, S. Rodt, K. Potschke, U.W. Pohl, and D. Bimberg, Phys. Rev. Lett. 95, 257402 (2005). [99] A.J. Bennett, M.A. Pooley, R.M. Stevenson, M.B. Ward, R.B. Patel, A. Boyer de la Giroday, N. Skold, I. Farrer, C.A. Nicoll, D.A. Ritchie, and A.J. Shields, Nat. Phys. 6, 947 (2010). [100] M. Bayer, G. Ortner, O. Stern, A. Kuther, A.A. Gorbunov, A. Forchel, P. Hawrylak, S. Fafard, K. Hinzer, T.L. Reinecke, S.N. Walck, J.P. Reithmaier, F. Klopf, and F. Schafer, Phys. Rev. B 65, 195315 (2002). [101] E. Poem, Y. Kodriano, C. Tradonsky, N.H. Lindner, B.D. Gerardot, P.M. Petroff, and D. Gershoni, Nat. Phys. 6, 993 (2010). [102] O. Benson, C. Santori, M. Pelton, and Y. Yamamoto, Phys. Rev. Lett. 84, 2513 (2000). [103] R.M. Stevenson, R.J. Young, P. Atkinson, K. Cooper, D.A. Ritchie, and A.J. Shields, Nature 439, 179 (2006). [104] G. Juska, V. Dimastrodonato, L.O. Mereni, A. Gocalinska, and E. Pelucchi, Nat. Photon. 7, 527 (2013). [105] T. Kuroda, T. Mano, N. Ha, H. Nakajima, H. Kumano, B. Urbaszek, M. Jo, M. Abbarchi, Y. Sakuma, K. Sakoda, I. Sue- mune, X. Marie, and T. Amand, Phys. Rev. B 88, 041306(R) (2013). Copyright line will be provided by the publisher [106] M. Larqu´e, I. Robert-Philip, and A. Beveratos, Phys. Rev. A 77, 042118 (2008). [107] R.J. Young, R.M. Stevenson, A.J. Shields, P. Atkinson, K. Cooper, D.A. Ritchie, K.M. Groom, A.I. Tartakovskii, and M.S. Skolnick, Phys. Rev. B 72, 113305 (2005). [108] K. Kowalik, O. Krebs, A. Lemaıtre, S. Laurent, P. Senellart, P. Voisin and J.A. Gaj, Appl. Phys. Lett. 86, 041907 (2005). [109] R. Trotta, E. Zallo, C. Ortix, P. Atkinson, J.D. Plumhof, J. van den Brink, A. Rastelli, and O. G. Schmidt, Phys. Rev. Lett. 109, 147401 (2012). [110] A.J. Shields, Nat. Photon. 1, 215 (2007). [111] P. Borri, W. Langbein, S. Schneider, U. Woggon, R.L. Sellin, D. Ouyang, and D. Bimberg, Phys. Rev. Lett. 87, 157401 (2001). [112] E.A. Muljarov and R. Zimmermann, Phys. Rev. Lett. 93, 237401 (2004). [113] A.J. Ramsay, A.V. Gopal, E.M. Gauger, A. Nazir, B.W. Lovett, A.M. Fox, and M. S. Skolnick, Phys. Rev. Lett. 104, 017402 (2010). [114] K.H. Madsen, P. Kaer, A. Kreiner-Møller, S. Stobbe, A. Nysteen, J. Mørk, and P. Lodahl, Phys. Rev. B 88, 045316 (2013). [115] P. Kaer, N. Gregersen, and J. Mørk, New J. Phys. 15, 035027 (2013). [116] P. Kaer and J. Mørk, Phys. Rev. B 90, 035312 (2014). [117] S. Ates, S.M. Ulrich, S. Reitzenstein, A. Loffler, A. Forchel, and P. Michler, Phys. Rev. Lett. 103, 167402 (2009). [118] E.B. Flagg, S.V. Polyakov, T. Thomay, and G.S. Solomon, Phys. Rev. Lett. 109, 163601 (2012). [119] O. Gazzano , S. Michaelis de Vasconcellos, C. Arnold, A. Nowak, E. Galopin, I. Sagnes, L. Lanco, A. Lemaıtre, and P. Senellart, Nat. Commun. 4, 1425 (2013). [120] T. Huber, A. Predojevi´c, D. Foger, G. Solomon, and G. Weihs, New J. Phys. 17, 123025 (2015). [121] S. Unsleber, D.P.S. McCutcheon, M. Dambach, M. Lermer, N. Gregersen, S. Hofling, J. Mørk, C. Schneider, and M. Kamp, Phys. Rev. B 91, 075413 (2015). [122] A.V. Kuhlmann, J. Houel, A. Ludwig, L. Greuter, D. Reuter, A.D. Wieck, M. Poggio, and R.J. Warburton, Nat. Phys. 9, 570 (2013). [123] K. Konthasinghe, J. Walker, M. Peiris, C.K. Shih, Y. Yu, M.F. Li, J.F. He, L.J. Wang, H.Q. Ni, Z.C. Niu, and A. Muller, Phys. Rev. B 85, 235315 (2012). [124] Y.-M. He, Y. He, Y.-J. Wei, D. Wu, M. Atature, C. Schnei- der, S. Hofling, M. Kamp, C.-Y. Lu, and J.-W. Pan, Nat. Nanotechnol. 8, 213 (2013). [125] W.B. Gao, P. Fallahi, E. Togan, A. Delteil, Y.S. Chin, J. Miguel-Sanchez, and A. Imamoglu, Nat. Commun. 4, 2744 (2013). [126] P. Gold, A. Thoma, S. Maier, S. Reitzenstein, C. Schneider, S. Hofling, and M. Kamp, Phys. Rev. B 89, 035313 (2014). [127] S. Rodt, A. Schliwa, K. Potschke, F. Guffarth, and D. Bim- berg, Phys. Rev. B 71, 155325 (2005). [128] X. Ding, Y. He, Z.-C. Duan, N. Gregersen, M.-C. Chen, S. Unsleber, S. Maier, C. Schneider, M. Kamp, S Hofling, C.-Y. Lu, and J.-W. Pan, Phys. Rev. Lett. 116 020401 (2016). [129] N. Somaschi, V. Giesz, L. De Santis, J.C. Loredo, M.P. Almeida, G. Hornecker, S.L. Portalupi, T. Grange, C. Ant´on, J. Demory, C. G´omez, I. Sagnes, N.D. Lanzillotti-Kimura, A. Lema´ıtre, A. Auffeves, A.G. White, L. Lanco, and P. Senel- lart, Nat. Photon. 10, 340 (2016). 21 [130] Y.-J. Wei, Y.-M. He, M.-C. Chen, Y.-N. Hu, Y. He, D. Wu, C. Schneider, M. Kamp, S. Hofling, C.-Y. Lu, and J.-W. Pan, Nano Lett. 14, 6515 (2014). [131] D. Birkedal, K. Leosson, and J.M. Hvam, Phys. Rev. Lett. 87, 227401 (2001). [132] A. Muller, E.B. Flagg, P. Bianucci, X.Y. Wang, D.G. Deppe, W. Ma, J. Zhang, G.J. Salamo, M. Xiao, and C.K. Shih, Phys. Rev. Lett. 99, 187402 (2007). [133] E.B. Flagg, A. Muller, J.W. Robertson, S. Founta, D.G. Deppe, M. Xiao, W. Ma, G.J. Salamo, and C.K. Shih, Nat. Phys. 5, 203 (2009). [134] A.N. Vamivakas, Y. Zhao, C.-Y. Lu, and M. Atature, Nat. Phys. 5, 198 (2009). [135] A. Moelbjerg, P. Kaer, M. Lorke, and J. Mørk, Phys. Rev. Lett. 108, 017401 (2012). [136] S.M. Ulrich, S. Ates, S. Reitzenstein, A. Loffler, A. Forchel, and P. Michler, Phys. Rev. Lett. 106, 247402 (2011). [137] Y. He, Y.-M. He, Y.-J. Wei, X. Jiang, M.-C. Chen, F.-L. Xiong, Y. Zhao, C. Schneider, M. Kamp, S. Hofling, C.-Y. Lu, and J.-W. Pan, Phys. Rev. Lett. 111, 237403 (2013). [138] M. Peiris, K. Konthasinghe, Y. Yu, Z.C. Niu, and A. Muller, Phys. Rev. B 89, 155305 (2014). [139] C.-Y. Lu, Y. Zhao, A.N. Vamivakas, C. Matthiesen, S. Falt, A. Badolato, and M. Atature, Phys. Rev.B 81, 035332 (2010). [140] X.D. Xu, B. Sun, P.R. Berman, D.G. Steel, A.S. Bracker, D. Gammon, and L.J. Sham, Science 317, 929 (2007). [141] B.R. Mollow, Phys. Rev. 188, 1969 (1969). [142] Y. He, Y.-M. He, J. Liu, Y.-J. Wei, H. Y. Ramrez, M. Atature, C. Schneider, M. Kamp, S. Hufling, C.-Y. Lu, and J.-W. Pan, Phys. Rev. Lett. 114, 097402 (2015). [143] H. Kamada, H. Gotoh, J. Temmyo, T. Takagahara, and H. Ando, Phys. Rev. Lett. 87, 246401 (2001). [144] T.H. Stievater, X.Q. Li, D.G. Steel, D. Gammon, D.S. Katzer, D. Park, C. Piermarocchi, and L.J. Sham, Phys. Rev. Lett. 87, 133603 (2001). [145] C. Matthiesen, M. Geller, C.H.H. Schulte, C. Le Gall, J. Hansom, Z.Y. Li, M. Hugues, E. Clarke, and M. Atature, Nat. Comm. 4, 1600 (2013). [146] C. Matthiesen, A.N. Vamivakas, and M. Atature, Phys. Rev. Lett. 108, 093602 (2012). [147] H.S. Nguyen, G. Sallen, C. Voisin, Ph. Roussignol, C. Diederichs, and G. Cassabois, Appl. Phys. Lett. 99, 261904 (2011). [148] D.P.S. McCutcheon and A. Nazir, Phys. Rev. Lett. 110, 217401 (2013). [149] M.J. Conterio, N. Skold, D.J.P. Ellis, I. Farrer, D.A. Ritchie, and A.J. Shields, Appl. Phys. Lett. 103, 162108 (2013). [150] R. Melet, V. Voliotis, A. Enderlin, D. Roditchev, X.L. Wang, T. Guillet, and R. Grousson, Phys. Rev. B 78, 073301 (2008). [151] H. Jayakumar, A. Predojevi´c, T. Huber, T. Kauten, G.S. Solomon, and G. Weihs, Phys. Rev. Lett. 110, 135505 (2013). [152] M.N. Makhonin, J.E. Dixon, R.J. Coles, B. Royall, I.J. Lux- moore, E. Clarke, M. Hugues, M.S. Skolnick, and A.M. Fox, Nano Lett. 14, 6997 (2014). [153] G. Reithmaier, M. Kaniber, F. Flassig, S. Lichtmannecker, K. Muller, A. Andrejew, J. Vuckovi´c, R. Gross, and J.J. Finley, Nano Lett. 15, 5208 (2015). [154] A. Faraon, I. Fushman, D. Englund, N. Stoltz, P. Petroff, and J. Vuckovi´c,, Nat. Phys. 4, 859 (2008). [155] A. Reinhard, T. Volz, M. Winger, A. Badolato, K. J. Hen- nessy, E. L. Hu, and A. Imamoglu, Nat. Photon. 6, 93 (2012). Copyright line will be provided by the publisher 22 C.P. Dietrich et al.: GaAs integrated quantum photonics [156] K. Konthasinghe, M. Peiris, Y. Yu, M.F. Li, J.F. He, L.J. Wang, H.Q. Ni, Z.C. Niu, C.K. Shih, and A. Muller, Phys. Rev. Lett. 109, 267402 (2012). [157] J. Bylander, I. Robert-Philip, I. Abram, Europ. Phys. Journ. [158] C.K. Hong, Z.Y. Ou, and L. Mandel, Phys. Rev. Lett. 59, D 22, 295 (2003). 2044 (1987). [159] R.B. Patel, A.J. Bennett, K. Cooper, P. Atkinson, C.A. Nicoll, D.A. Ritchie, and A.J. Shields, Phys. Rev. Lett. 100, 207405 (2008). [160] A.J. Bennett, R.B. Patel, C.A. Nicoll, D.A. Ritchie, and A.J. Shields, Nat. Phys. 5, 715 (2009). [161] S.V. Polyakov, A. Muller, E.B. Flagg, A. Ling, N. Borjem- scaia, E. Van Keuren, A. Migdall, and G.S. Solomon, Phys. Rev. Lett. 107, 157402 (2011). [162] K. Konthasinghe, M. Peiris, and A. Muller, Phys. Rev. A 90, 023810 (2014). [163] S. Ates, I. Agha, A. Gulinatti, I. Rech, M.T. Rakher, A. Badolato, and K. Srinivasan, Phys. Rev. Lett. 109, 147405 (2012). [164] E. Purcell, Phys. Rev. 69, 681 (1946). [165] D. Kleppner, Phys. Rev. Lett. 47, 233 (1981). [166] K.C. Balram, M. Davanc¸o, J.Y. Lim, J.D. Song, and K. Srinivasan, Optica 1, 414 (2014). [167] I.J. Luxmoore, R. Toro, O. Del Pozo-Zamudio, N.A. Wasley, E.A. Chekhovich, A.M. Sanchez, R. Beanland, A.M. Fox, M.S. Skolnick, H.Y. Liu, and A.I. Tartakovskii, Sci. Rep. 3, 1239 (2012). [168] A. Tandaechanurat, S. Ishida, D. Guimard, M. Nomura, S. Iwamoto, Y. Arakawa, Nat. Photon. 5, 91 (2011). [169] K. Srinivasan and O. Painter, Nature 450, 862 (2007). [170] C. Schneider, P. Gold, S. Reitzenstein, S. Hofling, and M. Kamp, arXiv:1510.05447 (2015). [171] M. Lermer, N. Gregersen, F. Dunzer, S. Reitzenstein, S. Hofling, J. Mørk, L. Worschech, M. Kamp, and A. Forchel, Phys. Rev. Lett. 108, 057402 (2012). [172] J.D. Joannopoulos, S.G. Johnson, J.N. Winn, and R.D. Meade, Photonic Crystals: Molding the Flow of Light, 2nd Ed. (Princeton University Press, 2008). [173] K. Nozaki, T. Tanabe, A. Shinya, S. Matsuo, T. Sato, H. Taniyama, and M. Notomi, Nat. Photon. 4, 477 (2010). [174] D. Englund, D. Fattal, E. Waks, G. Solomon, B.Y. Zhang, T. Nakaoka, Y. Arakawa, Y. Yamamoto, and J. Vuckovi´c, Phys. Rev. Lett. 95, 013904 (2005). [175] M. Shirane, S. Kono, J. Ushida, S. Ohkouchi, N. Ikeda, Y. Sugimoto, and A. Tomita, J. Appl. Phys. 101, 073107 (2007). [176] W.J. Fan, Z.B. Hao, E. Stock, J.B. Kang, Y. Luo, and D. Bimberg, Semicond. Sci. Technol. 26, 014014 (2011). [177] T.W. Saucer and V. Sih, Opt. Expr. 21, 20831 (2013). [178] Y. Akahane, T. Asano, B.-S. Song, and S. Noda, Nature 425, 944 (2003). 13, 1202 (2005). Lett. 33, 1908 (2008). [179] Y. Akahane, T. Asano, B.-S. Song, and S. Noda, Opt. Expr. [180] S. Combri´e, A. De Rossi, Q.V. Tran, and H. Benisty, Opt. [181] M. Minkov and V. Savona, Sci. Rep. 4, 5124 (2014). [182] H. Takagi, Y. Ota, N. Kumagai, S. Ishida, S. Iwamoto, and Y. Arakawa, Opt. Expr. 20, 28292 (2012). [183] A. Enderlin, Y. Ota, R. Ohta, N. Kumagai, S. Ishida, S. Iwamoto, and Y. Arakawa, Phys. Rev. B 86, 075314 (2012). [184] E. Viasnoff-Schwoob, C. Weisbuch, H. Benisty, S. Olivier, S. Varoutsis, I. Robert-Philip, R. Houdre, and C.J.M. Smith, Phys. Rev. Lett. 95, 183901 (2005). [185] G. Lecamp, P. Lalanne, and J.P. Hugonin, Phys. Rev. Lett. 99, 023902 (2007). (2007). 193901 (2007). [186] V.S.C. Manga Rao and S. Hughes, Phys. Rev. B 75, 205437 [187] V.S.C. Manga Rao and S. Hughes, Phys. Rev. Lett. 99, [188] S.J. Dewhurst, D. Granados, D.J. Ellis, A.J. Bennett, R.B. Patel, I. Farrer, D. Anderson, G.A. Jones, D.A. Ritchie, and A.J. Shields, Appl. Phys. Lett. 96, 031109 (2010). [189] D.G. Gevaux, A.J. Bennett, R.M. Stevenson, A.J. Shields, P. Atkinson, J. Griffiths, D. Anderson, G.A.C. Jones, and D.A. Ritchie, Appl. Phys. Lett. 88, 131101 (2006). [190] A. Faraon, D. Englund, I. Fushman, J. Vuckovi´c, N. Stoltz, and P. Petroff, Appl. Phys. Lett. 90, 213110 (2007). [191] A. Faraon and J. Vuckovi´c, Appl. Phys. Lett. 95, 043102 (2009). [192] H.C. Kim, T.C. Shen, D. Sridharan, G.S. Solomon, and E. Waks, Appl. Phys. Lett. 98, 091102 (2011). [193] J. Beetz, T. Braun, C. Schneider, S. Hofling, and M. Kamp, Semicond. Sci. Technol. 28, 122002 (2013). [194] S. Sun, H.C. Kim, G.S. Solomon, and E. Waks, Appl. Phys. Lett. 103, 151102 (2013). [195] P.W. Fry, I.E. Itskevich, D.J. Mowbray, M.S. Skolnick, J.J. Finley, J.A. Barker, E.P. O'Reilly, L.R. Wilson, I.A. Larkin, P.A. Maksym, M. Hopkinson, M. Al-Khafaji, J.P.R. David, A.G. Cullis, G. Hill, and J.C. Clark, Phys. Rev. Lett. 84, 733 (2000). [196] F. Hofbauer, S. Grimminger, J. Angele, G. Bohm, R. Meyer, M.-C. Amann, and J.J. Finley, Appl. Phys. Lett. 91, 201111 (2007). [197] N. Chauvin, C. Zinoni, M. Francardi, A. Gerardino, L. Balet, B. Alloing, L.H. Li, and A. Fiore, Phys. Rev. B 80, 241306 (2009). [198] A. Laucht, F. Hofbauer, N. Hauke, J. Angele, S. Stobbe, M. Kaniber, G. Bohm, P. Lodahl, M.-C. Amann, and J.J. Finley, New J. Phys. 11, 023034 (2009). [199] S.M. Thon, H. Kim, C. Bonato, J. Gudat, J. Hagemeier, P.M. Petroff, and D. Bouwmeester, Appl. Phys. Lett. 99, 161102 (2011). [200] F. Pagliano, Y.J. Cho, T. Xia, F. van Otten, R. Johne, and A. Fiore, Nat. Commun. 5, 5786 (2014). [201] M. Petruzzella, T. Xia, F. Pagliano, S. Birindelli, L. Midolo, Z. Zobenica, L.H. Li, E.H. Linfield, and A. Fiore, Appl. Phys. Lett. 107, 141109 (2015). [202] A. Faraon, D. Englund, D. Bulla, B. Luther-Davies, B.J. Eggleton, N. Stoltz, P. Petroff, and J. Vuckovi´c, Appl. Phys. Lett. 92, 043123 (2008). [203] K. Hennessy, A. Badolato, A. Tamboli, P.M. Petroff, E. Hu, M. Atature, J. Dreiser, and A. Imamoglu, Appl. Phys. Lett. 87, 021108 (2005). [204] K. Hennessy, C. Hogerle, E. Hu, A. Badolato, and A. Imamoglu, Appl. Phys. Lett. 89, 041118 (2006). [205] H.S. Lee, S. Kiravittaya, S. Kumar, J.D. Plumhof, L. Balet, L.H. Li, M. Francardi, A. Gerardino, A. Fiore, A. Rastelli, and O.G. Schmidt, Appl. Phys. Lett. 95, 191109 (2009). [206] F. Intonti, N. Caselli, S. Vignolini, F. Riboli, S. Kumar, A. Rastelli, O.G. Schmidt, M. Francardi, A. Gerardino, L. Balet, L.H. Li, A. Fiore, and M. Gurioli, Appl. Phys. Lett. 100, 033116 (2012). [207] A.Y. Piggott, K.G. Lagoudakis, T. Sarmiento, M. Bajcsy, G. Shambat, and J. Vuckovi´c, Opt. Expr. 22, 15017 (2014). Copyright line will be provided by the publisher [208] S. Mosor, J. Hendrickson, B.C. Richards, J. Sweet, G. Khitrova, H.M. Gibbs, T. Yoshie, A. Scherer, O.B. Shchekin, and D.G. Deppe, Appl. Phys. Lett. 87, 141105 (2005). [209] F. Intonti, S. Vignolini, F. Riboli, M. Zani, D.S. Wiersma, L. Balet, L.H. Li, M. Francardi, A. Gerardino, A. Fiore, and M. Gurioli, Appl. Phys. Lett. 95, 173112 (2009). [210] S. Vignolini, F. Riboli, D.S. Wiersma, L. Balet, L.H. Li, M. Francardi, A. Gerardino, A. Fiore, M. Gurioli, and F. Intonti, Appl. Phys. Lett. 96, 141114 (2010). [211] N.W.L. Speijcken, M.A. Dundar, A.C. Bedoya, C. Monat, C. Grillet, P. Domachuk, R. Notzel, B.J. Eggleton, and R.W. van der Heijden, Appl. Phys. Lett. 100, 261107 (2012). [212] A.F. Koenderink, M. Kafesaki, B.C. Buchler, and V. San- doghdar, Phys. Rev. Lett. 95, 153904 (2005). [213] S. Vignolini, F. Intonti, L. Balet, M. Zani, F. Riboli, A. Vinattieri, D.S. Wiersma, M. Colocci, L.H. Li, M. Francardi, A. Gerardino, A. Fiore, and M. Gurioli, Appl. Phys. Lett. 93, 023124 (2008). [214] M. Notomi, H. Taniyama, S. Mitsugi, and E. Kuramochi, Phys. Rev. Lett. 97, 023903 (2006). [215] L. Midolo, P.J. van Veldhoven, M.A. Dundar, R. Notzel, and A. Fiore, Appl. Phys. Lett. 98, 211120 (2011). [216] L. Midolo, F. Pagliano, T.B. Hoang, T. Xia, F.W.M. van Otten, L.H. Li, E. Linfield, M. Lermer, S. Hofling, and A. Fiore, Appl. Phys. Lett. 101, 091106 (2012). [217] Y. Sugimoto, Y. Tanaka, N. Ikeda, Y. Nakamura, and K. Asakawa, Opt. Expr. 12, 1090 (2004). [218] S. Fattah poor, T.B. Hoang, L. Midolo, C.P. Dietrich, L.H. Li, E.H. Linfield, J.F.P. Schouwenberg, T. Xia, F.M. Pagliano, F.W.M. Otten, and A. Fiore, Appl. Phys. Lett. 102, 131105 (2013). [219] T.H. Stievater, D. Park, W.S. Rabinovich, M.W. Pruessner, S. Kanakaraju, C.J.K. Richardson, and J.B. Khurgin, Opt. Expr. 18, 885 (2010). [220] J. Tatebayashi, R.B. Laghumavarapu, N. Nuntawong, and D.L. Huffaker, Electr. Lett. 43, 410 (2007). [221] J.H. Shin, Y.-C. Chang, and N. Dagli, Appl. Phys. Lett. 92, 201103 (2008). [222] M.M. de Lima Jr., M. Beck, R. Hey, and P.V. Santos, Appl. Phys. Lett. 89, 121104 (2006). [223] M. Poot and H.X. Tang, Appl. Phys. Lett. 104, 061101 (2014). [224] K.D. Jons, U. Rengstl, M. Oster, F. Hargart, M. Heldmaier, S. Bounouar, S.M. Ulrich, M. Jetter, and P. Michler, J. Phys. D: Appl. Phys. 48, 085101 (2015). [225] N. Prtljaga, R.J. Coles, J. O'Hara, B. Royall, E. Clarke, A.M. Fox, and M.S. Skolnick, Appl. Phys. Lett. 104, 231107 (2014). [226] U. Rengstl, M. Schwartz, T. Herzog, F. Hargart, M. Paul, S.L. Portalupi, M. Jetter, and P. Michler, Appl. Phys. Lett. 107, 021101 (2015). [227] J. Kasprzak, S. Reitzenstein, E.A. Muljarov, C. Kistner, C. Schneider, M. Strauss, S. Hofling, A. Forchel, and W. Langbein, Nat. Mater. 9, 304 (2010). [228] V. Loo, C. Arnold, O. Gazzano, A. Lema´ıtre, I. Sagnes, O. Krebs, P. Voisin, P. Senellart, and L. Lanco, Phys. Rev. Lett. 109, 166806 (2012). [229] T. Volz, A. Reinhard, M. Winger, A. Badolato, K.J. Hen- nessy, E.L. Hu, and A. Imamoglu, Nat. Photon. 6, 609 (2012). [230] D. Englund, A. Majumdar, M. Bajcsy, A. Faraon, P. Petroff, and J. Vuckovi´c, Phys. Rev. Lett. 108, 093604 (2012). [231] R. Bose, D. Sridharan, H. Kim, G.S. Solomon, and E. Waks, Phys. Rev. Lett. 108, 227402 (2012). 23 [232] R.J. Warburton, Nat. Mater. 12, 483 (2013). [233] J.R. Petta, A.C. Johnson, J.M. Taylor, E.A. Laird, A. Ya- coby, M.D. Lukin, C.M. Marcus, M.P. Hanson, A.C. Gossard, Science 309, 2180 (2005). [234] J. Berezovsky, M.H. Mikkelsen, N.G. Stoltz, L.A. Coldren, D.D. Awschalom, Science 320, 349 (2008). [235] M. Atature, J. Dreiser, A. Badolato, and A. Imamoglu, Nat. Phys. 3, 101 (2007). [236] I.J. Luxmoore, N.A. Wasley, A.J. Ramsay, A.C.T. Thijssen, R. Oulton, M. Hugues, S. Kasture, V.G. Achanta, A.M. Fox, and M.S. Skolnick, Phys. Rev. Lett. 110, 037402 (2013). [237] I.J. Luxmoore, N.A. Wasley, A.J. Ramsay, A.C.T. Thijssen, R. Oulton, M. Hugues, A.M. Fox, and M.S. Skolnick, Appl. Phys. Lett. 103, 241102 (2013). [238] A.B. Young, A.C.T. Thijssen, D.M. Beggs, P. Androvit- saneas, L. Kuipers, J.G. Rarity, S. Hughes, and R. Oulton, Phys. Rev. Lett. 115, 153901 (2015). [239] I. Sollner, S. Mahmoodian, S. Lindskov Hansen, L. Midolo, A. Javadi, G. Kirsanske, T. Pregnolato, H. El-Ella, E.H. Lee, J.D. Song, S. Stobbe, and P. Lodahl, Nat. Nanotechnol. 10, 775 (2015). [240] G.N. Gol'tsman, O. Okunev, G. Chulkova, A. Lipatov, A. Se- menov, K. Smirnov, B. Voronov, A. Dzardanov, C. Williams, and R. Sobolewski, Appl. Phys. Lett. 79, 705 (2001). [241] R.H. Hadfield, Nat. Photon. 3, 696 (2009). [242] C.M. Natarajan, M.G. Tanner, and R.H. Hadfield, Super- cond. Sci. Technol. 25, 063001 (2012). [243] A. Gaggero, S. Jahanmiri Nejad, F. Marsili, F. Mattioli, R. Leoni, D. Bitauld, D. Sahin, G.J. Hamhuis, R. Notzel, R. Sanjines, and A. Fiore, Appl. Phys. Lett. 97, 151108 (2010). [244] J.J. Renema, R. Gaudio, Q. Wang, Z. Zhou, A. Gaggero, F. Mattioli, R. Leoni, D. Sahin, M.J.A. de Dood, A. Fiore, and M.P. van Exter, Phys. Rev. Lett. 112, 117604 (2014). [245] J.J. Renema, Q. Wang, R. Gaudio, I. Komen, K. op 't Hoog, D. Sahin, A. Schilling, M.P. van Exter, A. Fiore, A. Engel, and M.J.A. de Dood, Nano Lett. 15, 4541 (2015). [246] A.J. Kerman, E.A. Dauler, W.E. Keicher, J.K.W. Yang, K.K. Berggren, G.N. Gol'tsman, and B. Voronov, Appl. Phys. Lett. 88, 111116 (2006). [247] F. Marsili, F. Najafi, E. Dauler, F. Bellei, X.L. Hu, M. Csete, R.J. Molnar, and K.K. Berggren, Nano Lett. 11, 2048 (2011). [248] R. Gaudio, K.P.M. op 't Hoog, Z. Zhou, D. Sahin, and A. Fiore, Appl. Phys. Lett. 105, 222602 (2014). [249] B. Calkins, P.L. Mennea, A.E. Lita, B.J. Metcalf, W.S. Kolthammer, A. Lamas-Linares, J.B. Spring, P.C. Humphreys, R.P. Mirin, J.C. Gates, P.G.R. Smith, I.A. Walm- sley, T. Gerrits, and S.W. Nam, Opt. Expr. 21, 22657 (2013). [250] S. Ferrari, O. Kahl, V. Kovalyuk, G.N. Gol'tsman, A. Ko- rneev, and W.H.P. Pernice, Appl. Phys. Lett. 106, 151101 (2015). [251] O. Kahl, S. Ferrari, V. Kovalyuk, G.N. Gol'tsman, A. Ko- rneev, and W.H.P. Pernice, Sci. Rep. 5, 10941 (2015). [252] F. Najafi, J. Mower, N.C. Harris, F. Bellei, A. Dane, C. Lee, X.L. Hu, P. Kharel, F. Marsili, S. Assefa, K.K. Berggren, and D. Englund, Nat. Commun. 6, 5873 (2015). [253] B. Baek, A.E. Lita, V. Verma, and S.W. Nam, Appl. Phys. Lett. 98, 251105 (2011). [254] F. Marsili, V.B. Verma, J.A. Stern, S. Harrington, A.E. Lita, T. Gerrits, I. Vayshenker, B. Baek, M.D. Shaw, R.P. Mirin, and S.W. Nam, Nat. Photon. 7, 210 (2013). [255] D. Sahin, A. Gaggero, J.-W. Weber, I. Agafonov, M.A. Ver- heijen, F. Mattioli, J. Beetz, M. Kamp, S. Hofling, M.C.M Copyright line will be provided by the publisher 24 C.P. Dietrich et al.: GaAs integrated quantum photonics van de Sanden, R. Leoni, and A. Fiore, IEEE J. Sel. Topics Quantum. Electron. 21, 3800210 (2015). [256] A.J. Kerman, E.A. Dauler, J.K.W. Yang, K.M. Rosfjord, V. Anant, K.K. Berggren, G.N. Gol'tsman, and B. Voronov, Appl. Phys. Lett. 90, 101110 (2007). [257] M.K. Akhlaghi, E. Schelew, J.F. Young, Nat. Commun. 6, 8233 (2015). [258] D. Sahin, A. Gaggero, T.B. Hoang, G. Frucci, F. Mattioli, R. Leoni, J. Beetz, M. Lermer, M. Kamp, S. Hofling, and A. Fiore, Opt. Expr. 21, 11162 (2013). [259] E.A. Dauler, A.J. Kerman, B.S. Robinson, J.K.W. Yang, G.G.B. Voronovc, S.A. Hamilton, and K.K. Berggren, J. Mod. Opt. 56, 364 (2009). [260] G. Reithmaier, S. Lichtmannecker, T. Reichert, P. Hasch, K. Muller, M. Bichler, R. Gross, and J.J. Finley, Sci. Rep. 3, 1901 (2013). [261] A. Divochiy, F. Marsili, D. Bitauld, A. Gaggero, R. Leoni, F. Mattioli, A. Korneev, V. Seleznev, N. Kaurova, O. Minaeva, G.N. Gol'tsman, K.G. Lagoudakis, M. Benkhaoul, F. Levy, and A. Fiore, Nat. Photon. 2, 302 (2008). [262] S. Jahanmirinejad and A. Fiore, Opt. Expr. 20, 5017 (2012). [263] S. Jahanmirinejad, G. Frucci, F. Mattioli, D. Sahin, A. Gag- gero, R. Leoni, and A. Fiore, Appl. Phys. Lett. 101, 072602 (2012). [264] Z. Zhou, S. Jahanmirinejad, F. Mattioli, D. Sahin, G. Frucci, A. Gaggero, R. Leoni, and A. Fiore, Opt. Expr. 22, 3475 (2014). [265] D. Sahin, A. Gaggero, Z. Zhou, S. Jahanmirinejad, F. Matti- oli, R. Leoni, J. Beetz, M. Lermer, M. Kamp, S. Hofling, and A. Fiore, Appl. Phys. Lett. 103, 111116 (2013). [266] P.J. Shadbolt, M.R. Verde, A. Peruzzo, A. Politi, A. Laing, M. Lobino, J.C.F. Matthews, M.G. Thompson, and J.L. O'Brien, Nat. Photon. 6, 45 (2012). [267] C. Zinoni, B. Alloing, L.H. Li, F. Marsili, A. Fiore, L. Lunghi, A. Gerardino, Yu.B. Vakhtomin, K.V. Smirnov, G.N. Gol'tsman, Appl. Phys. Lett. 91, 031106 (2007). [268] K. Takemoto, Y. Nambu, T. Miyazawa, K. Wakui, S. Hirose, T. Usuki, M. Takatsu, N. Yokoyama, K. Yoshino, A. Tomita, S. Yorozu, Y. Sakuma, and Y. Arakawa, Appl. Phys. Expr. 3, 092802 (2010). [269] H. Takesue, N. Matsuda, E. Kuramochi, W.J. Munro, and M. Notomi, Nat. Commun. 4, 2725 (2013). [270] P. Atkinson, O.G. Schmidt, S.P. Bremmer, and D.A. Ritchie, Compt. Rend. Phys. 9, 788 (2008). [271] C. Schneider, A. Huggenberger, T. Suunner, T. Heindel, M. Strauss, S Gopfert, P. Weinmann, S. Reitzenstein, L. Worschech, M. Kamp, S. Hofling, and A. Forchel, Nanotech- nology 20, 434012 (2009). [272] E. Pelucchi, S. Watanabe, K. Leifer, Q. Zhu, B. Dwir, P. De Los Rios, and E. Kapon, Nano Lett. 7, 1282 (2007). [273] J.H. Prechtel, A.V. Kuhlmann, J. Houel, L. Greuter, A. Lud- wig, D. Reuter, A.D. Wieck, and R.J. Warburton, Phys. Rev. X 3, 041006 (2013). [274] M. Alexe, V. Dragoi, M. Reiche, and U. Gosele, Electr. Lett. [275] S. Buckley, K. Rivoire, and J. Vuckovi´c, Rep. Prog. Phys. 36, 677 (2000). 75, 126503 (2012). [276] J.W. Silverstone, D. Bonneau, K. Ohira, N. Suzuki, H. Yoshida, N. Iizuka, M. Ezaki, C.M. Natarajan, M.G. Tanner, R.H. Hadfield, V. Zwiller, G.D. Marshall, J.G. Rarity, J.L. O'Brien, and M.G. Thompson, Nat. Photon. 8, 104 (2014). [277] A.L. Migdall, D. Branning, and S. Castelletto, Phys. Rev. A 66, 053805 (2002). [278] R.J.A. Francis-Jones and P.J. Mosley, arXiv:1503.06178 [279] H. Wada, H. Sasaki and T. Kamijoh, Solid State Electron. (2015). 43, 1655 (1999). Copyright line will be provided by the publisher 

1503.07854

2

1503

Thermophoresis of an Antiferromagnetic Soliton

We study dynamics of an antiferromagnetic soliton under a temperature gradient. To this end, we start by phenomenologically constructing the stochastic Landau-Lifshitz-Gilbert equation for an antiferromagnet with the aid of the fluctuation-dissipation theorem. We then derive the Langevin equation for the soliton's center of mass by the collective coordinate approach. An antiferromagentic soliton behaves as a classical massive particle immersed in a viscous medium. By considering a thermodynamic ensemble of solitons, we obtain the Fokker-Planck equation, from which we extract the average drift velocity of a soliton. The diffusion coefficient is inversely proportional to a small damping constant $\alpha$, which can yield a drift velocity of tens of m/s under a temperature gradient of $1$ K/mm for a domain wall in an easy-axis antiferromagnetic wire with $\alpha \sim 10^{-4}$.

cond-mat.mes-hall

cond-mat

Brownian thermophoresis of an antiferromagnetic soliton Se Kwon Kim,1 Oleg Tchernyshyov,2 and Yaroslav Tserkovnyak1 1Department of Physics and Astronomy, University of California, Los Angeles, California 90095, USA 2Department of Physics and Astronomy, The Johns Hopkins University, Baltimore, Maryland 21218, USA (Dated: June 17, 2021) We study dynamics of an antiferromagnetic soliton under a temperature gradient. To this end, we start by phenomenologically constructing the stochastic Landau-Lifshitz-Gilbert equation for an antiferromagnet with the aid of the fluctuation-dissipation theorem. We then derive the Langevin equation for the soliton's center of mass by the collective coordinate approach. An antiferromagentic soliton behaves as a classical massive particle immersed in a viscous medium. By considering a thermodynamic ensemble of solitons, we obtain the Fokker-Planck equation, from which we extract the average drift velocity of a soliton. The diffusion coefficient is inversely proportional to a small damping constant α, which can yield a drift velocity of tens of m/s under a temperature gradient of 1 K/mm for a domain wall in an easy-axis antiferromagnetic wire with α ∼ 10−4. PACS numbers: 75.78.-n, 66.30.Lw, 75.10.Hk Introduction. -- Ordered magnetic materials exhibit solitons and defects that are stable for topological rea- sons [1]. Well-known examples are a domain wall (DW) in an easy-axis magnet or a vortex in a thin film. Their dynamics have been extensively studied because of fun- damental interest as well as practical considerations such as the racetrack memory [2]. A ferromagnetic (FM) soli- ton can be driven by various means, e.g., an external magnetic field [3] or a spin-polarized electric current [4]. Recently, the motion of an FM soliton under a temper- ature gradient has attracted a lot of attention owing to its applicability in an FM insulator [5 -- 8]. A temperature gradient of 20 K/mm has been demonstrated to drive a DW at a velocity of 200 µm/s in an yttrium iron garnet film [9]. An antiferromagnet (AFM) is of a great current inter- est in the field of spintronics [10 -- 12] due to a few advan- tages over an FM. First, the characteristic frequency of an AFM is several orders higher than that of a typical FM, e.g., a timescale of optical magnetization switching is an order of ps for AFM NiO [13] and ns for FM CrO2 [14], which can be exploited to develop faster spintronic devices. Second, absence of net magnetization renders the interaction between AFM particles weak, and, thus, leads us to prospect for high-density AFM-based devices. Dynamics of an AFM soliton can be induced by an elec- tric current or a spin wave [15 -- 17]. A particle immersed in a viscous medium exhibits a Brownian motion due to a random force that is required to exist to comply with the fluctuation-dissipation theo- rem (FDT) [18, 19]. An externally applied temperature gradient can also be a driving force, engendering a phe- nomenon known as thermophoresis [20]. Dynamics of an FM and an AFM includes spin damping, and, thus, in- volves thermal fluctuations at a finite temperature [21]. The corresponding thermal stochastic field influences dy- namics of a magnetic soliton [8, 22, 23], e.g., by assisting a current-induced motion of an FM DW [24]. FIG. 1. (Color online) A thermal stochastic force caused by a temperature gradient pushes an antiferromagnetic domain wall to a colder region. The diffusion coefficient of the domain wall is inversely proportional to a small damping constant, which may give rise to a sizable drift velocity. In this Rapid Communication, we study the Brown- ian motion of a soliton in an AFM under a tempera- ture gradient. We derive the stochastic Landau-Lifshitz- Gilbert (LLG) equation for an AFM with the aid of the FDT, which relates the fluctuation of the staggered and net magnetization to spin damping. We then derive the Langevin equation for the soliton's center of mass by em- ploying the collective coordinate approach [16, 25]. We develop the Hamiltonian mechanics for collective coordi- nates and conjugate momenta of a soliton, which sheds light on stochastic dynamics of an AFM soliton; it can be considered as a classical massive particle moving in a vis- cous medium. By considering a thermodynamic ensemble of solitons, we obtain the Fokker-Planck equation, from which we extract the average drift velocity. As a case study, we compute the drift velocity of a DW in a quasi one-dimensional easy-axis AFM. Thermophoresis of a Brownian particle is a multi- faceted phenomenon, which involves several competing mechanisms. As a result, a motion of a particle depends on properties of its environment such as a medium or a temperature T [26]. For example, particles in pro- tein (e.g., lysozyme) solutions move to a colder region for T > 294 K and otherwise to a hotter region [20, 27]. 5 1 0 2 l u J 8 ] l l a h - s e m . t a m - d n o c [ 2 v 4 5 8 7 0 . 3 0 5 1 : v i X r a Thermophoresis of an AFM soliton would be at least as complex as that of a Brownian particle. We focus on one aspect of it in this Rapid Communication; the effect of thermal stochastic force on dynamics of the soliton. We discuss two other possible mechanisms, the effects of a thermal magnon current and an entropic force [5], later in the Rapid Communication. Main results. -- Before pursuing details of derivations, we first outline our three main results. Let us consider a bipartite AFM with two sublattices that can be trans- formed into each other by a symmetry transformation of the crystal. Its low-energy dynamics can be devel- oped in terms of two fields: the unit staggered spin field n ≡ (m1 − m2)/2 and the small net spin field m ≡ (m1 + m2)/2 perpendicular to n. Here, m1 and m2 are unit vectors along the directions of spin angular momentum in the sublattices. Starting from the standard Lagrangian description of the antiferromagnetic dynamics [28], we will show below that the appropriate theory of dissipative dynamics of antiferromagnets at a finite temperature is captured by the stochastic LLG equation s( n + βn × m) =n × (h + hth), s( m + βm × m + αn × n) =n × (g + gth) + m × (h + hth), (1a) (1b) in conjunction with the correlators of the thermal stochastic fields gth and hth, (cid:104)gth j (r(cid:48), t(cid:48))(cid:105) = 2kBT αsδijδ(r − r(cid:48))δ(t − t(cid:48)), (2a) i (r, t)gth (cid:104)hth j (r(cid:48), t(cid:48))(cid:105) = 2kBT βsδijδ(r − r(cid:48))δ(t − t(cid:48)), (2b) i (r, t)hth which are independent of each other [29]. This is our first main result. Here, α and β are the damping constants associated with n and m, g ≡ −δU/δn and h ≡ −δU/δm are the effective fields conjugate to n and m, U [n, m] ≡ the magnetic susceptibility), and s ≡ S/V is the spin angular momentum density (V is the volume per spin) per each sublattice. The potential energy U [n(r, t)] is a general functional of n, which includes the exchange U [n] +(cid:82) dV m2/2χ is the potential energy (χ represents energy(cid:82) dV Aij∂in · ∂jn at a minimum [28]. Slow dynamics of stable magnetic solitons can often be expressed in terms of a few collective coordinates parametrizing slow modes of the system. The center of mass R represents the proper slow modes of a rigid soli- ton when the translational symmetry is weakly broken. Translation of the stochastic LLG equation (1) into the language of the collective coordinates results in our sec- ond main result, a Langevin equation for the soliton's center of mass R: M R + Γ R = −∂U/∂R + Fth, (3) which adds the stochastic force Fth to Eq. (5) of Tveten et al. [17]. The mass and dissipation tensors are symmet- ric and proportional to each other: Mij ≡ ρ(cid:82) dV (∂in · 2 ∂jn) and Γij ≡ Mij/τ, where τ ≡ ρ/αs is the relaxation time, ρ ≡ χs2 is the inertia of the staggered spin field n. The correlator of the stochastic field Fth obeys the Einstein relation (cid:104)F th j (t(cid:48))(cid:105) = 2kBT Γijδ(t − t(cid:48)). i (t)F th (4) A temperature gradient causes a Brownian motion of an AFM soliton toward a colder region. In the absence of a deterministic force, the average drift velocity is pro- portional to a temperature gradient V ∝ kB∇T in the linear response regime. The form of the proportional- ity constant can be obtained by a dimensional analy- sis. Let us suppose that the mass and dissipation ten- sors are isotropic. The Langevin equation (3) is, then, characterized by three scalar quantities: the mass M , the viscous coefficient Γ, and the temperature T , which define the unique set of natural scales of time τ ≡ M/Γ, length l ≡ √ kBT M /Γ, and energy
≡ kBT . Using these scales to match the dimension of a velocity yields V = −cµ(kB∇T ), where µ ≡ Γ−1 is the mobility of an AFM soliton and c is a numerical constant. The explicit solution of the Fokker-Planck equation, indeed, shows c = 1. This simple case illustrates our last main result; a drift velocity of an AFM soliton under a temperature gradient in the presence of a deterministic force F is given by V = µF − µ(kB∇T ). (5) For a DW in an easy-axis one-dimensional AFM, the mobility is µ = λ/2αsσ, where λ is the width of the wall and σ is the cross-sectional area of the AFM. For a numerical estimate, let us take an angular momen- tum density s = 2 nm−1, a width λ = 100 nm, and a damping constant α = 10−4 following the previous stud- ies [17, 30]. For these parameters, the AFM DW moves at a velocity V = 32 m/s for the temperature gradient of ∇T = 1 K/mm. Stochastic LLG equation. -- Long-wave dynamics of an AFM on a bipartite lattice at zero temperature can de- scribed by the Lagrangian [28] (cid:90) L = s dV m · (n × n) − U [n, m]. We use the potential energy U [n, m] ≡ (cid:82) dV m2/2χ + (6) U [n] throughout the Rapid Communication, which re- spects the sublattice exchange symmetry (n → −n, m → m). Minimization of the action subject to nonlinear con- straints n = 1 and n· m = 0 yields the equations of mo- tion for the fields n and m. Damping terms that break the time reversal symmetry can be added to the equa- tions of motion to the lowest order, which are first order in time derivative and zeroth order in spatial derivative. The resultant phenomenological LLG equations are given by s( n + βn × m) = n × h, s( m + βm × m + αn × n) = n × g + m × h (7a) (7b) [16, 30, 31]. The damping terms can be derived from the Rayleigh dissipation function which is described by the position R = (X, Y ) [34, 35]. Translation from the field language into that of collective coordinates can be done as follows. If the staggered spin field n is encoded by coordinates q as n(r, t) = n[r; q(t)], time dependence of n reflects evolution of the coordi- n = qi ∂n/∂qi. With the canonical momenta p nates: defined by (cid:90) (cid:90) 3 R = dV (αs n2 + βs m2)/2, (8) which is related to the energy dissipation rate by − U = 2R. The microscopic origin of damping terms does not concern us here but it could be, e.g., caused by thermal phonons that deform the exchange and anisotropy inter- action. At a finite temperature, thermal agitation causes fluc- tuations of the spin fields n and m. These thermal fluc- tuations can be considered to be caused by the stochas- tic fields gth and hth with zero mean, which are con- jugate to n and m, respectively; their noise correlators are then related to the damping coefficients by the FDT. The standard procedure to construct the noise sources yields the stochastic LLG equation (1). The correlators of the stochastic fields are obtained in the following way [18, 32]. Casting the linearized LLG equation (7) into the form {h, g} = γ⊗{ n, m} provides the kinetic coefficients γ. Symmetrizing the kinetic coefficients γ produces the correlators (2) of the stochastic fields consistent with the FDT. Langevin equation. -- For slow dynamics of an AFM, the energy is mostly dissipated through the temporal variation of the staggered spin field n due to m2 (cid:39) (ατ )2n2 (cid:28) n2 (from Eq. (7)), which allows us to set β = 0 to study long-term dynamics of the magnetic soli- ton [17]. At this point, we switch to the Hamiltonian for- malism of an AFM [33], which sheds light on the stochas- tic dynamics of a soliton. The canonical momentum field π conjugate to the staggered spin field n is π ≡ δL/δ n = sm × n. (9) The stochastic LLG equations (1) can be interpreted as Hamilton's equations, n = δH/δπ = π/ρ, π = −δH/δn− δR/δ n + gth, (10) (cid:90) with the Hamiltonian H ≡ dV π · n − L = (cid:90) π2 2ρ dV + U [n]. (11) Long-time dynamics of magnetic texture can often be captured by focusing on a small subset of slow modes, which are parametrized by the collective coordinates q = {q1, q2,···}. A classical example is a DW in a one- dimensional easy-axis magnet described by the position of the wall X and the azimuthal angle Φ [3, 33]. An- other example is a skyrmion in an easy-axis AFM film, pi ≡ ∂L ∂ qi = dV · π, ∂n ∂qi (12) Hamilton's equations (10) translate into M q = p, p + Γ q = F + Fth, where F ≡ −∂U/∂q is the deterministic force and F th (cid:82) dV ∂qin · gth is the stochastic force. Hamilton's equa- (13) i ≡ tions (13) can be derived from the Hamiltonian in the collective coordinates and conjugate momenta, H ≡ pTM−1p/2 + U (q), (14) with the Poisson brackets {qi, pj} = δij, {qi, qj} = {pi, pj} = 0. An AFM soliton, thus, behaves as a classi- cal particle moving in a viscous medium. We focus on a translational motion of a rigid AFM soliton by choosing its center of mass as the collective coordinates q = R; n(r, t) = n(r − R(t)). Eliminat- ing momenta from Hamilton's equations (13) yields the Langevin equation for the soliton's center of mass: τ R + R = µF + η, (15) where η ≡ µFth is the stochastic velocity. Here the mo- bility tensor of the soliton µ ≡ Γ−1 relates a deterministic force to a drift velocity (cid:104) R(cid:105) = µF at a constant temper- ature [36]. The mobility is inversely proportional to a damping constant, which can be a small number for an AFM, e.g., α ∼ 10−4 for NiO [37]. The correlator (2) of thermal stochastic fields is translated into the correlator of the stochastic velocity, (16) Z ≡(cid:82) Πi[dpidxi/2π] exp(−H/kBT ), which provides the (cid:104)ηi(t) ηj(t(cid:48))(cid:105) = 2kBT µijδ(t − t(cid:48)) ≡ 2Dijδ(t − t(cid:48)). From Eq. (16), we see that diffusion coefficient and the mobility of the soliton respect the Einstein-Smoluchowski relation: D = µkBT , which is expected on general grounds. It can also be explicitly verified as follows. A system of an ensemble of magnetic solitons at ther- mal equilibrium is described by the partition function autocorrelation of the velocity, (cid:104) xi xj(cid:105)/2 = M−1 ij kBT /2 (the equipartition theorem). In the absence of an ex- ternal force, multiplying τ xi + xi = ηi (15) by xj and symmetrizing it with respect to indices i and j give the equation, τ d2(cid:104)xixj(cid:105)/dt2 + d(cid:104)xixj(cid:105)/dt = 2τ(cid:104) xi xj(cid:105), where the first term can be neglected for long-term dynamics t (cid:29) τ . This equation in conjunction with the autocorre- lation of the velocity allows us to obtain the diffusion co- efficient Dij in Eq. (16), (cid:104)xixj(cid:105) = 2kBT τ M−1 ij t = 2Dijt, without prior knowledge about the correlator (2) of the stochastic fields. Average dynamics. -- An AFM soliton exhibits Brown- ian motion at a finite temperature. The following Fokker- Planck equation for an ensemble of solitons in an inho- mogeneous medium describes the evolution of the density ρ(R, t) at time t (cid:29) τ : +∇·j = 0, with j ≡ µFρ−D∇ρ−DT(kB∇T ), (17) ∂ρ ∂t where DT ≡ µρ is the thermophoretic mobility (also known as the thermal diffusion coefficient) [20, 38]. A steady-state current density j = µFρ0−DT(kB∇T ) with a constant soliton density ρ(r, t) = ρ0 solves the Fokker- Planck equation (17), from which the average drift veloc- ity of a soliton can be extracted [39]: V = µF − µ(kB∇T ). (18) Let us take an example of a DW in a quasi one- dimensional easy-axis AFM with the energy U [n] = z)/2. A DW in the equilibrium is n(0) = (sin θ cos Φ, sin θ sin Φ, cos θ) with cos θ = (cid:82) dV (A∂xn2 − Kn2 tanh[(x − X)/λ], where λ ≡ (cid:112)A/K is the width of the wall. The position X and the azimuthal angle Φ parametrize zero-energy modes of the DW, which are en- gendered by the translational and spin-rotational symme- try of the system. Their dynamics are decoupled, ΓXΦ = 0, which allows us to study the dynamics of X separately from Φ. The mobility of the DW is µ = λ/2αsσ, where σ is the cross-sectional area of the AFM. The average drift velocity (18) is given by V = − 1 2α kBλ∇T sσ . (19) Discussion -- The deterministic force F on an AFM soliton can be extended to include the effect of an elec- tric current, an external field, and a spin wave [15 -- 17]. It depends on details of interaction between the soliton and the external degrees of freedom, whose thorough under- standing would be necessary for a quantitative theory for the deterministic drift velocity µF. The Brownian drift velocity V (18) is, however, determined by local property of the soliton. We have focused on the thermal stochastic force as a trigger of thermophoresis of an AFM soliton in this Rapid Communication. There are two other possi- ble ingredients of thermophoresis of a magnetic soliton. One is a thermal magnon current, scattering with which could exert a force on a soliton [40]. The other is an entropic force, which originates from thermal softening of the order-parameter stiffness [5]. Effects of these two mechanisms have not been studied for an AFM soliton; full understanding of its thermophoresis is an open prob- lem. In order to compare different mechanisms of thermally- driven magnetic soliton motion, let us address a closely 4 related problem of thermophoresis of a DW in a quasi one-dimensional FM wire with an easy-xz-plane easy- z-axis [3], which has attracted a considerable scrutiny recently. To that end, we have adapted the approach de- veloped in this Rapid Communication to the FM case, which leads to the conclusion that a DW drifts to a colder region by a Brownian stochastic force at the ve- locity given by the same expression for an AFM DW, V B = −kBλ∇T /2αsσ [41]. A thermal magnon cur- rent pushes a DW to a hotter region at the velocity V M = kB∇T /6π2sλm, where λm ≡ (cid:112)A/sT is the thermal-magnon wavelength [7]. According to Schlick- eiser et al. [5], an entropic force drives a DW to a hotter region at the velocity V E = kB∇T /4sa, where a is the lattice constant. The Brownian stochastic force, there- fore, dominates the other forces for a thin wire, σ (cid:28) λa/α (supposing rigid motion) [42]. Within the framework of the LLG equations that are first order in time derivative, the thermal noise is white as long as slow dynamics of a soliton is concerned, i.e., the highest characteristic frequency of the natural modes parametrized by the collective coordinates is much smaller than the temperature scale, ω (cid:28) kBT . The thermal noise could be colored in general [26], e.g., for fast excitations of magnetic systems, which may be ex- amined in the future. In addition, local energy dissi- pation (8) allowed us to invoke the standard FDT at the equilibrium to derive the stochastic fields. It would be worth pursuing to understand dissipative dynamics of general magnetic systems, e.g., with nonlocal energy dissipation with the aid of generalized FDTs at the out- of-equilibrium [43]. We have studied dynamics of an AFM soliton in the Hamiltonian formalism. Hamiltonian's equations (13) for the collective coordinates and conjugate momenta can be derived from the Hamiltonian (14) with the conventional Poisson bracket structure. By replacing Poisson brackets with commutators, the coordinates and conjugate mo- menta can be promoted to quantum operators. This may provide a one route to study the effect of quantum fluc- tuations on dynamics of an AFM soliton [44]. After the completion of this work, we became aware of two recent reports. One is on thermophoresis of an AFM skyrmion [35], whose numerical simulations sup- port our result on diffusion coefficient. The other is on thermophoresis of an FM DW by a thermodynamic magnon recoil [45]. We are grateful for useful comments on the manuscript to Joseph Barker as well as insightful discussions with Scott Bender, So Takei, Gen Tatara, Oleg Tretiakov, and Jiadong Zang. This work was supported by the US DOE- BES under Award No. DE-SC0012190 and in part by the ARO under Contract No. 911NF-14-1-0016 (S.K.K. and Y.T.) and by the US DOE-BES under Award No. DE- FG02-08ER46544 (O.T.). [1] A. Kosevich, B. Ivanov, and A. Kovalev, Phys. Rep. 194, 117 (1990). [2] S. S. P. Parkin, M. Hayashi, and L. Thomas, Science 320, 190 (2008). [3] N. L. Schryer and L. R. Walker, J. Appl. Phys. 45, 5406 (1974). [4] L. Berger, Phys. Rev. B 54, 9353 (1996); J. Slonczewski, J. Magn. Magn. Mater. 159, L1 (1996). [5] D. Hinzke and U. Nowak, Phys. Rev. Lett. 107, 027205 (2011); F. Schlickeiser, U. Ritzmann, D. Hinzke, and U. Nowak, Phys. Rev. Lett. 113, 097201 (2014). [6] G. E. W. Bauer, E. Saitoh, and B. J. van Wees, Nat. Mater. 11, 391 (2012). [7] P. Yan, X. S. Wang, and X. R. Wang, Phys. Rev. Lett. 107, 177207 (2011); A. A. Kovalev and Y. Tserkovnyak, Europhys. Lett. 97, 67002 (2012). [8] L. Kong and J. Zang, Phys. Rev. Lett. 111, 067203 (2013); A. A. Kovalev, Phys. Rev. B 89, 241101 (2014). [9] W. Jiang, P. Upadhyaya, Y. Fan, J. Zhao, M. Wang, L.-T. Chang, M. Lang, K. L. Wong, M. Lewis, Y.-T. Lin, J. Tang, S. Cherepov, X. Zhou, Y. Tserkovnyak, R. N. Schwartz, and K. L. Wang, Phys. Rev. Lett. 110, 177202 (2013); J. Chico, C. Etz, L. Bergqvist, O. Eriks- son, J. Fransson, A. Delin, and A. Bergman, Phys. Rev. B 90, 014434 (2014). [10] I. Zuti´c, J. Fabian, and S. Das Sarma, Rev. Mod. Phys. 76, 323 (2004). [11] J. Sinova and I. Zutic, Nat. Mater. 11, 368 (2012). [12] A. H. MacDonald and M. Tsoi, Philos. Trans. R. Soc., A 369, 3098 (2011); E. V. Gomonay and V. M. Loktev, Low Temp. Phys. 40, 17 (2014). [13] M. Fiebig, N. P. Duong, T. Satoh, B. B. V. Aken, K. Miyano, Y. Tomioka, and Y. Tokura, J. Phys. D: Appl. Phys. 41, 164005 (2008). [14] Q. Zhang, A. V. Nurmikko, A. Anguelouch, G. Xiao, and A. Gupta, Phys. Rev. Lett. 89, 177402 (2002). [15] K. M. D. Hals, Y. Tserkovnyak, and A. Brataas, Phys. Rev. Lett. 106, 107206 (2011). [16] E. G. Tveten, A. Qaiumzadeh, O. A. Tretiakov, and A. Brataas, Phys. Rev. Lett. 110, 127208 (2013). [17] E. G. Tveten, A. Qaiumzadeh, and A. Brataas, Phys. Rev. Lett. 112, 147204 (2014); S. K. Kim, Y. Tserkovnyak, and O. Tchernyshyov, Phys. Rev. B 90, 104406 (2014). [18] L. D. Landau, E. M. Lifshitz, and L. P. Pitaevskii, Sta- tistical Physics, Part 1, 3rd ed. (Pergamon Press, New York, 1980). [19] M. Ogata and Y. Wada, J. Phys. Soc. Jpn. 55, 1252 (1986). [20] R. Piazza and A. Parola, J. Phys.: Condens. Matter 20, 153102 (2008). [21] W. F. Brown, Phys. Rev. 130, 1677 (1963); R. Kubo and N. Hashitsume, Prog. Theor. Phys. Suppl. 46, 210 (1970); J. L. Garc´ıa-Palacios and F. J. L´azaro, Phys. Rev. B 58, 14937 (1998); J. Foros, A. Brataas, G. E. W. Bauer, and Y. Tserkovnyak, Phys. Rev. B 79, 214407 (2009); S. Hoffman, K. Sato, and Y. Tserkovnyak, Phys. Rev. B 88, 064408 (2013); U. Atxitia, P. Nieves, and O. Chubykalo-Fesenko, Phys. Rev. B 86, 104414 (2012). [22] B. A. Ivanov, A. K. Kolezhuk, and E. V. Tartakovskaya, J. Phys.: Condens. Matter 5, 7737 (1993). 5 [23] C. Schutte, J. Iwasaki, A. Rosch, and N. Nagaosa, Phys. Rev. B 90, 174434 (2014). [24] R. A. Duine, A. S. N´unez, and A. H. MacDonald, Phys. Rev. Lett. 98, 056605 (2007). [25] O. A. Tretiakov, D. Clarke, G.-W. Chern, Y. B. Baza- liy, and O. Tchernyshyov, Phys. Rev. Lett. 100, 127204 (2008). [26] S. Hottovy, G. Volpe, and J. Wehr, Europhys. Lett. 99, 60002 (2012). [27] S. Iacopini and R. Piazza, Europhys. Lett. 63, 247 (2003). [28] A. F. Andreev and V. I. Marchenko, Sov. Phys. Usp. 23, 21 (1980). [29] Employing the quantum FDT would change the corre- lator of the stochastic fields to (cid:104)hth j (r(cid:48), ω(cid:48))(cid:105) = [2πδijαsω/ tanh(ω/2kBT )]δ(r− r(cid:48))δ(ω− ω(cid:48)) in the fre- quency space [18]. We focus on slow dynamics of an AFM soliton in the manuscript, ω (cid:28) kBT , which allows us to replace ω/ tanh(ω/2kBT ) with 2kBT , yielding Eq. (2). [30] S. Takei, B. I. Halperin, A. Yacoby, and Y. Tserkovnyak, i (r, ω)hth Phys. Rev. B 90, 094408 (2014). [31] B. A. Ivanov and D. D. Sheka, Phys. Rev. Lett. 72, 404 (1994); N. Papanicolaou, Phys. Rev. B 51, 15062 (1995); H. V. Gomonay and V. M. Loktev, Phys. Rev. B 81, 144427 (2010); A. C. Swaving and R. A. Duine, Phys. Rev. B 83, 054428 (2011). [32] Y. Tserkovnyak and C. H. Wong, Phys. Rev. B 79, 014402 (2009). [33] H. J. Mikeska, J. Phys. C: Solid St. Phys. 13, 2913 (1980); F. D. M. Haldane, Phys. Rev. Lett. 50, 1153 (1983). [34] I. Raicevi´c, D. Popovi´c, C. Panagopoulos, L. Benfatto, M. B. Silva Neto, E. S. Choi, and T. Sasagawa, Phys. Rev. Lett. 106, 227206 (2011); X. Zhang, Y. Zhou, and M. Ezawa, arXiv:1504.01198. [35] J. Barker and O. A. Tretiakov, arXiv:1505.06156. [36] The relation between the mobility and dissipation tensor can also be understood by equating the (twice) Rayleigh function, 2R = VTΓV, to the dissipation governed by the mobility, − E = V · F = VTµ−1V. [37] T. Kampfrath, A. Sell, G. Klatt, A. Pashkin, S. Mahrlein, T. Dekorsy, M. Wolf, M. Fiebig, A. Leitenstorfer, and R. Huber, Nature Photon. 5, 31 (2011). [38] N. van Kampen, IBM J. Res. Dev. 32, 107 (1988); T. Kuroiwa and K. Miyazaki, J. Phys. A: Math. Theor. 47, 012001 (2014). [39] M. Braibanti, D. Vigolo, and R. Piazza, Phys. Rev. Lett. 100, 108303 (2008). [40] The effect is, however, negligible at a temperature lower than the magnon gap (∼ 40 K for NiO [37]). Also for our example -- a DW in a 1D easy-axis AFM -- the conserva- tive force and torque exerted by magnons vanish [17]. [41] Unlike 1D domain walls, Brownian motions are drasti- cally distinct between 2D FM and AFM solitons due to the gyrotropic force, which significantly slows down fer- romagnetic diffusion [23, 35]. [42] A DW in a wire with a large crosssection σ (cid:29) a2 forms a 2D membrane. Its fluctuations foment additional soft modes of the dynamics, which needs to be taken into account to understand the dynamics of such a DW [25]. [43] M. Baiesi, C. Maes, and B. Wynants, Phys. Rev. Lett. 103, 010602 (2009); U. Seifert and T. Speck, Europhys. Lett. 89, 10007 (2010). [44] S.-Z. Lin and L. N. Bulaevskii, Phys. Rev. B 88, 060404 (2013). [45] P. Yan, Y. Cao, and J. Sinova, arXiv:1504.00651. 

1701.08016

1

1701

Interaction effects in a chaotic graphene quantum billiard

We investigate the local electronic structure of a Sinai-like, quadrilateral graphene quantum billiard with zigzag and armchair edges using scanning tunneling microscopy at room temperature. It is revealed that besides the $(\sqrt{3}\times\sqrt{3})R30${\deg} superstructure, which is caused by the intervalley scattering, its overtones also appear in the STM measurements, which are attributed to the Umklapp processes. We point out that these results can be well understood by taking into account the Coulomb interaction in the quantum billiard, accounting for both the measured density of state values and the experimentally observed topography patterns. The analysis of the level-spacing distribution substantiates the experimental findings as well. We also reveal the magnetic properties of our system which should be relevant in future graphene based electronic and spintronic applications.

cond-mat.mes-hall

cond-mat

Interaction effects in a chaotic graphene quantum billiard Imre Hagym´asi,1 P´eter Vancs´o,2, 3 Andr´as P´alink´as,4 and Zolt´an Osv´ath4 1Strongly Correlated Systems "Lendulet" Research Group, Institute for Solid State Physics and Optics, MTA Wigner Research Centre for Physics, Budapest H-1525 P.O. Box 49, Hungary 22D Nanoelectronics "Lendulet" Research Group, Institute of Technical Physics and Materials Science, HAS Centre for Energy Research, Budapest H-1525 P.O. Box 49, Hungary 3Department of Physics, University of Namur, 61 rue de Bruxelles, 5000 Namur, Belgium 4Institute of Technical Physics and Materials Science, HAS Centre for Energy Research, Budapest H-1525 P.O. Box 49, Hungary √ 3 × √ We investigate the local electronic structure of a Sinai-like, quadrilateral graphene quantum bil- liard with zigzag and armchair edges using scanning tunneling microscopy at room temperature. 3)R30° superstructure, which is caused by the intervalley It is revealed that besides the ( scattering, its overtones also appear in the STM measurements, which are attributed to the Umk- lapp processes. We point out that these results can be well understood by taking into account the Coulomb interaction in the quantum billiard, accounting for both the measured density of state values and the experimentally observed topography patterns. The analysis of the level-spacing dis- tribution substantiates the experimental findings as well. We also reveal the magnetic properties of our system which should be relevant in future graphene based electronic and spintronic applications. I. INTRODUCTION Quantum dots and quantum billiards have been in the focus of mesoscopic systems for the last three decades. The investigation of irregular shaped quantum billiards revealed how the classicaly chaotic behavior manifests in their energy spectrums.1 -- 4 Although many properties of quantum dots with two-dimensional electron gas are now well understood, the appearance of new two-dimensional materials has renewed the interest in such systems both experimentally and theoretically. A paradigmatic exam- ple is graphene, the two-dimensional honeycomb lattice of carbon atoms, which has been an actively researched area since its discovery in 2004.5 Recent developments of nanofabrication and growth techniques make it now possible to create nanostructures with well-defined crys- tallographic edges.6 Since they serve as building blocks of future nanodevices, it is crucial to understand their properties. As the Hamiltonian is different and various edge configurations can be present, it is not trivial how these effects alter the properties of quantum dots. An illustrative example is a classicaly chaotic neutrino bil- liard, investigated by Berry and Mondragon,7 where the energy spectrum turned out to be governed by the Gaus- sian Unitary Ensemble (GUE) instead of the Gaussian Orthogonal Ensemble (GOE) as one would naively ex- pect, since time-reversal symmetry seems to be present at first glance. The Hamiltonian of graphene is similar to that of the neutrino billiards, however, the various edge types can lead to further effects.8,9 This is the reason why special attention has been paid to graphene based billiards.8 -- 18 It has been shown that in the absence of intervalley scattering, which case is identical to the neu- trino billiard, the level-spacing distribution follows GUE statistics.8,9 However, when the valleys are not indepen- dent the time-reversal symmetry is restored, the level- spacing distribution obeys GOE statistics.8,9 Although many properties of bulk graphene can be un- derstood within a noninteracting picture, interaction ef- fects can become important if one considers finite sam- ples. Even a weak electron-electon interaction can lead to drastic effects, for example, in zigzag graphene nanorib- bons the paramagnetic ground state becomes unstable against magnetic ordering.19 -- 27 This is due to the fact that the zigzag edge states form a flat band at the Fermi energy and the corresponding large density of states is not favorable energetically in the presence of the interac- tion, therefore a symmetry-breaking ground state occurs. The same scenario can happen in graphene quantum dots possessing zigzag edges.28,29 In this paper we examine a quadrilateral shaped graphene quantum dot, which is a truncated triangle having three zigzag and one armchair edges. This peculiar shape resembles the theoretically well-studied Sinai billiard.30 In this joint experimental and theoretical study, we use scanning tunneling microscopy (STM) and perform theoretical calculations to understand the exper- imental results. Our main finding is that the electron- electron interaction must be taken into account to re- produce the STM images and local density of states (LDOS) measurements which highlight the important role of the interactions in graphene quantum dots even at room temperature. The results may have implications in the nanoscale engineering of the electronic and magnetic properties of graphene based functional surfaces. The paper is organized as follows. In Sec. II. A the in- vestigated quantum dot is introduced together with the experimental details, while in Sec. II. B the applied theo- retical methods are described. In Sec. III. A we examine the properties of the noninteracting quantum billiard us- ing the elements of quantum chaos. In Sec. III. B we discuss the experimental results together with our theo- retical findings. Sec. III. C presents the magnetic proper- ties of the quantum dot based on the theoretical results. 7 1 0 2 n a J 7 2 ] l l a h - s e m . t a m - d n o c [ 1 v 6 1 0 8 0 . 1 0 7 1 : v i X r a Finally, in Sec. IV. our conclusions are presented. II. METHODS A. Experimental details Graphene grown by chemical vapour deposition onto electro-polished copper foil31 was transferred onto highly oriented pyrolytic graphite (HOPG) substrate using ther- mal release tape. An etchant mixture consisting of CuCl2 aqueous solution (20%) and hydrochloric acid (37%) in 4:1 volume ratio was used. After etching the copper foil, the tape holding the graphene was rinsed in dis- tilled water, then dried and pressed onto the HOPG surface. The tape/graphene/HOPG sample stack was placed onto a hot plate at 95 °C. At this temperature the tape released the graphene and it could be removed easily. Graphene nanostructures were obtained by an- nealing the graphene/HOPG sample at 650 °C for two hours in argon atmosphere. STM and tunneling spec- troscopy (STS) measurements were performed using a DI Nanoscope E operating under ambient conditions. The investigated graphene quantum dot is a truncated trian- gle, as shown in Fig. 1 (a). Atomic resolution images obtained on the dot (inset of Fig. 1 (a)) reveal that the nanostructure has three zigzag edges and one armchair edge. The formation of predominantly zigzag edges is due to the fact that these edges are more resistive to thermal oxidation.32 B. Theory To describe the graphene quantum dot, we use the π- band model of graphene, where a lattice site can host two electrons at most with opposite spins. We con- sider nearest-neighbor hopping terms only, and model the Coulomb repulsion by using a local Hubbard interaction term. Thus, we arrive at the Hamiltonian: H = −t † c iσcjσ + U ni↑ ni↓ − µ niσ, (1) (cid:88) (cid:104)ij(cid:105)σ (cid:88) iσ (cid:88) i where the summation (cid:104)ij(cid:105) extends over all nearest- † neighbor pairs. The operator c iσ (ciσ) creates (annihi- lates) an electron at site i with spin σ and niσ is the corresponding particle number operator. The nearest- neighbor hopping amplitude is given by t, U is the onsite Hubbard interaction strength and µ is the chemical po- tential. Throughout the paper we set t = 3 eV, which defines the energy scale of the system and tune the chem- ical potential close to the half-filled case to account for the small p-doping in the experiment. We assume a weak Coulomb repulsion U/t = 1, in agreement with previous experiments.33 Since the Hamiltonian (1) can be solved exactly for very small systems, we must use approxima- tions to obtain results for realistic systems. The most 2 FIG. 1. (a) STM image of the investigated graphene quan- tum dot. Tunneling parameters: U = 200 mV, I = 1 nA. The atomic resolution inset image shows the crystallographic orientation in the dot. (b) Height profile taken along the line section 1 in (a), showing monolayer thickness. commonly used one is the mean-field approach, whose application is justified for moderate values of electron- electron interaction in case of graphene nanodisks.29 By neglecting the fluctuation terms in the Hamiltonian (1), we obtain an effective single-particle Hamiltonian HMF = −t † c iσcjσ + U (cid:104)ni¯σ(cid:105)niσ − µ niσ, (2) (cid:88) (cid:104)ij(cid:105)σ (cid:88) iσ (cid:88) iσ where the unknown electron densities, (cid:104)ni¯σ(cid:105) are deter- mined by using the standard self-consistent procedure and µ is determined by the conservation of the electron number. Having obtained the mean-field solution one can calculate then the LDOS in the quantum dot as a function of energy from the Green function: ρi(E) = − 1 π Im Gii(E) = − 1 π Im((E + i0+)I − HMF)−1 ii , (3) where I is the identity operator. 3 III. RESULTS A. Noninteracting case Before diving into the details of the experimental and theoretical results, it is worth investigating what we can learn from the noninteracting tight-binding picture. Therefore, we perform tight-binding calculations for this peculiar geometry, a truncated triangle with zigzag and armchair edges, shown in Fig. 1 (a). To account for the edge roughness and to obtain a more realistic geometry, we removed randomly certain edge atoms as shown in Fig. 2. Thus our model contains 4310 atoms. We cal- FIG. 3. The energy spectrum of the quantum dot near the Fermi-energy with and without the Hubbard term. The role of the interaction is discussed in Sec. II. B. is the average density of states in the interval (E, E+dE). In graphene quantum billiards, we can approximate this quantity by its bulk value: (cid:104)ρ(E)(cid:105) ≈ ρbulk(E) = 1 π A (vF )2E, (4) where A is the area of the quantum billiard, vF is the Fermi velocity. According to our previous definition in 3√3vF /a ≈ 3 eV and a is the lattice Sec. II. B, t = 2 spacing of graphene. In completely integrable systems, the energy levels are uncorrelated, and their level-spacing statistics follows Poisson distribution: P (S) = exp(−S). (5) In chaotic systems with (or without) time-reversal sym- metry the level-spacing obeys the GOE (or GUE) of ran- dom matrices:3,4 (cid:18) (cid:19) πS2 4 4S2 π − PGOE(S) = PGUE(S) = − (cid:18) S exp π 2 32 π2 S2 exp , (cid:19) . (6) (7) It is worth noting that in integrable systems energy lev- els lie close to each other, since the probability distribu- tion is maximal at S = 0. In contrast, chaotic systems exhibit level repulsion since the corresponding probabil- ity densities vanish as S → 0, which is often referred as avoided level crossings. The zero-energy states, which are essentially degenerate, are due to the zigzag edges and localized on segments of the zigzag boundaries. Since they would cause an artificial contribution in the level distribution, we consider the interval 0.1 (cid:46) E/t (cid:46) 1, which contains approximately 400 eigenvalues. The level spacing distribution for our noninteracting quadrilateral FIG. 2. The geometry used in the model calculations. The dot consists of 4310 atoms. culate the energy spectrum by diagonalizing the tight- binding Hamiltonian, which is shown in Fig. 3 (U/t = 0). It is observed that the zero-energy states, which corre- spond to the Fermi level, are multiply degenerate. This may not surprise us if we recall that zigzag-shaped nan- odisks also have degenerate zero-energy states, which are attributed to the presence of edge-localized states.28 It is worth mentioning that this degeneracy is still present but partially resolved in our case due to the armchair edge and edge defects. To reveal the properties of integrability, we examine the level-spacing distribution, P (S). This quantity re- flects the nature of the corresponding classical dynam- ics of the system and follows certain university classes depending on the Hamiltonian.3,4 By definition, P (S)dS gives the probability that the distance between the neigh- boring values of the unfolded spectrum, (cid:104)ρ(E)(cid:105)(Ei+1 − Ei), lies in the interval (S, S + dS), where Ei are the eigenvalues of the tight-binding Hamiltonian and (cid:104)ρ(E)(cid:105) En(eV)-1.0-0.50.00.51.0n210021202140216021802200U/t=0U/t=1graphene billiard is shown in Fig. 4 together with the distribution functions Eqs. (5)-(7) corresponding to the university classes. In contrast to the equilateral triangle 4 FIG. 4. The level-spacing distribution obtained from the un- folded spectrum. The solid red, blue and black lines corre- spond to the Poisson, GOE and GUE distributions. billiard which is an integrable system13 our results clearly indicate that our quantum billiard is a classicaly chaotic system. Moreover, the calculated level-spacings are in a very good agreement with the GOE statistics and not with the GUE as one might expect based on the connec- tion with the neutrino billiard. This confirms what has been mentioned in Sec. I that the two valleys are coupled to each other due to the armchair edge and the defects on the zigzag edges, which restore the time-reversal sym- metry of the quantum billiard. B. Experimental results and their theoretical explanations In the previous subsection we have demonstrated that our quantum dot is a chaotic billiard based on its level- spacing distribution and the corresponding university class is GOE, which suggests strong intervalley scatter- ing. In what follows we show direct experimental evi- dence which supports this scenario and also demonstrate how the electron-electron interaction modifies the mea- sured properties of the system. First of all, we discuss the measured dI/dV spectra which correspond to the LDOS of the systems. The ex- perimental results measured in the middle of the quan- tum dot and on the HOPG substrate are shown in Fig. 5. The measured HOPG LDOS exhibits the known linear behavior, but it is asymmetric with respect to the Dirac point and has a finite amount of p-doping (80 meV) due to impurities in the system. It is easily seen that the quantum dot exhibits similar behavior to the HOPG far FIG. 5. The measured LDOS as a function of the voltage, the solid black and red lines correspond to the HOPG substrate and the quantum dot, respectively. The dotted blue line rep- resents the theoretical result for the LDOS of the quantum dot. In the theoretical calculation a Gaussian broadening 0.1 eV is used to account for the finite temperature. from the Fermi energy, however, higher LDOS appears near the Dirac point. If we recall the energy spectrum of the noninteracting system in Fig. 3, we observe that a gap ∆ ∼ 0.2 eV is opened, which is not visible in the measured LDOS. (We also investigated the role of the next-nearest-neighbor (NNN) hopping term on the spec- trum of the quantum dot, since it may result in simi- lar effects34. We found that the inclusion of NNN hop- ping lifts the degeneracy at the Fermi level, but the ob- served band gap quantitatively remains unchanged. See Appendix for details.) This is the first indication that the noninteracting model does not describe the dot accu- rately. We know that such highly degenerate levels near the Fermi-energy are very sensitive to even weak electron- electron interaction, just like in graphene nanoribbons with zigzag edges. Switching on the interaction, we can see from Fig. 3 that the energy levels near the Fermi- energy are no longer degenerate even for a small value of the Hubbard-U , U/t = 1, while the high-lying states are not modified. (Note that this weak interaction does not alter the energy levels, where the level-spacing dis- tribution was taken.) Thus, the obtained spectrum with electron-electron interaction is much closer to what we expect from the LDOS measurements in Fig. 5. In or- der to compare the results we also plot the calculated LDOS inside the quantum dot taking into account the finite p-doping (Fig. 5). The measured and the calcu- lated values are in good agreement, however, some dis- crepancy is observed slightly below the Fermi energy, where the calculated minimum does not appear. This can be understood considering that the LDOS of HOPG is asymmetric, therefore its slope below the Fermi energy P(s)00.20.40.60.81s00.511.522.53QDPoissonGOEGUEdI/dU(a.u.)20406080100U(mV)-400-2000200400HOPGQDtheory5 FIG. 6. The topographic STM images within the quantum dot. Each column corresponds to different voltages where the measurement was done: (a), (b) and (c) belong to U = 100 mV, 25 mV and -50 mV, respectively. The first row shows √ 3 × √ the measured images, the second one contains the simulated images, while the third row displays the Fourier spectrum of 3)R30° the measured images. The spots inside the white circles in the Fourier spectrum are the components of the ( superstructure, while the green circled spots are their overtones. The non-circled spots correspond to the periodicity of the atomic lattice. is larger than above it. Since the HOPG also contributes to the tunneling current, their larger LDOS values can vanish the calculated LDOS minimum of the dot below the Fermi energy. In the following we discuss the measured topographic STM images. In Fig. 6 we can see superstructure pat- terns within the quantum dot measured at various volt- ages. Regarding the voltages, U = 100, 25 and -50 mV, where the STM images were taken, and the energy spec- trum in Fig. 3, then it is evident that the noninteracting tight-binding description fails. Since these energy val- ues lie in the gap there is no way to account for the different STM images. In contrast, Fig 6 shows clearly different superstructure patterns. In Fig. 6 (a) the well- known (√3 × √3)R30° superstructure (marked by white circle) appears on the top of the atomic structure. This is more spectacular from the Fourier spectrum in Fig. 6 (a), where the components of the superstructure are clearly seen (marked by white circles). Since this structure orig- inates from the intervalley scattering between the K and K(cid:48) valleys due to the presence of the armchair edge,35,36 it confirms our previous theoretical findings based on the level-spacing distribution (Fig. 4) where the intervalley scattering was predicted by the GOE distribution. In Fig. 6 (b)-(c) besides the (√3 × √3)R30° superstruc- ture, stripes parallel to armchair edge and rings are ob- served in the measured STM images. In order to simulate these distinct STM images, we consider a finite electron- electron interaction using the procedure described in Sec. II. B. With the help of the calculated LDOS of the interacting system, we simulate the STM images with the simple Tersoff-Hamann approximation.37 Taking into account the electron-electron interaction, we can quan- titatively reproduce the measured STM images, includ- ing the (√3 × √3)R30° superstructure in Fig. 6 (a), the stripes in Fig. 6 (b), and the rings in Fig. 6 (c) at the dif- ferent voltages. If we take a closer look to the measured Fourier spectra, it can be easily seen that the Fourier components of the (√3×√3)R30° superstructure are al- ways present (inner hexagon, components marked with white circles). Moreover, their overtones (marked with green circles) appear with various amplitudes which re- (a) (b) (c) sults in the formation of different, more complex patterns in the topography images. The emergence of the over- tones measured by STM can be attributed to the Umk- lapp scattering processes, where an electron is scattered outside the first Brillouin Zone (BZ). Thus, in our quan- tum dot besides the usual scattering between K and K(cid:48) valleys in first BZ (K − K(cid:48) = q), electrons also scatter from the first BZ K valley to the second BZ K(cid:48) valley (K − K(cid:48) = q + G). This latter is equal to the shift of the K(cid:48) point back to the first BZ by a reciprocal lattice vector G. Beacuse the Umklapp processes are absent in the noninteracting case and they appear if electron- electron interaction is present in the half-filled Hubbard model,38 it serves as a further evidence for the presence of electron-electron interaction in our system. C. Magnetic properties Previously, we demonstrated that the inclusion of the electron-electron interaction provides a good agreement with the experimental results in terms of the measured LDOS and topographic STM images. Since our quantum dot is a quite complex system including doping and edge defects, it is worth revealing the effects of the interac- tion expicitly by considering the magnetic moments at the lattice sites. It has been mentioned earlier that the paramagnetic state becomes unstable against magnetic ordering if zigzag edges are present in the system,27,29,39 however, it also known that the presence of doping can vanish the appeared magnetic moments.40 To clarify the presence of the magnetism in our system we calculated the magnetic moments by taking into account the mea- sured p-doping (80 meV). We found that magnetic mo- ments are still present in the doped system (Fig. 7 (b)), although their values are smaller than in the half-filled case (Fig. 7 (a)). Quantitatively, the average momen- tum of up- and down-spin electrons in the doped quan- tum dot is decreased to one fourth of the undoped value. This reduction can be seen also in Fig. 7 (b), where the magnetic moments disappeared along the shortest zigzag side of the triangle and are localized only at the longer zigzag edge segments. We note that for larger doping values (120 meV) the magnetic moments totally vanish at the edges, which is consistent with previous findings.40 Our calculations indicate that only the zigzag edges are magnetic and no magnetization occurs along the armchair edge and around the defects in the zigzag edges. It is in agreement with the a priori expectations based on nanoribbons, since armchair nanoribbons are not magnetic, and edge defects in zigzag ribbons sup- press the magnetism.33,41 This follows from the fact that here the sublattice symmetry is restored, while zigzag edges contain atoms only from one sublattice. We also observe spin-collinear domain walls42 along the edges due to presence of the defects. Overall, we can conclude that the magnetism can be present in the system even at the presence of small doping and edge defects based on the 6 FIG. 7. Magnetic moments in the quantum dot for U/t = 1. Panels (a) and (b) correspond to the half-filled case and 80 meV p-doping, respectively. The blue and red colors encode the up and down spins, respectively. The magnitude of the spins depends on the radii of the circles, r, as exp(−r). The largest magnetic moments in both cases are Sz ∼ 0.12. mean-field calculations. Our analysis above gives further insight into the role of the electron-electron interaction in graphene quantum dots. IV. CONCLUSIONS We carried out a joint experimental and theoretical in- vestigation to analyze the properties of a quadrilateral Appendix: The role of next-nearest-neighbor hopping 7 In order to understand the lack of energy gap in the dI/dV spectra and the role of the interaction it is impor- tant to consider other effects that may alter the energy spectrum. The most important candidate is the next- nearest-neighbor (NNN) hopping, since it is known to break the electron-hole symmetry and shifts the disper- sionless edge states away from the zero energy.34 There- fore, we study the following Hamiltonian: (cid:48) (cid:88) (cid:88) (cid:88) (cid:104)ij(cid:105)σ † c iσcjσ−t † c iσcjσ +U (cid:104)(cid:104)ij(cid:105)(cid:105)σ ni↑ ni↓, (A.1) H = −t where (cid:104)(cid:104)ij(cid:105)(cid:105) denotes the summation over all NNN pairs. For the value of t(cid:48) we use t(cid:48) = 0.1 eV from the tight- binding fit to experimental data.43,44 To reveal the effect of NNN hopping, we calculated the energy spectrum of the Hamiltonian (A.1) in the noninteracting case, which is shown in Fig. 8. It is obviously seen that the NNN hop- i shaped graphene quantum billiard. First, we examined the quantum billiard in a tight-binding picture to grasp the properties of integrability. It turned out that the billiard is chaotic, moreover, its level-spacing distribu- tion agree well with the GOE distribution. This indi- cates strong intervalley scattering, which also manifests in the appearance of the (√3 × √3)R30° superstructure observed in the STM images measured under ambient conditions. We pointed out that by taking into account the interaction among the electrons, we were able to elucidate the spectroscopy (LDOS) measurements and STM images simultaneously. Using the mean-field ap- proximation of the Hubbard model we reproduced the (√3 × √3)R30°, stripe and ring superstructures as well. (√3 × √3)R30° superstructure, which also confirms the It was also revealed that the latter two structures ap- pear due to the overtones of the Fourier components of importance of the interaction even at room-temperature. Furthermore, we showed that the weak electron-electron interaction does not alter the higher energy levels (> 0.2 eV), only the ones near the Dirac point (lifted degen- eracy). Thus the chaotic nature and the time-reversal symmetry of the graphene billiard is preserved. We also demonstrated that edge-magnetism appears along the zigzag edges, and the magnetic moments disappear inside the quantum dot. The ability of tailoring the electronic and magnetic properties of graphene quantum dots can open new avenues for information coding at nanoscale. ACKNOWLEDGMENTS I. H. and P. V. contributed equally to this work. The research leading to these results has received fund- ing from the People Programme (Marie Curie Actions) of the European Unions Seventh Framework Programme under REA grant agreement No. 334377 and Lendulet programme of the Hungarian Academy of Sciences grant LP2014-14. Financial support from the National Re- search, Development and Innovation Office (NKFIH) through the OTKA Grant Nos. K119532 and K120569 are acknowledged. FIG. 8. The energy spectrum of the quantum dot near the Fermi-energy with and without NNN hopping terms. For better comparison, the result for the interacting case is shown too. ping lifts the degeneracy of the edge states, and they are slightly shifted towards higher energies. However, they still lie very close to each other, and the gap ∆ ∼ 0.2 eV in the spectrum hardly decreases. In contrast, the in- teraction completely removes the degeneracy of the zero- energy states and the gap disappears in the spectrum. 1 O. Bohigas, M. J. Giannoni, and C. Schmit, Phys. Rev. 4 M. L. Mehta, Random Matrices, 3rd ed. (Elsevier, Ams- Lett. 52, 1 (1984). 2 M. V. Berry, Proc. R. Soc. London Ser. A 400, 229 (1985). 3 H. A. Weidenmuller and G. E. Mitchell, Rev. Mod. Phys. 81, 539 (2009). terdam, 2004). 5 K. S. Novoselov, A. K. Geim, S. V. Morozov, D. Jiang, Y. Zhang, S. V. Dubonos, I. V. Grigorieva, and A. A. Firsov, Science 306, 666 (2004). En(eV)-1.0-0.50.00.51.0n210021202140216021802200U/t=0,t′=0U/t=1,t′=0U/t=0,t′=0.03t8 6 L. Tapaszt´o, G. Dobrik, P. Lambin, and L. P. Bir´o, Nat. 26 M. Golor, S. Wessel, and M. J. Schmidt, Phys. Rev. Lett. Nano. 3, 397 (2008). 112, 046601 (2014). 7 M. V. Berry and R. J. Mondragon, Proc. R. Soc. London 27 I. Hagym´asi and O. Legeza, Phys. Rev. B 94, 165147 Ser. A 412, 53 (1987). 8 J. Wurm, A. Rycerz, I. Adagideli, M. Wimmer, K. Richter, and H. U. Baranger, Phys. Rev. Lett. 102, 056806 (2009). 9 J. Wurm, K. Richter, and I. Adagideli, Phys. Rev. B 84, 205421 (2011). 10 L. Huang, Y.-C. Lai, and C. Grebogi, Chaos 21, 013102 (2011). 11 S. K. Hamalainen, Z. Sun, M. P. Boneschanscher, A. Up- pstu, M. Ijas, A. Harju, D. Vanmaekelbergh, and P. Lil- jeroth, Phys. Rev. Lett. 107, 236803 (2011). 12 L. A. Ponomarenko, F. Schedin, M. I. Katsnelson, R. Yang, E. W. Hill, K. S. Novoselov, and A. K. Geim, Science 320, 356 (2008). 13 A. Rycerz, Phys. Rev. B 85, 245424 (2012). 14 A. Rycerz, Phys. Rev. B 87, 195431 (2013). 15 L. Ying, G. Wang, L. Huang, and Y.-C. Lai, Phys. Rev. B 90, 224301 (2014). 16 W. Jolie, F. Craes, M. Petrovi´c, N. Atodiresei, V. Caciuc, S. Blugel, M. Kralj, T. Michely, and C. Busse, Phys. Rev. B 89, 155435 (2014). 17 J. G. G. S. Ramos, I. M. L. da Silva, and A. L. R. Barbosa, Phys. Rev. B 90, 245107 (2014). 18 J. G. G. S. Ramos, M. S. Hussein, and A. L. R. Barbosa, Phys. Rev. B 93, 125136 (2016). 19 Y.-W. Son, M. L. Cohen, and S. G. Louie, Phys. Rev. Lett. 97, 216803 (2006). 20 L. Yang, C.-H. Park, Y.-W. Son, M. L. Cohen, and S. G. Louie, Phys. Rev. Lett. 99, 186801 (2007). 21 T. Wassmann, A. P. Seitsonen, A. M. Saitta, M. Lazzeri, and F. Mauri, Phys. Rev. Lett. 101, 096402 (2008). 22 J. Fern´andez-Rossier, Phys. Rev. B 77, 075430 (2008). 23 O. V. Yazyev, Phys. Rev. Lett. 101, 037203 (2008). 24 J. Jung and A. H. MacDonald, Phys. Rev. B 79, 235433 (2009). 25 H. Feldner, Z. Y. Meng, T. C. Lang, F. F. Assaad, S. Wes- sel, and A. Honecker, Phys. Rev. Lett. 106, 226401 (2011). (2016). 28 M. Ezawa, Phys. Rev. B 76, 245415 (2007). 29 H. Feldner, Z. Y. Meng, A. Honecker, D. Cabra, S. Wessel, and F. F. Assaad, Phys. Rev. B 81, 115416 (2010). 30 H.-J. Stockmann, Quantum Chaos - An Introduction (Uni- versity Press, Cambridge, 1999). 31 A. P´alink´as, P. Sule, M. Szendro, G. Moln´ar, C. Hwang, L. P. Bir´o, and Z. Osv´ath, Carbon 107, 792 (2016). 32 P. Nemes-Incze, G. Magda, K. Kamar´as, and L. P. Bir´o, Nano Res. 3, 110 (2010). 33 G. Z. Magda, X. Jin, I. Hagym´asi, P. Vancs´o, Z. Osv´ath, P. Nemes-Incze, C. Hwang, L. P. Bir´o, and L. Tapaszt´o, Nature 514, 608 (2014). 34 M. Wimmer, A. R. Akhmerov, and F. Guinea, Phys. Rev. B 82, 045409 (2010). 35 Y. Niimi, T. Matsui, H. Kambara, K. Tagami, M. Tsukada, and H. Fukuyama, Phys. Rev. B 73, 085421 (2006). 36 K.-i. Sakai, K. Takai, K.-i. Fukui, T. Nakanishi, and T. Enoki, Phys. Rev. B 81, 235417 (2010). 37 J. Tersoff and D. R. Hamann, Phys. Rev. B 31, 805 (1985). 38 G. Ehlers, J. S´olyom, O. Legeza, and R. M. Noack, Phys. Rev. B 92, 235116 (2015). 39 A. Valli, A. Amaricci, A. Toschi, T. Saha-Dasgupta, K. Held, and M. Capone, Phys. Rev. B 94, 245146 (2016). 40 P. Potasz, A. D. Gu¸clu, A. W´ojs, and P. Hawrylak, Phys. Rev. B 85, 075431 (2012). 41 W. Wu, Z. Zhang, P. Lu, and W. Guo, Phys. Rev. B 82, 085425 (2010). 42 O. V. Yazyev and M. I. Katsnelson, Phys. Rev. Lett. 100, 047209 (2008). 43 R. S. Deacon, K.-C. Chuang, R. J. Nicholas, K. S. and A. K. Geim, Phys. Rev. B 76, 081406 Novoselov, (2007). 44 A. H. Castro Neto, F. Guinea, N. M. R. Peres, K. S. Novoselov, and A. K. Geim, Rev. Mod. Phys. 81, 109 (2009). 

1305.3880

2

1305

Bandgap Engineering of Strained Monolayer and Bilayer MoS2

We report the influence of uniaxial tensile mechanical strain in the range 0-2.2% on the phonon spectra and bandstructures of monolayer and bilayer molybdenum disulfide (MoS2) two-dimensional crystals. First, we employ Raman spectroscopy to observe phonon softening with increased strain, breaking the degeneracy in the E' Raman mode of MoS2, and extract a Gr\"uneisen parameter of ~1.06. Second, using photoluminescence spectroscopy we measure a decrease in the optical band gap of MoS2 that is roughly linear with strain, ~45 meV% strain for monolayer MoS2 and ~120 meV% strain for bilayer MoS2. Third, we observe a pronounced strain-induced decrease in the photoluminescence intensity of monolayer MoS2 that is indicative of the direct-to-indirect transition of the character of the optical band gap of this material at applied strain of ~1.5%, a value supported by first-principles calculations that include excitonic effects. These observations constitute the first demonstration of strain engineering the band structure in the emergent class of two-dimensional crystals, transition-metal dichalcogenides.

cond-mat.mes-hall

cond-mat

Bandgap Engineering of Strained Monolayer and Bilayer MoS2 Hiram J. Conley,† Bin Wang,† Jed I. Ziegler,† Richard F. Haglund Jr.,† Sokrates T. Pantelides,†,‡ and Kirill I. Bolotin∗,† Department of Physics and Astronomy, Vanderbilt University, and Materials Science and Technology Division, Oak Ridge National Laboratory E-mail: kirill.bolotin@vanderbilt.edu KEYWORDS: MoS2 , strain, bandgap engineering, photoluminescence, Grüneisen parameter Abstract We report the influence of uniaxial tensile mechanical strain in the range 0–2.2% on the phonon spectra and bandstructures of monolayer and bilayer molybdenum disulfide (MoS2 ) two-dimensional crystals. First, we employ Raman spectroscopy to observe phonon softening with increased strain, breaking the degeneracy in the E (cid:48) Raman mode of MoS2 , and extract a Grüneisen parameter of ∼1.06. Second, using photoluminescence spectroscopy we measure a decrease in the optical band gap of MoS2 that is approximately linear with strain, ∼45 meV/% strain for monolayer MoS2 and ∼120 meV/% strain for bilayer MoS2 . Third, we observe a pronounced strain-induced decrease in the photoluminescence intensity of monolayer MoS2 that is indicative of the direct-to-indirect transition of the character of the optical band gap of this material at applied strain of ∼1%. These observations constitute the first demonstration of strain engineering the band structure in the emergent class of two-dimensional crystals, transition-metal dichalcogenides. ∗To whom correspondence should be addressed †Department of Physics and Astronomy, Vanderbilt University ‡Materials Science and Technology Division, Oak Ridge National Laboratory 1 Monolayer1 molybdenum disulfide (MoS2 ), along with other monolayer transition metal dichalco- genides (MoSe2 , WS2 , WSe2 ) have recently been the focus of extensive research activity that follows the footsteps of graphene, a celebrated all-carbon cousin of MoS2 . 1 Unlike semimetallic graphene, monolayer MoS2 is a semiconductor with a large direct band gap. 2,3 The presence of a band gap opens a realm of electronic and photonic possibilities that have not been previously exploited in two-dimensional crystals and allows fabrication of MoS2 transistors with an on/off ra- tio exceeding 1 × 108 , 4,5 photodetectors with high responsivity, 6 and even LEDs. 7 Moreover, the direct nature of the band gap causes MoS2 to exhibit photoluminescence at optical wavelengths 2,3 with intensity that is tunable via electrical gating. 8–10 Finally, strong Coulomb interactions be- tween electrons and holes excited across the band gap of MoS2 lead to the formation of tightly bound excitons that strongly affect the optical properties of this material. 9,10 It has been well established that straining a two dimensional material shifts its phonon modes, allowing a simple method to detect strain in these materials. These shifts, that are due to the anharmonicity of molecular potentials, can be probed with micro-Raman spectroscopy. 11,12 Very recently, it has been proposed that mechanical strain can strongly perturb the band structure of MoS2 . It has been predicted that straining MoS2 modifies the band gap energy and the carrier effective masses. Moreover, at strains larger than 1% the lowest lying band gap changes from direct to indirect. 13–18 It has been suggested that strain engineering of the band structure of MoS2 could be used to increase carrier mobility of MoS2 , to create tunable photonic devices and solar cells, 19 and even to control the magnetic properties of MoS2 . 13,14 While strain perturbs the band structure of all materials, two-dimensional materials such as MoS2 can sustain strains greater than 11%, 20 allowing exceptional control of material properties by strain engineering. Here, we investigate the influence of uniaxial tensile strain from 0% to 2.2% on the phonon spectra and band gaps of both monolayer and bilayer MoS2 , by employing a four point bending apparatus (Fig. 1). First, with increasing strain, for both mono- and bilayer MoS2 we observe splitting of the Raman peak due to the E (cid:48) phonon mode into two distinct peaks that shift by 4.5 1Monolayer in this paper refers to one molecular molybdenum disulfide (MoS2 ) layer, or one layer of molybdenum atoms sandwiched between two layers of sulphur atoms. It is also sometimes refered to as trilayer MoS2 . 2 Figure 1: Straining MoS2 devices (a) Optical image of a bilayer MoS2 flake with titanium clamps attaching it to SU8/polycarbonate substrate. (b,c) Schematic of the beam bending apparatus used to strain MoS2 . (d) Photograph of bending apparatus with MoS2 under strain. and 1 cm−1 /% strain. Second, a linear redshift of 45meV/% strain of the position of the A peak in photoluminescence for monolayer MoS2 (53meV/% strain for bilayer MoS2 ) indicates a cor- responding reduction in band gap energy of these materials. Finally, we observe a pronounced strain-induced decrease in intensity of the photoluminescence of monolayer MoS2 . Our modelling and first-principles calculations indicate that this decrease is consistent with a transition of an op- tical band gap of MoS2 from direct to indirect at ∼1% strain, while the fundamental (or transport) band gap remains direct in the investigated regime of strain. Fabrication of controllably strained devices starts by mechanically exfoliating 21 MoS2 onto a layer of cross-linked SU8 photoresist deposited onto a polycarbonate beam. The number of layers of MoS2 is verified using Raman microscopy. 22 Titanium clamps are then evaporated through a shadow mask to prevent MoS2 from slipping against the substrate (Fig. 1a). Uniaxial strain is applied to MoS2 by controllably bending the polycarbonate beam in a four-point bending apparatus (Fig. 1c,d). Assuming that as-fabricated exfoliated devices before bending are virtually strain- free, 23 we can calculate that upon bending the substrate with radius of curvature R, the induced strain in these devices is ε = τ /R, where 2τ = 2–3mm is the thickness of the substrate (Fig. 1b). 12 Overall, we fabricated four monolayer and three bilayer MoS2 devices. The strained devices are probed with a confocal microscope (Thermo Scientific DXR) that is used to collect both Raman 3 R5 μma)b)d)c)MoS2and photoluminescence spectra. We employ a 532nm laser excitation source with average power ∼100 µW, which does not damage our samples. We first investigate the evolution of the Raman spectra of MoS2 with strain (Fig. 2). In un- strained monolayer MoS2 devices, consistent with previous reports, 22,24 we observe the A(cid:48) mode due to out-of-plane vibrations at 403 cm−1 and the doubly degenerate E (cid:48) mode due to in-plane vibrations of the crystal at 384 cm−1 . With increased strain, the A(cid:48) peak shows no measurable shift in position while the degenerate E (cid:48) peak splits into two subpeaks (in contrast to a previous report 25 ) that we label as E (cid:48)+ and E (cid:48)− (Fig. 2), as strain breaks the symmetry of the crystal. The A(cid:48) mode maintains its intensity as strain increases, while the total integrated intensity of the E (cid:48) peaks now splits between the E (cid:48)+ and E (cid:48)− peaks. The E (cid:48)− peak shifts by 4.5 ± 0.3 cm−1/% strain for monolayer devices and 4.6 ± 0.4 cm−1/% strain for bilayer devices, while the E (cid:48)+ peak shifts by 1.0 ± 1 cm−1/% strain for monolayer devices and 1.0 ± 0.9 cm−1/% strain for bilayer devices, consistent with our first- principles calculations (dashed lines in Fig. 2; details of the calculations can be found in the supplementary materials). For applied strain in the range 0–2%, the peak positions shift at nearly identical rates for all measured devices and do not exhibit hysteresis in multiple loading/unloading cycles, indicating that MoS2 does not slip against the substrate and that the strain does not generate a significant number of defects. Bilayer devices behave in a similar manner but fail, either due to breaking or slipping of the MoS2 , at strains larger than 1%. The strain dependence of the Raman E (cid:48) mode enables us to calculate parameters characterizing anaharmonicity of molecular potentials, the Grüneisen parameter, γ , and the shear deformation potential, β : γE (cid:48) = − ∆ωE (cid:48)+ + ∆ωE (cid:48)− 2ωE (cid:48) (1 − ν )ε ∆ωE (cid:48)+ − ∆ωE (cid:48)− ωE (cid:48) (1 + ν )ε βE (cid:48) = (1) (2) Here ω is the frequency of the Raman mode, ∆ω is the change of frequency per unit strain, and ν is Poisson’s ratio, which for a material adhering to a substrate is the Poisson ratio of the substrate, 4 Figure 2: Phonon softening of single layer MoS2 (a) Evolution of the Raman spectrum as a device is strained from 0 to 1.6%. (b) The peak location of the the E (cid:48)+ and E (cid:48)− Raman modes, extracted by fitting the peaks to a Lorentzian, as their degeneracy is broken by straining MoS2 . Different colors represent individual devices. Dashed lines are the results of our first-principles calculations after subtraction of 9 cm−1 to account for underestimating phonon energies. 0.33. 12 This yields a Grüneisen parameter of 1.1 ± 0.2, half that of graphene (1.99) 12 and compa- rable to that of hexagonal boron nitride (0.95–1.2). 26,27 The shear deformation potential is 0.78 ± 0.1 for both monolayer and bilayer MoS2 . Next, we investigate the evolution of the band structure of MoS2 with strain through photolu- minescence (PL) spectroscopy. The principal PL peak (A peak) in unstrained direct-gap monolayer MoS2 at 1.82±0.02 eV (Fig. 3a) is due to a direct transition at the K point (Fig. 3a, inset). 2,3 The B peak, due to a direct transition between the conduction band and a lower lying valence band, is obscured in our devices by background PL of polycarbonate/SU8. The PL spectra of unstrained indirect-band gap bilayer MoS2 devices are characterized by a similar A peak at 1.81±0.02 eV that originates from the same direct transition, but that is now less intense as it originates from hot luminescence. In addition, we observe an I peak at 1.53±0.03eV (Fig. 3c), which originates from the transition across the indirect band gap of bilayer MoS2 between the Γ and K points, (Fig. 3c, inset). Applied strain significantly changes the PL spectra (Fig. 3a,c). For all measured monolayer devices, the A peak redshifts approximately linearly with strain, at a rate of 45 ± 7 meV/% strain (Fig. 3b). For bilayer devices, the A and I peaks are redshifted by 53 ± 10 and 129 ± 20 meV/% strain respectively (Fig. 3d). While the intensity of the A peak in monolayer devices decreases to a third of its original size with an applied strain of 2%, in bilayer devices the intensity of this peak is virtually strain-independent (Figs. 4a). 5 Raman Intensity (a.u.)Raman Shift (cm-1)380400420A'1E'+Raman Shift (cm-1)375380385Strain00.010.02E'-E'-E'+a)b)strain0.4%0.8%1.2%1.6%0%Figure 3: Photoluminescence spectra of strained MoS2 (a) PL spectra of a representative mono- layer device as it is strained from 0 to 1.8%. Strain independent PL background was subtracted. (b) Evolution of the position of the A peak of the PL spectrum (Lorentzian fits) with strain for several monolayer devices (colors represent different devices) with GW0 -BSE calculations (dashed line) of expected peak position after 25 meV offset. (c) PL spectra of a representative bilayer device as strain is increased from 0 to 0.6%. (d) PL peak position versus strain for the A and I peaks of bilayer devices (colors represent different devices) with good agreement to our GW0 -BSE calcula- tions (dashed lines). Insets in (a) and (b) contains schematic representations of the band structure for both monolayer and bilayer MoS2 devices that are progressively strained from 0% (black) to ∼5% (maroon) and ∼8% (red). To understand our experimental results, we compare them to the results of GW0 -BSE calcula- tions; details are given in supplementary materials. Crucially, these calculations capture the effect of strong electron-electron interactions in MoS2 leading to the formation of excitons with binding energies significantly exceeding kBT at room temperature. 28,29 This is important because PL spec- troscopy probes the optical band gap, the difference between the fundamental (or transport) band gap and the exciton binding energy. The observed redshift of the PL peaks is indicative of strain-induced reduction of band gaps in both monolayer and bilayer MoS2 . Indeed, our GW0 -BSE calculations for a monolayer predict a reduction of the optical band gap at a rate of ∼59 meV/% strain (dashed line in Fig. 3b), in close agreement with the measured PL peak shift. In bilayer devices, the calculated rates of reduction for the direct (67 meV/% strain) and indirect (94 meV/% strain) optical band gaps (dashed lines in Fig. 3d) are also in close agreement with measured redshift rates for A and I peaks, 53 ± 10 and 6 Intensity (a.u.)PL energy (eV)1.41.61.80 %0.2 %0.4 %0.6 %c)d)IAa)b)A Peak Position (eV)1.71.751.81.85Strain00.010.02KAAIKAIIntensity (a.u.)PL energy (eV)1.61.820%0.6%1.2%1.8%A and I Peak Positions (eV)1.41.51.781.81.82Strain00.0050.01A PeakI Peakd)Figure 4: Direct to indirect band gap transition in MoS2 (a) Evolution of intensity of the A peak of strained monolayer MoS2 (solid shapes) with a fit (dashed curve) to the rate equations consistent with a degenerate direct and indirect optical band gap at 1.3±0.6% strain (supplementary material). PL intensity of bilayer A peak (unfilled circles) with no measurable change in intensity. Each color represents a distinct device. (b) GW0 calculations of the fundamental band gaps of strained monolayer MoS2 , with an expected degeneracy at ∼5% strain. Optical band gap calculated by including the exciton binding energy yields a degeneracy at ∼ 0.1%. 129 ± 20 meV/% strain respectively. We interpret the rapid decrease in PL intensity of monolayer MoS2 with strain as a signature of the anticipated strain-induced transformation of the optical band gap of this material from direct to indirect. 13,14,18,19 Indeed, at zero strain the energy difference between the minimum of the con- duction band at the K point and the local maximum of the valence band at the Γ point (the indirect gap) is higher in energy than the direct gap at the K point (Fig. 3a, inset, black curve). However, we calculate that the indirect gap reduces with strain faster than the direct gap (59 vs. 94 meV/% strain). As a result, if we ignore the effect of excitons, at ε ∼5% the indirect gap overtakes the direct gap and monolayer MoS2 becomes an indirect-gap material. With excitonic effects included, our calculations indicate that the direct and indirect optical gaps (fundamental gaps minus binding energy of corresponding excitons) become degenerate at a much lower strain, 0.1% (Fig. 4b). We however note that the accuracy of this value sensitively depends on the precise binding energy of the direct and indirect excitons that have not yet been measured experimentally. As monolayer MoS2 is strained and transitions from a direct to an indirect band gap material, we expect a marked decrease in the intensity of the A peak, as a majority of the excitons would not reside in this higher energy excitonic state, in good agreement the decrease in intensity in Fig. 4a. Quantitatively, a simple model describing direct and indirect excitons in monolayer MoS2 as a two-level system yields an acceptable fit to our experimental data (dashed curve in Fig. 4a), with a 7 EindirectEdirectEdirect-EexcitonFundamental GapOptical GapEindirect-Eexcitonb)a)bilayermonolayerNormalized Integrated Intensity00.51Strain00.010.02Band Gap Energy (eV)22.5Strain00.010.02direct-to-indirect band gap transition at 1.3 ± 0.6% strain (details in supplementary information). The observed PL spectra warrant two more comments. First, while for strains where the indirect band gap of monolayer MoS2 is lower in energy than the direct band gap, we do not observe a peak corresponding to an indirect transition in its PL spectrum. This is likely due to the much smaller intensity of the indirect photoluminescence compared to the intensity of hot luminescence of the A peak. Second, in the range of strains from 1.3–5%, monolayer MoS2 enters a curious regime where its fundamental band gap is direct, while the optical band gap is indirect. In conclusion, we have observed strain-induced phonon softening, band gap modulation and a transition from an optically direct to an optically indirect material in strained MoS2 samples. These observations support a view of strain engineering as an enabling tool to both explore novel physics in MoS2 (and other two-dimensional transition metal dichalcogenides such as MoSe2 , WS2 , WSe2 ) and to tune its optical and electronic properties. An interesting avenue of research would be to explore the regime of degenerate direct and indirect bands – that play key roles in a plethora of spin-related properties of the MoS2 . 30,31 Among the potential applications of the strain-dependent photoluminescence of MoS2 and its cousins are nanoscale stress sensors and tunable photonic devices – LEDs, photodetectors, and electro-optical modulators. Note: While the manuscript was under review, two studies reporting modification of the band- gap with strain became available. 32,33 The measured band gap shifts with strain are in general agreement with our results, however due to the much lower range of strains employed in the devices of Ref. [32,33] they do not measure the Grüneisen parameter, probe the direct-to-indirect optical band gap transition, or observe contribution of excitonic effects. Acknowledgement This research was supported by NSF DMR-1056859, NSF EPS-1004083, and ONR N000141310299. BW and JZ were supported by DTRA HDTRA1-1-10-1-0047. We thank John Fellenstein for help in designing the four point bending apparatus, Branton Campbell for teaching us about phonon naming conventions, and Ashwin Ramasubramaniam for discussions about the first-principles cal- 8 culations. Supporting Information Available: Supporting information, including bilayer Raman data, de- tails of first principles calculations, and a simple two-level model to calculate the intensity of strained MoS2 , is available free of charge via the Internet at http://pubs.acs.org. References (1) Wang, Q. H.; Kalantar-Zadeh, K.; Kis, A.; Coleman, J. N.; Strano, M. S. Electronics and optoelectronics of two-dimensional transition metal dichalcogenides. Nature Nanotechnology 2012, 7, 699–712. (2) Mak, K.; Lee, C.; Hone, J.; Shan, J.; Heinz, T. Atomically Thin MoS_{2}: A New Direct-Gap Semiconductor. Physical Review Letters 2010, 105, 2–5. (3) Splendiani, A.; Sun, L.; Zhang, Y.; Li, T.; Kim, J.; Chim, C.-Y.; Galli, G.; Wang, F. Emerging photoluminescence in monolayer MoS2. Nano Letters 2010, 10, 1271–5. (4) Radisavljevic, B.; Radenovic, a.; Brivio, J.; Giacometti, V.; Kis, a. Single-layer MoS2 tran- sistors. Nature Nanotechnology 2011, 6, 147–50. (5) Lin, M.-W.; Liu, L.; Lan, Q.; Tan, X.; Dhindsa, K. S.; Zeng, P.; Naik, V. M.; Cheng, M. M.-C.; Zhou, Z. Mobility enhancement and highly efficient gating of monolayer MoS 2 transistors with polymer electrolyte. Journal of Physics D: Applied Physics 2012, 45, 345102. (6) Yin, Z.; Li, H.; Li, H.; Jiang, L.; Shi, Y.; Sun, Y.; Lu, G. Single-layer MoS2 phototransistors. ACS Nano 2012, 6, 74–80. (7) Sundaram, R. S.; Engel, M.; Lombardo, A.; Krupke, R.; Ferrari, A. C.; Avouris, P.; Steiner, M. Electroluminescence in Single Layer MoS2. Nano Letters 2013, 13, 1416–1421. (8) Newaz, A.; Prasai, D.; Ziegler, J.; Caudel, D.; Robinson, S.; Jr., R. H.; Bolotin, K. Electrical control of optical properties of monolayer MoS2. Solid State Communications 2013, 155, 49 – 52. 9 (9) Ross, J. S.; Wu, S.; Yu, H.; Ghimire, N. J.; Jones, A. M.; Aivazian, G.; Yan, J.; Man- drus, D. G.; Xiao, D.; Yao, W.; Xu, X. Electrical control of neutral and charged excitons in a monolayer semiconductor. Nature communications 2013, 4, 1474. (10) Mak, K. F.; He, K.; Lee, C.; Lee, G. H.; Hone, J.; Heinz, T. F.; Shan, J. Tightly bound trions in monolayer MoS2. Nature materials 2013, 12, 207–11. (11) Huang, M.; Yan, H.; Chen, C.; Song, D.; Heinz, T. F.; Hone, J. Phonon softening and crys- tallographic orientation of strained graphene studied by Raman spectroscopy. Proceedings of the National Academy of Sciences of the United States of America 2009, 106, 7304–8. (12) Mohiuddin, T.; Lombardo, a.; Nair, R.; Bonetti, a.; Savini, G.; Jalil, R.; Bonini, N.; Basko, D.; Galiotis, C.; Marzari, N.; Novoselov, K.; Geim, a.; Ferrari, a. Uniaxial strain in graphene by Raman spectroscopy: G peak splitting, Grüneisen parameters, and sample orientation. Physical Review B 2009, 79, 1–8. (13) Lu, P.; Wu, X.; Guo, W.; Zeng, X. C. Strain-dependent electronic and magnetic properties of MoS(2) monolayer, bilayer, nanoribbons and nanotubes. Physical chemistry chemical physics : PCCP 2012, 14, 13035–40. (14) Pan, H.; Zhang, Y.-W. Tuning the Electronic and Magnetic Properties of MoS2 Nanoribbons by Strain Engineering. The Journal of Physical Chemistry C 2012, 116, 11752–11757. (15) Yue, Q.; Kang, J.; Shao, Z.; Zhang, X.; Chang, S.; Wang, G.; Qin, S.; Li, J. Mechanical and electronic properties of monolayer MoS2 under elastic strain. Physics Letters A 2012, 376, 1166–1170. (16) Li, T. Ideal strength and phonon instability in single-layer MoS2 . Phys. Rev. B 2012, 85, 235407. (17) Scalise, E.; Houssa, M.; Pourtois, G.; Afanas’ev, V.; Stesmans, A. Strain-induced semicon- 10 ductor to metal transition in the two-dimensional honeycomb structure of MoS2. Nano Re- search 2012, 5, 43–48. (18) Shi, H.; Pan, H.; Zhang, Y.-W.; Yakobson, B. I. Quasiparticle band structures and optical properties of strained monolayer MoS2 and WS2 . Phys. Rev. B 2013, 87, 155304. (19) Feng, J.; Qian, X.; Huang, C.; Li, J. Strain-engineered artificial atom as a broad-spectrum solar energy funnel. Nature Photonics 2012, 6, 2–8. (20) Bertolazzi, S.; Brivio, J.; Kis, A. Stretching and breaking of ultrathin MoS2. ACS Nano 2011, 5, 9703–9. (21) Novoselov, K. S.; Jiang, D.; Schedin, F.; Booth, T. J.; Khotkevich, V. V.; Morozov, S. V.; Geim, A. K. Two-dimensional atomic crystals. Proceedings of the National Academy of Sci- ences of the United States of America 2005, 102, 10451–10453. (22) Li, H.; Zhang, Q.; Yap, C. C. R.; Tay, B. K.; Edwin, T. H. T.; Olivier, A.; Baillargeat, D. From Bulk to Monolayer MoS2: Evolution of Raman Scattering. Advanced Functional Materials 2012, 22, 1385–1390. (23) Chen, C.; Rosenblatt, S.; Bolotin, K. I.; Kalb, W.; Kim, P.; Kymissis, I.; Stormer, H. L.; Heinz, T. F.; Hone, J. Performance of monolayer graphene nanomechanical resonators with electrical readout. Nature Nanotechnology 2009, 4, 861–7. (24) Lee, C.; Yan, H.; Brus, L.; Heinz, T.; Hone, J.; Ryu, S. Anomalous lattice vibrations of single-and few-layer MoS2. ACS nano 2010, 4, 2695–2700. (25) Rice, C.; Young, R. J.; Zan, R.; Bangert, U.; Wolverson, D.; Georgiou, T.; Jalil, R.; Novoselov, K. S. Raman-scattering measurements and first-principles calculations of strain- induced phonon shifts in monolayer MoS_{2}. Physical Review B 2013, 87, 081307. (26) Kern, G.; Kresse, G.; Hafner, J. Ab initio calculation of the lattice dynamics and phase dia- gram of boron nitride. Physical Review B 1999, 59, 8551–8559. 11 (27) Sanjurjo, J. A.; López-Cruz, E.; Vogl, P.; Cardona, M. Dependence on volume of the phonon frequencies and the ir effective charges of several III-V semiconductors. Phys. Rev. B 1983, 28, 4579–4584. (28) Cheiwchanchamnangij, T.; Lambrecht, W. R. Quasiparticle band structure calculation of monolayer, bilayer, and bulk MoS_{2}. Physical Review B 2012, 85, 1–4. (29) Ramasubramaniam, A. Large excitonic effects in monolayers of molybdenum and tungsten dichalcogenides. Phys. Rev. B 2012, 86, 115409. (30) Mak, K. F.; He, K.; Shan, J.; Heinz, T. F. Control of valley polarization in monolayer MoS(2) by optical helicity. Nature Nanotechnology 2012, 7, 494–498. (31) Zeng, H.; Dai, J.; Yao, W.; Xiao, D.; Cui, X. Valley polarization in MoS2 monolayers by optical pumping. Nature Nanotechnology 2012, 7, 490–3. (32) He, K.; Poole, C.; Mak, K. F.; Shan, J. Experimental Demonstration of Continuous Electronic Structure Tuning via Strain in Atomically Thin MoS2. Nano Letters 2013, 13, 2931–2936. (33) Wang, G.; Zhu, C.; Liu, B.; Marie, X.; Feng, Q.; Wu, X.; Fan, H.; Tan, P.; Amand, T.; Urbaszek, B. Strain tuning of optical emission energy and polarization in monolayer and bilayer MoS2. arXiv preprint 2013, arXiv:1306.3442. 12 

1702.06210

2

1702

Valley dependent anisotropic spin splitting in silicon quantum dots

Spin qubits hosted in silicon (Si) quantum dots (QD) are attractive due to their exceptionally long coherence times and compatibility with the silicon transistor platform. To achieve electrical control of spins for qubit scalability, recent experiments have utilized gradient magnetic fields from integrated micro-magnets to produce an extrinsic coupling between spin and charge, thereby electrically driving electron spin resonance (ESR). However, spins in silicon QDs experience a complex interplay between spin, charge, and valley degrees of freedom, influenced by the atomic scale details of the confining interface. Here, we report experimental observation of a valley dependent anisotropic spin splitting in a Si QD with an integrated micro-magnet and an external magnetic field. We show by atomistic calculations that the spin-orbit interaction (SOI), which is often ignored in bulk silicon, plays a major role in the measured anisotropy. Moreover, inhomogeneities such as interface steps strongly affect the spin splittings and their valley dependence. This atomic-scale understanding of the intrinsic and extrinsic factors controlling the valley dependent spin properties is a key requirement for successful manipulation of quantum information in Si QDs.

cond-mat.mes-hall

cond-mat

Valley dependent anisotropic spin splitting in silicon quantum dots Rifat Ferdous,1 Erika Kawakami,2 Pasquale Scarlino,2 Micha l P. Nowak,2, 3 D. R. Ward,4 D. E. Savage,4 M. G. Lagally,4 S. N. Coppersmith,4 Mark Friesen,4 Mark A. Eriksson,4 Lieven M. K. Vandersypen,2 and Rajib Rahman1 1Network for Computational Nanotechnology, Purdue University, West Lafayette, IN 47907, USA 2QuTech and Kavli Institute of Nanoscience, TU Delft, Lorentzweg 1, 2628 CJ Delft, The Netherlands 3AGH University of Science and Technology, Faculty of Physics and Applied Computer Science, al. Mickiewicza 30, 30-059 Krak´ow, Poland 4University of Wisconsin-Madison, Madison, Wisconsin 53706, USA (Dated: August 18, 2017) ABSTRACT Spin qubits hosted in silicon (Si) quantum dots (QD) are attractive due to their exceptionally long coherence times and compatibility with the silicon transistor plat- form. To achieve electrical control of spins for qubit scal- ability, recent experiments have utilized gradient mag- netic fields from integrated micro-magnets to produce an extrinsic coupling between spin and charge, thereby elec- trically driving electron spin resonance (ESR). However, spins in silicon QDs experience a complex interplay be- tween spin, charge, and valley degrees of freedom, influ- enced by the atomic scale details of the confining inter- face. Here, we report experimental observation of a valley dependent anisotropic spin splitting in a Si QD with an integrated micro-magnet and an external magnetic field. We show by atomistic calculations that the spin-orbit interaction (SOI), which is often ignored in bulk silicon, plays a major role in the measured anisotropy. Moreover, inhomogeneities such as interface steps strongly affect the spin splittings and their valley dependence. This atomic- scale understanding of the intrinsic and extrinsic factors controlling the valley dependent spin properties is a key requirement for successful manipulation of quantum in- formation in Si QDs. INTRODUCTION How microscopic electronic spins in solids are af- fected by the crystal and interfacial symmetries has been a topic of great interest over the past few decades and has found potential applications in spin-based elec- tronics and computation1–7. While the coupling be- tween spin and orbital degrees of freedom has been ex- tensively studied, the interplay between spin and the momentum space valley degree of freedom is a topic of recent interest. This spin-valley interaction is ob- served in the exotic class of newly found two-dimensional materials8–10, in carbon nanotubes11 and in silicon12–14 -the old friend of the electronics industry. Progress in silicon qubits in the last few years has come with the demonstrations of various types of qubits with exception- ally long coherence times, such as single spin up/down qubits15,16, two-electron singlet-triplet qubits17,18, three- electron exchange-only19 and hybrid spin-charge qubits20 and also hole spin qubits21 realized in Si QDs. The pres- ence of the valley degree of freedom has enabled valley based qubit proposals22 as well, which have potential for noise immunity. To harness the advantages of different qubit schemes, quantum gates for information encoded in different bases are required9,23,24. A controlled coherent interaction between multiple degrees of freedom, like val- ley and spin, might offer a building block for promising hybrid systems. Although bulk silicon has six-fold degenerate conduc- tion band minima, in quantum wells or dots, electric fields and often in-plane strain in addition to vertical confinement results in only two low lying valley states (labeled as v− and v+ in Figure 1b) split by an energy gap known as the valley splitting. An interesting inter- play between spin and valley degrees of freedom, which gives rise to a valley dependent spin splitting, has been observed in recent experiments15,25–27. SOI enables the control of spin resonance frequencies by gate voltage, an effect measured in refs. [16, 25]. However, the ESR fre- quencies and their Stark shifts were found to be different for the two valley states25. In another work, an inho- mogeneous magnetic field, created by integrated micro- magnets in a Si/SiGe quantum dot device, was used to electrically drive ESR15. Magnetic field gradients gener- ated in this way act as an extrinsic spin-orbit coupling and thus can affect the ESR frequency28. Remarkably, although SOI is a fundamental effect arising from the crystalline structure, the ESR frequency differences be- tween the valley states observed in refs. [15] and [25] have different signs when the external fields are oriented in the same direction with respect to the crystal axes. To un- derstand and achieve control over the coupled behavior between spin and valley degrees of freedom, several ques- tions need to be addressed, such as 1) What causes the device-to-device variability?, 2) Can an artificial source of interaction, like inhomogeneous B-field, completely over- power the SOI effects of the intrinsic material?, 3) What knobs and device designs can be utilized to engineer the valley dependent spin splittings? RESULTS Here we report experimentally measured anisotropy in and fv+ and the ESR frequencies of the valley states fv− arXiv:1702.06210v2 [cond-mat.mes-hall] 17 Aug 2017 2 FIG. 1. Valley dependent anisotropic ESR in a Si QD with integrated micro-magnets a, False-color image of the experimental device showing the estimated location of the quantum dot (magenta colored circle) and two Co micro-magnets (green semi- transparent rectangles). The external magnetic field (Bext) was rotated clockwise in-plane, from the [110] (θ = 0◦) crystal orientation towards [1¯10] (θ = 90◦). b, Lowest energy levels of a Si QD in an external magnetic field. The valley-split levels v− and v+ are found to have unequal Zeeman splittings (EZS(v±) = hfv±), with ESR frequencies fv− 6= fv+ . In the experiment, all the measured spin splittings are much larger than the valley splitting and are therefore above the anticrossing point of the spin and valley states. c, Both measured (red circles) and calculated fv− − fv+ as a function of θ, for Bext =0.8 T. The anisotropy in fv− − fv+ is governed by both internal (intrinsic SOI) and external (micro-magnetic field) factors. The anisotropy due to the intrinsic SOI, calculated from atomistic tight binding method, for a specifically chosen vertical electric field and interface step configuration, is labeled as 'Bext (TB)'. The micro-magnetic field is separated into a homogeneous (Bθ micro) and an inhomogeneous (∆Bθ) part. The inclusion of Bθ micro (TB)'), shifts the curve away from micro + ∆Bθ (TB)') and shifts the the experiment. The addition of ∆Bθ introduces additional anisotropy (labeled 'Bext + Bθ curve towards the experiment. An effective-mass calculation, with fitted SOI and dipole coupling parameters, is also presented , as a function of θ, for Bext =0.8 T. Calculation with a cyan solid line. d, Both measured (red circles) and calculated fv− with the intrinsic SOI shows negligible change in GHz scale, while the addition of Bθ micro results in anisotropy close to the experimental data. ∆Bθ has negligible effect on fv− is mainly dictated by the homogeneous part of the micro-magnetic field. micro in this case (labeled 'Bext + Bθ . Hence, the anisotropy of fv− their differences fv− − fv+, as a function of the direc- tion of the external magnetic field (Bext) in a quantum dot formed at a Si/SiGe heterostructure with integrated micro-magnets. At specific angles of the external B-field, we also measure the spin splittings of the two valley states as a function of the B-field magnitude. By performing spin-resolved atomistic tight binding (TB) calculations of the quantum dots confined at ideal versus non-ideal interfaces, we evaluate the contribution of the intrinsic SOI with and without the spatially varying B-fields from the micro-magnets to the spin splittings, thereby relating these quantities to the microscopic nature of the inter- face and elucidating how spin, orbital and valley degrees of freedom are intertwined in these devices. Finally, by combining all the effects together, we explain the exper- imental measurements and address the questions raised in the previous paragraph. Fig. 1 shows the experimental device, energy levels of interest and measured anisotropic spin splittings com- pared with the final theoretical results. Details of the device shown in Fig. 1a and the measurement technique of the spin resonance frequency can be found in ref. [15]. The external magnetic field is swept from the [110] to [1¯10] crystal orientation. A schematic of the energy level structure is shown in Fig. 1b depicting the v− and v+ valley states with different spin splittings, where v− is de- fined as the ground state. In the experiment, the lowest valley-orbit excitation is well below the next excitation, justifying this four-level schematic in the energy range of interest. The atomistic calculation with SOI alone (labeled 'Bext (TB)') for a QD at a specifically chosen non-ideal interface and vertical electric field (Ez) qualitatively cap- tures the experimental trend of fv− − fv+ in Fig. 1c, but fails to reproduce the anisotropy of the measured fv− in Fig. 1d in the larger GHz scale. The differences between the experimental data and the SOI-only calculations in both figures arise from the micro-magnets present in the experiment. We can separate the contribution from the micro-magnet into two parts, a homogeneous (spatial av- erage, Bθ micro) and an inhomogeneous (spatially varying, ∆Bθ) magnetic field. The superscript θ here indicates that the micro-magnetic fields depend on the direction of Bext (supplementary section S4). The inclusion of the homogeneous part of the micro-magnetic field creates an anisotropy in the total magnetic field (supplementary Fig. S7), which captures the anisotropy of fv− in Fig. 1d very well (fv− ≈ gµ(cid:12)(cid:12)Bext + Bθ the Land´e g-factor, µ is the Bohr magneton and h is the Planck constant), but quantitative match with the exper- micro(cid:12)(cid:12) /h, where g is Bext+Bmicro+∆B(TB) Bext(TB) exp Bext+Bmicro+∆B(EM) Bext(TB) Bext+Bmicro(TB) exp exp Bext (TB) exp micro (TB) Bext+Bθ Bext+Bmicro+∆B(TB) Bext+Bmicro+∆B(EM) Bext+Bmicro+∆B(EM) Bext+Bmicro(TB) micro+ΔBθ (EM) Bext+Bθ micro+ΔBθ (TB) Bext+Bθ Bext+Bmicro+∆B(TB) Bext(TB) Bext+Bmicro+∆B(TB) Bext+Bmicro(TB) Bext(TB) Bext(TB) Bext+Bmicro(TB) Bext+Bmicro(TB) 23 23 d 30 19 19 60 0 0 θ ◦ 90 30 30 θ° θ ◦ 60 60 90 90 22 22 21 21 20 20 fv−(GHz) fv- (GHz) 024 −2 −4 0 90 90 90 fv−−fv+(MHz) 024 60 60 60 −2 fv−−fv+(MHz) θ ◦ 30 θ° θ ◦ 024 −2 −4 0 30 30 4 024 2 fv−−fv+(MHz) −2 -2 −4 -4 0 0 0 c fv- - fv+ (MHz) fv−−fv+(MHz) Energy hfv+ 024 024 −2 −2 −4 hfv- fv−−fv+(MHz) fv−−fv+(MHz) B-field a y,[010] [110] Bext θ ϕ 400nm b v+ v- x,[100] - [110] 3 dBext On the other hand, comparing the calculated fv− −fv+ from SOI alone (labeled 'Bext (TB)') for the chosen Ez and interface condition, with experimental data, in both the top and bottom panels of Fig. 2a, it is clear that the experimental B-field dependence of fv− − fv+ (the slope, d(fv−−fv+ ) ) is captured from the effect of intrinsic SOI, except for a shift between the SOI curve and the experi- mental data (different shift for θ = 0◦ and θ = 90◦). The addition of Bmicro alone does not result in the necessary shift to match the experiment. Only after adding ∆B can a quantitative match with the experiment be achieved. Again the experiment-theory agreement is conditional on the interface condition and Ez. Moreover, we see that the addition of ∆B does not change the dependency on Bext. Therefore, to properly explain the observed experi- mental behavior, we can ignore neither the SOI, which is responsible for the change in fv− −fv+ with Bext, nor the inhomogeneous B-field which shifts fv− − fv+ regardless of Bext. DISCUSSION dBext d(fv−−fv+ ) To obtain a quantitative agreement between the exper- iment and the atomistic TB calculations, simultaneously in the anisotropy (Fig. 1c) and the Bext (Fig. 2a) depen- dence of fv− − fv+, the only knobs we have to adjust are 1) Ez and 2) interfacial geometry i.e. how many atomic steps at the interface lie inside the dot and where they are located relative to the dot center. These adjust- ments have to be done iteratively since the steps and Ez not only affect the intrinsic SOI but also the influence of the inhomogeneous B-field. It is easy to separate out the contribution of the SOI from the micro-magnet in the Bext dependence of fv− − fv+. It will be shown in Figs. 3 and 4 that, the slope, originates from the SOI, while the micro-magnetic field shifts fv− − fv+ independent of Bext. First we individually match the ex- perimental "slope" from the SOI and the "shift" from the contribution of the micro-magnet for some combinations of the two knobs. Finally both effects together quantita- tively match the experiment for Ez = 6.77 MVm−1, and an interface with four evenly spaced monoatomic steps at -24.7 nm, -2.9 nm, 18.7 nm, 40.4 nm from the dot center along the x ([100]) direction. This combination also pre- dicts a valley splitting of 34.4 µeV in close agreement with the experimental value, given by 29 µeV27. To describe the QD, a 2D simple harmonic (parabolic confinement) potential was used with orbital energy splittings of 0.55 meV and 9.4 meV characterizing the x and y ([010]) con- finement respectively. As the interface steps are parallel to y direction, the orbital energy splitting along y has negligible effects, but the strong y confinement signifi- cantly reduces simulation time. To further our understanding, we have complemented the atomistic calculations with an effective mass (EM) based analytic model with Rashba and Dresselhaus-like FIG. 2. Measured ESR frequencies, (fv± ) and their dif- ferences for the two valley states as a function of the ex- ternal B-field magnitude Bext along two crystal directions, and comparison with theoretical calculations. a, fv− − fv+ with Bext along [110] (θ = 0◦) (bottom panel) and b, fv− and [1¯10] (θ = 90◦) (top panel). As in Figs. 1c and 1d, the calculations progressively include SOI (labeled 'Bext (TB)'), homogeneous (labeled 'Bext + Bmicro (TB)'), and gradient (labeled 'Bext + Bmicro + ∆B (TB)') B-field of the micro- magnet. The cyan solid lines are the effective mass calcu- lations and the red circles are the experimental data. The dependence (slope) of fv− − fv+ on Bext in (a) comes from the SOI, while the micro-magnetic fields provide a shift inde- pendent of Bext. imental data in Fig. 1c is not obtained. Next, we also in- corporate the inhomogeneous part of the micro-magnetic field, and witness a close quantitative agreement in the anisotropy of fv− − fv+, while the anisotropy of fv− is unaffected. This experiment-theory agreement of fig. 1c is achieved for a specific choice of interface condition and Ez, whose influence will be discussed later. Here, we conclude that mainly the intrinsic SOI and the extrinsic inhomogeneous B-field govern the anisotropy of fv−−fv+ on the MHz scale, while the anisotropy in the total homo- geneous magnetic field introduced by the micro-magnet dictates the anisotropy of fv− (and fv+) on the larger GHz scale. In Fig. 2, we show the measurements of the spin split- tings as a function of the magnitude of Bext (Bext), to- gether with the theoretical calculations. The bottom panels show fv− − fv+ (Fig. 2a) and fv− (Fig. 2b) for Bext along [110] (θ = 0◦), whereas the top pan- els correspond to the B-field along [1¯10] (θ = 90◦). In Fig. 2b, fv− depends on Bext through g−µBtot/h, with Btot = Bext + Bmicro. The addition of Bmicro causes a change in Btot and shifts fv− to coincide with the ex- perimental data. The contributions of ∆B and SOI are negligible here in the GHz scale. - 1.3 θ=90°, [110] Bext+Bmicro+∆B(TB) Bext(TB) Bext(TB) Bext+Bmicro(TB) exp Bext (TB) Bext+Bmicro (TB) Bext+Bmicro+∆B(EM) Bext+Bmicro(TB) Bext+Bmicro+ΔB (EM) Bext+Bmicro+∆B(TB) 40 b 40 Bext(TB) 30 Bext+Bmicro(TB) 30 20 20 10 10 40 40 30 30 θ ◦ 20 20 θ=0°, [110] 0.5 0.7 0.9 1.1 1.3 1.5 90 0.5 −2 −4 0 90 fv−(GHz) fv- (GHz) 024 30 60 30 60 90 0.9 Bext (T) B e x t(T) θ ◦ fv−−fv+(MHz) - Bext+Bmicro+∆B(EM) exp exp Bext+Bmicro+∆B(TB) Bext+Bmicro+∆B(EM) Bext+Bmicro+ΔB (TB) Bext(TB) Bext+Bmicro+∆B(TB) Bext+Bmicro(TB) Bext(TB) Bext+Bmicro(TB) θ=90°, [110] θ ◦ 024 fv−−fv+(MHz) 0 0 −2 -2 −4 -4 −6 -6 60 −8 -8 0.5 0.7 0.9 1.1 1.3 1.5 θ ◦ 60 0.5 θ ◦ θ=0°, [110] 90 90 0.9 Bext (T) B e x t(T) 60 −2 −4 0 1.3 −2 −4 0 30 fv- - fv+ (MHz) fv−−fv+(MHz) 30 30 a 024 2468 6 8 4 fv−−fv+(MHz) 2 4 FIG. 3. Effect of the intrinsic SOI on fv± in a Si QD. a, Calculated fv± as a function of θ, in a QD with ideal (flat) interface, for Bext =0.8 T, without any micro-magnet. The anisotropies in these curves are in the MHz range and will appear flat on a GHz scale, like the SOI line (labeled Bext (TB)) of Fig. 1d. b, Schematic of a QD wavefunction near a monoatomic step at the interface. The distance between the dot center and the step edge is denoted by x0. c, Computed Dresselhaus parameters β± as a function of x0. β± changes sign between the two sides of the step. d, QD wave-functions subjected to multiple interface steps. Four different cases are shown (c1(5), c2(3), c3(4), c4(4)) that are used in Figs. 3e, 3f and also in Figs. 4c and 4d. The number in parentheses is the total number of steps within the QD. Though c3 and c4 has the same number of steps, the location of the steps are different. c3 is the step configuration used in Figs. 1 and 2. e, Calculated fv− − fv+ as a function of θ, for different interface conditions, for Bext =0.8 T. Interface steps affect both the magnitude and sign of fv− − fv+ . f, fv− − fv+ with respect to Bext along [110](θ = 0◦)/ [¯1¯10] (θ = 180◦) (bottom panel) and [1¯10] (θ = 90◦)/[¯110] (θ = 270◦) (top panel). fv− − fv+ for c3 (red lines with circular marker), in both Figs. 3e and 3f, corresponds to the SOI lines (blue dashed lines with diamond marker) of Figs. 1c and 2a. The parabolic confinement and Ez used here are the same as that of Figs. 1 and 2, except for Fig. 3c, where a smaller dot (with a parabolic confinement in both x and y corresponding to orbital energy splitting of 9.4 meV) is used to accommodate for large variation in dot location. SOI terms [supplementary section S1], as used in ear- lier works25,30–32. We have also developed an analytic model to capture the effects of the inhomogeneous mag- netic field [supplementary section S2]. The contributions of the SOI and ∆B on fv− − fv+ obtained from these models are shown in equations 1 and 2 respectively. gµ ≈ ∆(cid:0)fv− − fv+(cid:1)∆B h ( cos φ(cid:18)(hx−i − hx+i) + sin φ (hx−i − hx+i) dBφ y dx dBφ x dx + (hy−i − hy+i) dBφ y + (hy−i − hy+i) dBφ x dy (cid:19) dy !) ∆(cid:0)fv− − fv+(cid:1)SOI 4π ehzi ≈ h2 Bextn (β− − β+) sin 2φ − (α− − α−)o (1) (2) Here, α± and β± are the Rashba and Dresselhaus-like coefficients respectively, hzi is the spread of the electron wavefunction along z, hx±i and hy±i are the intra-valley dipole matrix elements, φ is the angle of the external magnetic field with respect to the [100] crystal orien- 0 45 90135180225270315 0 45 90135180225270315 −75 −75 −80 −80 −85 −85 fv(GHz) fv(GHz) v− v− β- v+ β+ v+ −5 -5 −10 -10 −20 −10 -10 -20 0 0 10 20 10 20 x0 (nm) x 0( n m ) 5 05 0 c 10 10 β(10−15eV·m) β (10-15 eV m) . b β1 x0 electron wave- function β2 βeff = b1β1 + b2β2 ; β2 = -β1 0 45 90135180225270315 0 45 90135180225270315 - θ ◦ θ ◦ - [110] - [110] −50 −50 −55 [110] −55 −60 −60 −65 −65 −70 −70 −75 −75 −80 −80 −85 −85 −50 −55 -55 −60 −65 -65 −70 −75 -75 −80 −85 -85 v− v− v- 0 45 90135180225270315 0 v+ v+ v+ 180 270 θ° θ ◦ fv(GHz) fv(GHz) [110] 90 - c1(5) ideal −30 0 30 −30 0 30 30 -30 0 c2(3) c1(5) −30 0 30 −30 0 30 30 -30 0 ideal ideal e c2(3) c1(5) ideal c1(5) c1(5) c4(4) c3(4) c3(4) ideal c3(4) c3(4) c4(4) c4(4) c4(4) c1(5) 0 0.9 Bext (T) B e x t( T ) −20 90°/270° 0°/180° 20 fv−−fv+(MHz) fv−−fv+(MHz) fv- - fv+ (MHz) c2(3) c2(3) c3(4) c2(3) c1(5) c2(3) f 30 30 0 0 −30 -30 −60 -60 60 60 30 30 0 0 −30 -30 90 90 180 270 180 270 θ° θ ◦ 0.5 0.7 0.9 1.1 1.3 1.5 0.5 1.3 ideal 30 30 20 20 10 10 0 0 −10 20 -10 −20 -20 −30 0 -30 0 0 fv−−fv+(MHz) fv- - fv+ (MHz) fv−−fv+(MHz) 0 20 20 0 fv−−fv+(MHz) fv−−fv+(MHz) 0 x ([100]) (nm) −20 −20 −20 −20 fv−−fv+(MHz) c3(4) −30 0 30 30 -30 0 −30 0 30 20 c4(4) −30 0 30 30 -30 −30 0 30 0 fv - gµBext (MHz) a fv(GHz) v− v+ d z ([001]) i dj tation and dBφ are the magnetic field gradients along different directions (i, j = x, y, z) for a specific angle φ. It is clear from these expressions that to match fv− − fv+ the difference in SOI and dipole moment pa- rameters between the valley states are relevant (but not their absolute values). The parameters used to match the experiment are α− − α+ = −2.5370 × 10−15 eV·m, β− − β+ = 9.4564 × 10−19 eV·m, hx−i − hx+i = −0.169 nm, hy−i − hy+i = 0 nm and hzi = 2.792 nm. These fitting parameters in the EM calculations enable us to obtain an even better match with the experimental data compared to TB in Figs. 1c and 2a (cyan solid lines). Here we want to point out that the accurateness of the numerically calculated micro-magnetic field values de- pends on our estimation of the dot location. But as we calculate (β− − β+) and (α− − α−) independently by comparing the measured for [110] and [1¯10] with equation 1(supplementary section S5), any uncertainty in the estimated dot location or the micro- magnetic field values does not effect the extracted SOI parameters. d(fv−−fv+ ) dBext As shown in Figs. 1 and 2, three physical attributes play a key role in explaining the experimental data, 1) SOI, 2) Bmicro, and 3) ∆B. Each of these contribute to fv± , and only their sum can accurately reproduce the experimental data for a specific interface condition and vertical electric field, the two knobs mentioned in ear- lier paragraph. In Figs. 3 and 4, we show separately the effects of 1) and 3) respectively. We show how the contributions of SOI and ∆B are modulated by the na- ture of the confining interface (knob 2). The influence of Ez (knob 1) on the effects of SOI and ∆B are shown in the supplementary Figs. S3 and S4 respectively. We also show how Bmicro modifies the total homogeneous B-field in the supplementary Fig. S7. Fig. 3a shows the angular dependence of fv± for a Si QD with a smooth interface calculated from TB. and fv+ show a 180◦ periodicity but they are Both fv− 90◦ out of phase. From analytic effective mass study [supplementary equation S11], we understand that the anisotropic contribution from the Dresselhaus-like inter- action, caused by interface inversion asymmetry32, re- sults in this angular dependence in fv± . Moreover, the different signs of the Dresselhaus coefficients β± for the valley states, give rise to a 90◦ phase shift between fv− and fv+. It is important to notice that the change in fv± is in MHz range. So, in GHz scale, like the blue curve (diamond markers) in Fig. 1 d, this change is not visible. However, if we compare fv− and fv+ for this ideal inter- face case, we see fv− < fv+ at θ =90◦, which does not explain the experimentally mea- sured anisotropy. We now discuss the remaining physical parameters needed to obtain a complete understanding of the experiment. > fv+ at θ =0◦ and fv− It is well-known that the interface between Si/SiGe or Si/SiO2 has atomic-scale disorder, with monolayer atomic steps being a common form of disorder33. To 5 understand how such non-ideal interfaces can affect SOI, we first introduce a monolayer atomic step as shown in Fig. 3b and vary the dot position laterally relative to the step, as defined by the variable x0. By fitting the EM solutions to the TB results [supplementary equation S15], we have extracted the Dresselhaus-like coefficient β± and plotted it in Fig. 3c as a function of x0. It is seen that β± changes sign as the dot moves from the left to the right of the step edge. Both the sign and magnitude of β± depends on the distribution of the wavefunction between the neighboring regions with one atomic layer shift between them, as shown in fig. 3b. A monoatomic shift of the vertical position of the interface results in a 90◦ rotation of its atomic arrangements about the [001] axis, which results in a sign inversion of the Dresselhaus coefficient of that region32. A dot wavefunction spread over a monoatomic step therefore samples out a weighted average of two βs with opposite signs30,31. Next, we investigate the anisotropy of fv− − fv + (Fig. 3e) with various step configurations shown in Fig. 3d. fv− − fv + in Fig. 3e exhibits a 180◦ periodicity, with ex- trema at the [110], [1¯10], [¯1¯10], [¯110] crystal orientations. Both the sign and magnitude of fv− − fv + depends on the interface condition. Since β± decreases when a QD wavefunction is spread over a step edge, the smooth inter- face case (green curve) has the highest amplitude. Fig. 3f shows that the slope of fv− − fv + with Bext changes sign for a 90◦ rotation of Bext and is strongly dependent on the step configuration. The step configuration labeled c3 in Fig. 3d is used to match the experiment in Figs. 1 and 2. So the curves for c3 in both Figs. 3e and 3f correspond to the SOI results of Figs. 1c and 2a. It is key to note here that, as Ez also influences(cid:12)(cid:12)fv− − fv+(cid:12)(cid:12) and(cid:12)(cid:12)(cid:12) shown in supplementary Fig. S3, a different combination of interface steps and Ez can also produce these same SOI results of Figs. 1c and 2a, but might not result in the necessary contribution from micro-magnet to match the experiment. Now the dependence of fv− − fv+ on the interface condition will cause device-to-device variability, while the dependence on the direction and magnitude of Bext can provide control over the difference in spin split- tings. These results thus give us answers to the questions 1 and 3 asked in paragraph 3. d(fv−−fv+ ) dBext (cid:12) (cid:12) (cid:12), Fig. 4 illustrates how the inhomogeneous magnetic field alone changes fv− − fv+ (denoted as ∆(fv− − fv+)∆B). Since ∆Bθ vectorially adds to Bext, an anisotropic ∆(fv− − fv+)∆B is seen in Fig. 4c with and without the various step configurations portrayed in Fig. 3d. We also see that ∆(fv− − fv+)∆B in Fig. 4c is negligible for a flat interface, but is significant when interface steps are present. This can be understood from Figs. 4a and 4b, and/or equation 2. Interface steps generate strong valley- orbit hybridization34,35 causing the valley states to have non-identical wavefunctions, and hence different dipole moments, (hx−i − hx+i) 6= 0 and/or (hy−i − hy+i) 6= 0, as opposed to a flat interface case, which has hx±i = hy±i = 0. Thus the spatially varying magnetic field has 6 A comparison between Figs. 3f and 4d (also between equations 1 and 2) reveals that any dependence of fv− − fv+ on Bext can only come from the SOI. This indicates that the experimental B-field dependency in Fig. 2a can not be explained without the SOI. So the effect of the SOI cannot be ignored even in the presence of a micro-magnet and this answers question 2 raised in the third paragraph. However, engineering the micro-magnetic field will allow us to engineer the anisotropy of fv− − fv+ (question 3, paragraph 3). Also, the influence of interface steps will cause additional device-to-device variability (question 1, paragraph 3). FIG. 5. Calculated Rabi frequencies (fRabi) with SOI only, inhomogeneous B-field only and both SOI and inhomogeneous B-field for different direction of the external magnetic field for both v−(panel A) and v+(panel B) valley states. Interface condition, vertical e-field and parabolic confinement for the dot used in these simulations are the same as that used to match the experimental data in Figs. 1 and 2. The valley and orbital splittings that we get from the simulations are around 34.4 µeV and 0.48 meV respectively. The dot radius is around 35 nm. The fastest Rabi frequencies for SOI only are around 1MHz, which are least five times smaller compared to that of the gradient B-field for θ = 0◦. It is important to note here that the micro-magnet geometry was designed to maximize the Rabi frequency at θ = 0◦. The details of the fRabi calculation are discussed in supplementary section S6. Now the understanding of an enhanced SOI effect com- pared to bulk, brings forward an important question, whether it is possible to perform electric-dipole spin resonance (EDSR) without the requirement of micro- magnets. Here, we predict that (Fig. 5) for similar driv- ing amplitudes as used here the SOI-only EDSR can of- fer Rabi frequencies close to 1 MHz, which is around five times smaller than the micro-magnet based EDSR. More- over, the Rabi frequency of the SOI-EDSR will strongly depend on the interface condition36 (supplementary sec- tion S6) and can be difficult to control or improve. On the other hand, with improved design (stronger trans- verse gradient field) we can gain more advantage of the micro-magnets and drive even faster Rabi oscillations. However, we also predict that, both the SOI and inhomo- geneous B-field contribute to the Ez dependence of fv± (supplementary section S3) and make the qubits suscep- tible to charge noise25. As these two have comparable contribution, both of their effects will add to the charge FIG. 4. Effect of inhomogeneous magnetic field on fv− − fv+ . a, 1D cut of the wavefunctions of the two valley states close to a step edge, highlighting their spatial differences. A large vertical E-field, Ez =30 MVm−1 is used here to show a mag- nified effect. b, The change in fv− − fv + due to the inho- mogeneous B-field (∆B) alone as a function of the distance x0 between the dot center and a step edge, as defined in Fig. 3b. c, Angular dependence of ∆(fv− − fv +)∆B for the var- ious step configurations of Fig. 3d (same color code) com- puted from atomistic TB. d, ∆(fv−−fv +)∆B as a function of Bext. ∆(fv− − fv +)∆B shows negligible dependence on Bext. ∆(fv− − fv +)∆B for c3 (red lines with circular marker), in Figs. 4c and 4d, corresponds to the contribution of ∆B (the difference between the black solid curve/lines with circular markers and green dashed curve/lines with square markers) of Figs. 1c and 2a. The fields used in the simulations of c and d are the same as that of Figs. 1 and 2, whereas the fields used for b are the same as that of Fig. 3c. a different effect on the two wavefunctions, thereby con- tributing to the difference in ESR frequencies between the valley states. Fig. 4b shows ∆(fv− − fv+)∆B as a function of the dot location relative to a step edge, x0 (as in Fig. 3b) and illustrates that ∆B has the largest contribution to fv− − fv+ when the step is in the vicinity of the dot. Also, ∆(fv− − fv+)∆B is almost indepen- dent of Bext, as shown in Fig. 4d. The curves labeled c3 in both Figs. 4c and 4d correspond to the contribu- tion of ∆B in Figs. 1c and 2a. Now Ez also influences (cid:12) (cid:12)∆(fv− − fv+)∆B(cid:12)(cid:12), as shown in supplementary Fig. S4. Thus a different combination of interface steps and Ez can also produce these same ∆B results of Figs. 1c and 2a, but might not result in the necessary SOI contribu- tion to match the experiment. 0 15 30 15 30 0 x0 (nm) x 0( n m ) c2(3) c3(4) c1(5) c2(3) θ=90° θ=0° 0.5 0.7 0.9 1.1 1.3 1.5 0.5 1.3 0.9 Bext (T) B e x t( T ) 6 4 2 0 −2 -2 θ ◦ θ ◦ −4 -4 −6 -6 −30 −15 -15 -30 0246 b ∆(fv−−fv+)∆B(MHz) Δ(fv- - fv+ )ΔB (MHz) 0 45 90135180225270315 0 45 90135180225270315 fv(GHz) fv(GHz) −65 −65 x=-3.15 nm −70 −70 −75 −75 −80 −80 −85 −85 step at v− v− v- v+ v+ v+ 5 10 0 0 10 5 x (nm) x ( n m ) −10 −5 -5 -10 1.2 1 1 0.8 0.8 0.6 0.6 0.4 0.4 0.2 0.2 0 0 a Ψ2!10−7" Ψ2 (10-7) c1(5) c1(5) c c2(3) ideal c2(3) c2(3) 10 10 ideal ideal c3(4) c1(5) c3(4) c3(4) c2(3) c4(4) c4(4) c4(4) 05 ideal c1(5) c1(5) d 5 0 −5 -5 10 10 05 0 −5 −10 -10 Δ(fv- - fv+ )ΔB (MHz) ∆(fv−−fv+)∆B(MHz) 20 0 90 90 20 0 60 60 fv−−fv+(MHz) fv−−fv+(MHz) θ° θ ◦ 20 0 30 30 5 05 fv−−fv+(MHz) −10 -10 0 0 0 −5 -5 Δ (fv- - fv+ )ΔB (MHz) ∆(fv−−fv+)∆B(MHz) 30 30 90 θ° θ ◦ 30 0 60 60 90 90 60 90 1.5 1 0 0.5 0 0 60 10 10 frabi(v−)(MHz) 05 5 θ ◦ frabi (v+) (MHz) frabi(v−)(MHz) 30 (SOI) (SOI+gradB) ΔB b (gradB+SOI) (SOI) SOI (SOI+gradB) v+ (gradB+SOI) ΔB+SOI (SOI) (SOI+gradB) v- 1.5 1 frabi(v−)(MHz) 0.5 90 60 60 90 0 0.5 0 0 30 30 θ° θ ◦ 0246 2 0 1 a 6 4 1.5 frabi(v−)(MHz) frabi (v-) (MHz) frabi(v−)(MHz) noise induced dephasing of the spin qubits in the presence of micro-magnets. The coupled spin and valley behavior observed in this work may in principle enable us to simultaneously use the quantum information stored in both spin and val- ley degrees of freedom of a single electron. For example, a valley controlled not gate9 can be designed in which the spin basis can be the target qubit, while the valley information can work as a control qubit. If we choose such a direction of the external magnetic field, where the valley states have different spin splittings, an applied microwave pulse in resonance with the spin splitting of v−, will rotate the spin only if the electron is in v−. So we get a NOT operation of the spin quantum informa- tion controlled by the valley quantum information. Spin transitions conditional to valley degrees of freedom are also shown in ref. [15] and an inter-valley spin transition, which can entangle spin and valley degrees of freedom, is observed in ref. [27]. CONCLUSION To conclude, we experimentally observe anisotropic be- havior in the electron spin resonance frequencies for dif- ferent valley states in a Si QD with integrated micro- magnets. We analyze this behavior theoretically and find that intrinsic SOI introduces 180◦ periodicity in the dif- ference in the ESR frequencies between the valley states, but the inhomogeneous B-field of the micro-magnet also modifies this anisotropy. Interfacial non-idealities like steps control both the sign and magnitude of this differ- ence through both SOI and inhomogeneous B-field. We also measure the external magnetic field dependence of the resonance frequencies. We show that the measured magnetic field dependence of the difference in resonance frequencies originates only from the SOI. We conclude that even though the SOI in bulk silicon has been typi- cally ignored as being small, it still plays a major role in determining the valley dependent spin properties in in- terfacially confined Si QDs. These understandings help us answer the questions raised in paragraph 3, which are crucial for proper operation of various qubit schemes based on silicon quantum dots. 7 METHODS For the theoretical calculations, we used a large scale atomistic tight binding approach with spin re- solved sp3d5s* atomic orbitals with nearest neighbor interactions37. Typical simulation domains comprise of 1.5-2 million atoms to capture realistic sized dots. Spin- orbit interactions are directly included in the Hamilto- nian as a matrix element between p-orbitals following the prescription of Chadi38. The advantage of this ap- proach is that no additional fitting parameters are needed to capture various types of SOI such as Rashba and Dres- selhaus SOI in contrast to k.p theory. We introduce monoatomic steps as a source of non-ideality consistent with other works33,35,39. The Si interface was modeled with Hydrogen passivation, without using SiGe. This in- terface model is sufficient to capture the SOI effects of a Si/SiGe interface discussed in refs. [30–32]. We have used the methodology of ref. [29] to model the micro- magnetic fields [supplementary section S4]. Full magne- tization of the micro-magnet is assumed. This causes the value of the magnetization of the micro-magnet to be saturated and makes it independent of Bext. However, a change in the direction of Bext changes the magneti- zation. We include the effect of inhomogeneous mag- netic field perturbatively, with the perturbation matrix elements, hψm 1 dj j ψni. Here, ψn and ψm are atomistic wave-functions calculated with homogeneous magnetic field. For further details about the numerical techniques, see NEMO3D reference [37]. Method details about the experiment can be found in ref. [15]. The dot location in this experiment is differ- ent from ref. [15]. The device was electrostatically reset by shining light using an LED and all the measurements were done with a new electrostatic environment (a new gate voltage configuration). The quantum dot location is estimated by the offsets of the magnetic field created by the micro-magnets extrapolated from the measurements shown in Fig. 2 and comparing to the simulation results shown in supplementary section S4. We also observed that the Rabi frequencies were different from ref. [15] when applying the same microwave power to the same gate, which qualitatively indicates that the dot location is different. 2 gµPi,j hψm dBφ 2 gµ∆Bφ ψni = 1 i [1] Datta, S. & Das, B. Electronic analog of the electro-optic 2184 (2005). modulator. Appl. Phys. Lett. 56, 665(R) (1990). [2] Wolf, S. et al. Spintronics: A spin based electronics vision for the future. Science 294, 1488-1495 (2001). [3] I. Zuti´c, Fabian, J. & Das Sarma, S. Spintronics: Fun- damentals and applications. Rev. Mod. Phys. 76, 323 (2004). [4] Loss, D. & Vincenzo, D. P. Quantum computation with quantum dots. Phys. Rev. A 57, 120 (1998). [5] Petta, J. et al. Coherent manipulation of coupled electron spins in semiconductor quantum dots. Science 309, 2180- [6] Koppens, F. et al. Driven coherent oscillations of a sin- gle electron spin in a quantum dot. Nature 442, 766-771 (2006). [7] Hanson, R., Kouwenhoven, L. P., Petta, J. R., Tarucha, S., Vandersypen & L. M. K. Spins in few electron quan- tum dots. Rev. Mod. Phys. 79, 1217 (2007). [8] Xiao, D., Liu, G.-B. , Feng, W., Xu, X. & Yao, W. Cou- pled spin and valley physics in monolayers of MoS2 and other group VI dichalcogenides. Phys. Rev. Lett. 108, 196802 (2012). [9] Gong, Z. et al. Magnetoelectric effects and valley- controlled spin quantum gates in transition metal dichalcogenide bilayers. Nature Communications 4, 2053 (2013). [10] Xu, X., Yao, W., Xiao, D. & Heinz, T. Spins and pseu- dospins in layered transition metal dichalcogenides. Na- ture Physics 10, 343350 (2014). [11] Laird, E. A., Pei, F. & Kouwenhoven, L. P. A valleyspin qubit in a carbon nanotube. Nature Nanotech. 8, 565568 (2013). [12] Renard, V. T. et al. Valley polarization assisted spin po- larization in two dimensions. Nature Communications 6, 7230 (2015) [13] Yang, C. et al. Spin-valley lifetimes in a silicon quan- tum dot with tunable valley splitting. Nature Communi- cations 4, 2069 (2013). [14] Hao, X., Ruskov, R., Xiao, M., Tahan, C. & Jiang, H. Electron spin resonance and spin-valley physics in a sil- icon double quantum dot. Nature Communications 5, 3860 (2014). [15] Kawakami, E. et al. Electrical control of a long-lived spin qubit in a Si/SiGe quantum dot. Nature Nanotechnology 9, 666670 (2014). [16] Veldhorst, M. et al. An addressable quantum dot qubit with fault tolerant control fidelity. Nature Nanotechnol- ogy 9, 981985 (2014). [17] Maune, B. M. et al. Coherent singlet-triplet oscillations in a silicon-based double quantum dot. Nature 481, 344347 (2011). [18] Wu, X. et al. Two-axis control of a singlet-triplet qubit with integrated micromagnet. Proc. Natl. Acad. Sci. USA 111, 11938 (2014). [19] Eng, K. et al. Isotopically enhanced triple quantum dot qubit. Science Advances 1, no. 4, e1500214 (2015). [20] Kim, D. et al. Quantum control and process tomography of a semiconductor quantum dot hybrid qubit. Nature 511, 70-74 (2014). [21] Maurand, R. et al. A CMOS silicon spin qubit. Nature Communications 7, 13575 (2016). [22] Culcer, D., Sariava, A. L., Koiller, B., Hu, X. & Das Sarma, S. Valley-based noise resistant quantum computa- tion using Si quantum dots. Phys. Rev. Lett. 108, 126804 (2012). [23] Rohling, N. & Burkard, G. Universal quantum computing with spin and valley states. New J. Phys. 14, 083008 (2012). [24] Rohling, N., Russ, M. & Burkard, G. Hybrid spin and val- ley quantum computing with singlet-triplet qubits. Phys. Rev. Lett. 113, 176801 (2014). [25] Veldhorst, M. et al. Spin-orbit coupling and operation of multivalley spin qubits. Phys. Rev. B 92, 201401(R) (2015). [26] Scarlino, P. et al. Second-harmonic coherent driving of a spin qubit in SiGe/Si quantum dot. Phys. Rev. Lett. 115, 106802 (2015). [27] Scarlino, P. et al., Dressed photon-orbital states in a quantum dot: Inter-valley spin resonance, Phys. Rev. B 95, 165429 (2017).. [28] Tokura, Y., van der Wiel, W. G., Obata, T. & Tarucha, S. Coherent single electron spin control in a slanting Zee- man field. Phys. Rev. Lett. 96, 047202 (2006). [29] Goldman, J. R., Ladd, T. D., Yamaguchi, F. & Ya- mamoto, Y. Magnet designs for a crystal lattice quantum computer. Appl. Phys. A 71, 11 (2000). 8 [30] Golub, L. E., & Ivchenko, E. L. Spin splitting in sym- metrical SiGe quantum wells. Phys. Rev. B 69, 115333 (2004) [31] Nestoklon, M. O., Golub, L. E. & Ivchenko, E. L. Spin and valley-orbit splittings in SiGe/Si heterostructures. Phys. Rev. B 73, 235334 (2006). [32] Nestoklon, M. O., Ivchenko, E. L., Jancu, J. -M. & Voisin, P. Electric field effect on electron spin splitting in SiGe/Si quantum wells. Phys. Rev. B 77, 155328 (2008). [33] Zandvliet, H. J. W. & Elswijk, H. B. Morphology of monatomic step edges on vicinal Si(001). Phys. Rev. B 48, 14269 (1993). [34] Gamble, J. K., Eriksson, M. A., Coppersmith, S. N. & Friesen, M. Disorder-induced valley-orbit hybrid states in Si quantum dots. Phys. Rev. B 88, 035310 (2013). [35] Friesen, M., Eriksson, M. A. & Coppersmith, S. N. Magnetic field dependence of valley splitting in realis- tic Si/SiGe quantum wells, Appl. Phys. Lett. 89, 202106 (2006). [36] Huang, W., Veldhorst, M., Zimmerman, N. M., Dzurak, A. S. & Culcer, D. Electrically driven spin qubit based on valley mixing. Phys. Rev. B 95, 075403 (2017). [37] Klimeck, G. et al. Atomistic simulation of realistically sized nanodevices using NEMO 3D: Part I-models and benchmarks. IEEE Trans. Electron Dev. 54, 2079–2089 (2007). [38] Chadi, D. J. Spin-orbit splitting in crystalline and com- positionally disordered semiconductors. Phys. Rev. B 16, 790 (1977). [39] Kharche, N., Prada, M., Boykin, T. B. & Klimeck, G. Valley splitting in strained silicon quantum wells modeled with 2 miscuts, step disorder, and alloy disorder. Appl. Phys. Lett. 90, 092109 (2007). ACKNOWLEDGMENTS This work was supported in part by ARO (W911NF- 12-0607); development and maintenance of the growth fa- cilities used for fabricating samples is supported by DOE (DE-FG02-03ER46028). This research utilized NSF- supported shared facilities (MRSEC DMR-1121288) at the University of Wisconsin-Madison. Computational re- sources on nanoHUB.org, funded by the NSF grant EEC- 0228390, were used. M.P.N. acknowledges support from ERC Synergy Grant. R.F. and R.R. acknowledge discus- sions with R. Ruskov, C. Tahan, and A. Dzurak. AUTHOR CONTRIBUTIONS STATEMENT R.F. performed the g-factor calculations, explained the underlying physics and developed the theory with guid- ance from R.R. R.F., R.R., E.K., P.S., and M.P.N. ana- lyzed the simulation results and compared with experi- mental data in consultation with L.M.K.V., M.F., S.N.C. and M.A.E.. E.K. and P.S. performed the experiment and analyzed the measured data. D.R.W. fabricated the sample. D.E.S. and M.G.L. grew the heterostructure. R.F. and R.R. wrote the manuscript with feedback from all the authors. R.R. and L.M.K.V. initiated the project, and supervised the work with S.N.C, M.F. and M.A.E. ADDITIONAL INFORMATION The authors declare that they have no competing fi- nancial interests. 9 SUPPLEMENTARY MATERIALS FOR 'VALLEY DEPENDENT ANISOTROPIC SPIN SPLITTING IN SILICON QUANTUM DOTS' S1. Analytic effective mass model to explain the effect of spin-orbit interaction (SOI) on the valley dependent g-factor anisotropy in a silicon quantum dot To explain our atomistic tight-binding (TB) results of the valley dependent g-factor and its anisotropy in a sil- icon quantum dot (QD), we have also performed analytic effective mass calculations as shown here. The electron Hamiltonian can be written as, 1 H = 2 2m k2 + V (r) + HZ + HSO (Eq. S1) Here, m is the electron effective mass, which assumes a value of 0.19m0 and 0.91m0 for the transverse (x or [100], y or [010]) and longitudinal (z or [001]) components in Si respectively (m0 being the free electron mass). V (r) is the potential defining the quantum dot, and has been discussed in the main text. HZ = 1 2 gµσ · B is the Zeeman term, with g the electron g-factor, µ the Bohr magneton, and B the applied magnetic field at the dot location. HSO = β (σxkx − σyky) + α (σxky − σykx) is the SOI term, where β (α) is the strength of Dresselhaus (Rashba) interaction, σx, σy are the Pauli spin matrices, and kx (ky) is electron canonical momentum along x (y) direction. k = −i∇− e A  , where A is the vector potential and B = ∇ × A. Now, we treat HP = 1 2 gµσ · B + β (σxkx − σyky) + α (σxky − σykx) as a perturbation to H0 = 2 2m k2 + V (r). The unperturbed Hamiltonian H0 yields the spin degenerate eigenstates of a Si QD. Typically, in these devices, orbital splitting (EOS) is much larger than valley (EVS) and spin (EZS) splittings. So, we only consider the four lowest energy states (cid:12)(cid:12)(cid:12)ψ↓v−E, (cid:12)(cid:12)(cid:12)ψ↑v−E, (cid:12)(cid:12)(cid:12)ψ↓v+E, (cid:12)(cid:12)(cid:12)ψ↑v+E as the basis for the perturbation calculation. We can write the perturbation Hamiltonian as, 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 HP = σx(cid:18) 1 2 gµBx + βkx + αky(cid:19) + σy(cid:18) 1 2 gµBy − βky − αkx(cid:19) + σz(cid:18) 1 2 gµBz(cid:19) (Eq. S2) 19 After diagonalizing S2, we obtain the spin splittings for different valley states, (Eq. S3) +(cid:18) 1 2 gµBx + β±(cid:10)k±x(cid:11) + α±(cid:10)k±y(cid:11)(cid:19)2 +(cid:18) 1 2 gµBy − β±(cid:10)k±y(cid:11) − α±(cid:10)k±x(cid:11)(cid:19)2) 1 2 2 gµBz) 1 2 EZS(±) = 2(( Here, we replaced β, α with β±, α± to denote the two v+ and v− valley states1,2. Now, the expectation values of the momentum operator in a magnetic field B are, 20 21  ∂ ∂x − e (cid:10)k±x(cid:11) =(cid:10)ψv±(cid:12)(cid:12) − i (cid:10)k±y(cid:11) =(cid:10)ψv±(cid:12)(cid:12) − i ∂y(cid:12)(cid:12)ψv±(cid:11) ≈ 0. For a magnetic field in the x-y plane, we assume, Az = 0, Ax = zBy  (cid:12)(cid:12)ψv±(cid:11) ≈ e hA±x i  (cid:12)(cid:12)ψv±(cid:11) ≈ e(cid:10)A±y(cid:11) y − e (Eq. S5) (Eq. S4) ∂ ∂ Ax Ay Since(cid:10)ψv±(cid:12)(cid:12)− i ∂ ∂x(cid:12)(cid:12)ψv±(cid:11) ≈(cid:10)ψv±(cid:12)(cid:12)− i ∂ and Ay = −zBx, where Bx and By are the x and y components of the magnetic field respectively. Then, arXiv:1702.06210v2 [cond-mat.mes-hall] 17 Aug 2017 22 23 24 25 26 By  (cid:10)k±x(cid:11) ≈ e hzi (cid:10)k±y(cid:11) ≈ −e hzi  Bx (Eq. S6) (Eq. S7) Here, hzi is the dipole moment along [001] crystal direction. We assumed hz−i = hz+i = hzi. As the electrons in the v− and v+ valley states have only z valley component, we can replace g here with g⊥, the g-factor perpendicular to the valley axis4. So, for an in-plane magnetic field, EZS(±) = 2((cid:18) 1 2 g⊥µBx + β± e hzi  By − α± e hzi  Bx(cid:19)2 +(cid:18) 1 2 g⊥µBy + β± e hzi  Bx − α± e hzi  2 2 By(cid:19)2) 1 (Eq. S8) In the presence of only external magnetic field, Bext, Bx = Bextcosφ and By = Bextsinφ, with φ being the B-field angle relative to the [100] crystal orientation for a counter clockwise rotation of Bext. However, to be consistent with the angle θ defined in the experiment (i.e. an angle relative to the [110] crystal orientaiton with clockwise rotation of Bext), we use the relation θ = 45◦ − φ. Now, using the definition of Bx and By from above, we can obtain the angular dependence of the spin splitting for different valley states, β±α± sin 2φ) 1 β± sin 2φ − 8(cid:18)ehzi g⊥µ(cid:19)2 α± + 4ehzi g⊥µ 2 + α± 2 (Eq. S9) 2. So, we can ignore the second order terms. Thus simplifying equation (S9) 27 28 29 30 31 32 33 2(cid:1) − 4ehzi EZS(±) = g⊥µBext(1 + 4(cid:18)ehzi g⊥µ(cid:19)2(cid:0)β± gµ (cid:17) ∗ α± gµ (cid:17) ∗ β± >(cid:16)ehzi Now, 1 (cid:29)(cid:16)ehzi EZS(±) = g⊥µBext(cid:26)1 − 4ehzi g⊥µ g⊥µ we get, α± + 4ehzi g⊥µ β± sin 2φ(cid:27) 1 2 (Eq. S10) 34 After doing a series expansion and ignoring higher order terms, we can simplify this expression even further, EZS(±) = g⊥µBext(cid:26)1 − 2ehzi g⊥µ α± + 2ehzi g⊥µ β± sin 2φ(cid:27) (Eq. S11) 35 36 37 So, from this equation we can see that, without the Dresselhaus contribution, there is no angular dependence or anisotropy in the spin splitting (or g-factor) for the different valley states. Now, for Bext along the [110] and [1¯10] crystal orientations, we get, E[110] ZS(±) − g⊥µBext = 2 (β± − α±)e hzi  ZS(±) − g⊥µBext = 2 (−β± − α±)e hzi E[110]  Bext Bext (Eq. S12) (Eq. S13) 38 39 Equation (S12) matches the analytic prediction of ref. [1]. We can extract α± and β± from the atomistic calculations as follows, α± = −(cid:16)E[110] ZS(±) + E[1¯10] ZS(±) − 2g⊥µBext(cid:17) 4e hzi Bext E[1¯10] ZS(±) − E[1¯10] ZS(±) 4e hzi Bext β± = (Eq. S14) (Eq. S15) 40 41 42 43 44 45 46 47 48 49 In Fig. S1, we compared fv−−fv+ calculated from this analytic model with the atomistic tight-binding results for the ideal interface case, shown in Figs. 3e and 3f of the main text. Both the anisotropic behavior of fv− − fv+ with respect to the angle of Bext (Fig. S1A), and the dependence on the magnitude Bext along [110] and [1¯10] crystal orientations (Fig. S1B) show perfect agreement between analytic and atomistic calculations. From the atomistic calculations of Fig. 3a and equation (S14) and (S15), we extracted β− = 10.117 × 10−15eV·m, β+ = −9.3621 × 10−15eV·m, α− = −1.2568 × 10−15eV· m, α+ = −1.5920 × 10−15eV· m for g⊥=1.9937 and hzi = 2.7925 nm, calculated using hzi ≈ 1.5587lz, where, lz =(cid:16) (S7) for hk±x i and(cid:10)k±y(cid:11). micro (θ), micro (θ). As, Bx/y (cid:29) Bz and hx/yi (cid:29) hzi, we can still use equation (S6) and To include the homogeneous fields from the micro-magnets, in equation (S3), we use, Bx = Bext cos θ + Bx 3 1. Here, Ez = 6.77 MVm−1 is the vertical electric field. 2mleEz(cid:17) 1 micro (θ) and Bz = Bz By = Bext sin θ + By 2 3 Supplementary Fig. S1. Effect of spin-orbit interaction: comparison between analytic effective mass and atomistic tight-binding calculations. A, Comparison in angular dependence of fv− − fv+ for Bext =800mT. The angle of Bext is defined clockwise with respect to the [110] crystal orientation) B, Magnetic field dependence of fv− − fv+ for Bext along [110] and [1¯10] crystal orientations. The tight-binding calculations correspond to the ideal (interface) case of Figs. 3e and 3f of the main text. 60 60 40 40 20 20 0 0 −20 -20 −40 -40 −60 -60 0.5 0.7 0.9 1.1 1.3 1.5 0.5 E M [ 1 1 0 ] E M [ 1 1 0 ] E M [ 1 1 0 ] T B [ 1 1 0 ] EM [110] E M [ 1 1 0 ] T B [ 1 1 0 ] TB [110] T B [ 1 1 0 ] E M [ 1 ¯10 ] T B [ 1 1 0 ] E M [ 1 ¯10 ] EM [110] E M [ 1 ¯10 ] T B [ 1 ¯10 ] TB [110] E M [ 1 ¯10 ] T B [ 1 ¯10 ] T B [ 1 ¯10 ] T B [ 1 ¯10 ] 1.3 - - 0.9 Bext (T) B e x t( m T ) fv- - fv+ (MHz) B fv−−fv+(MHz) 1500 1500 1500 1500 75 30 30 75 75 50 20 75 20 50 50 25 50 10 10 25 25 0 25 0 0 0 0 −25 0 −10 −25 -10 −25 −50 −25 −50 −20 -20 −50 −75 −50 −75 −30 -30 −75 EM 500 1000 −75 EM 90 0 180 270 0 180 270 500 1000 90 EM 500 1000 TB TB θ° TB B e x t(m T) θ ◦ 1000 500 B e x t(m T) B e x t(m T) fv−−fv+(MHz) fv−−fv+(MHz) fv−−fv+(MHz) fv- - fv+ (MHz) fv−−fv+(MHz) fv−−fv+(MHz) A S2. Analytic model to explain the effect of gradient magnetic field on the valley dependent spin- splittings in a Si QD To develop an analytic model to capture the effect of the inhomogeneous magnetic field on the spin-splittings for different valley states, we assume the Hamiltonian in equation (S1) as our unperturbed Hamiltonian and include the magnetic field gradient as a perturbation, 4 gµXi σiXj dBi dj j 1 2 gµσ.∆B = 1 2 H ∆B P = dBi dj jv− ↓i dj hj−i dBi σiXj gµXi gµXi h↓ σi ↓iXj 1 2 ∆E∆B v−↓ = hv− ↓ 1 2 ∆E∆B v−↓ = Here, dBi dj are the magnetic field gradients along different directions, where i, j correspond to x,y,z co-ordinates. Now the lowest 4 eigen-states of the unperturbed Hamiltonian are v− ↓i, v− ↑i, v+ ↓i and v+ ↑i. The 1st order correction due to H ∆B P to v− ↓i is given by, To simplify our analytic calculation, we assume that the spin mixing due to SOI is very small and we can separate the spin part from the spatial part of the wavefunction. Using this approximation in Eq. S17, we get, Here, hj−i is the dipole moment along the j direction, hj−i = hv− j v−i. Now, for spins in an in-plane magnetic field, h↓ σx ↓i = − cos φ, h↓ σy ↓i = − sin φ and h↓ σz ↓i = 0. The external magnetic field fully magnetizes the micro-magnets. So, the inhomogeneous magnetic field from the micro-magnets ( dBi dj ) depends on the direction (φ) of the external magnetic field. Using these relations in equation (S19) we get, (Eq. S16) (Eq. S17) (Eq. S18) (Eq. S19) (Eq. S20) (Eq. S21) We can get similar expressions for v+ ↓i and v+ ↑i. So, the change in spin splitting of both the valleys due to the gradient magnetic field is, hj−i hj−i hj−i dBφ x dj  + sin φXj dj  + sin φXj dBφ x dBφ y hj−i hj±i dBφ x dj  + sin φXj dBφ y hj±i dBφ y dj  dj  dj  v−↑ = gµ cos φXj gµ cos φXj ZS(±) = gµ cos φXj 1 2 1 2 ∆E∆B ∆E∆B ∆E∆B v−↓ = − Now, for v− ↑i, h↑ σx ↑i = cos φ, h↑ σy ↑i = sin φ and h↑ σz ↓i = 0. So, For an electron in an ideal (smooth) interface, hx±i ≈ hy±i ≈ 0. But the presence of interface steps in realistic devices makes hx±i and/or hy±i non zero. We can ignore the 2nd order corrections because 1 dj hj±i (cid:28) EZS. Due to interface steps, hx−i 6= hx+i and/or hy−i 6= hy+i, but hz−i ≈ hz+i, for the electric field used in the experiment. So, in the presence of interface steps, the magnetic field gradient adds to the difference in ESR frequencies between the valley states, and from equation (S21), we get equation (3) of the main text. 2 gµ dBφ i 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 5 Supplementary Fig. S2. Effect of the gradient magnetic field on the anisotropy of fv− − fv+ : comparison between analytic effective mass and atomistic tight-binding calculations. For the latter, the contribution of the SOI is subtracted to obtain the change in fv− − fv+ due to ∆B only from Fig. 1c of the main text (difference between green dashed line and black solid line). In Fig. S2, we compared ∆(cid:0)fv− − fv+(cid:1)∆B calculated using equation (2) of the main text with atomistic tight- binding calculations. We used hx−i − hx+i ≈ −0.17 nm and hy−i − hy+i = 0 in equation (2). The atomistic results shown in Fig. S2 is the difference between the black curve (circular marker) and the green curve (square markers) shown in Fig. 1c. The analytic calculation qualitatively captures the tight-binding results. The little mismatch between the analytic and atomistic calculation is due to ignoring the spin mixing from SOI in our analytic model. 72 73 74 75 76 EM EM TB TB 60 60 90 90 30 30 θ° θ ◦ 3 3 2.5 2.5 2 2 ∆(fv−−fv+)∇B(MHz) Δ(fv- - fv+ )ΔB (MHz) 1.5 1.5 0 0 S3. Effect of the vertical electric field on the difference in ESR frequencies between valley states through both SOI and gradient magnetic field 6 The vertical electric field, Ez affects fv− − fv+, because it affects the SOI parameters2 and also the dipole moment parameters. The effect of Ez on fv− − fv+ due to the SOI and gradient magnetic field, from atomistic tight-binding calculations, are shown in supplementary Figs. S3 and S4 respectively. Here Ez only changes the magnitude of fv− − fv+ , but not its sign. Ez also effect the Bext dependence of fv− − fv+ through SOI, as shown in supplementary Fig. S3B. Here also Ez only changes the magnitude of the slope, d∆(fv−−fv+ )SOI but not its sign. dBext 77 78 79 80 81 82 83 Supplementary Fig. S3. Influence of the vertical electric field Ez on the change in fv− − fv+ due to SOI. A, Increase in (cid:12)∆(cid:0)fv− − fv+(cid:1)SOI(cid:12)(cid:12)(cid:12) with increasing Ez for Bext=0.8 T. B, Change in ∆(cid:0)fv− − fv+(cid:1)SOI with changing Bext for different Ez. (cid:12) (cid:12) (cid:12) Ez changes the magnitude of the slope, (cid:12)(cid:12)(cid:12)(cid:12) (cid:12) (cid:12) (cid:12). Bottom panels in both Figs. A and B corresponds to Bext along [110] (θ = 0◦) and in top panels Bext is along [1¯10] (θ = 90◦). Interface condition and parabolic confinement for the dot used in these simulations are the same as that used to match experimental data in Figs. 1 and 2 of main text. d∆(fv−−fv+ )SOI dBext 15 10 10 6.77 10 6.77 6.77 15 15 10 θ=90° 6.77 30 20 10 0 0 −10 −20 −30 0.9 Bext (T) B e x t(T) 0.5 0.7 0.9 1.1 1.3 1.5 0.5 0.7 0.9 1.1 1.3 1.5 0.5 1.3 0.5 0.7 0.9 1.1 1.3 1.5 ∆(fv−−fv+)SOI(MHz) θ=0° 30 30 20 20 10 30 10 0 20 0 10 0 0 0 −10 -10 −20 0 -20 Δ −30 −10 -30 −20 −30 ∆(fv−−fv+)SOI(MHz) ∆(fv−−fv+)SOI(MHz) (fv- - fv+)SOI (MHz) B 30 20 10 0 0 −10 −20 −30 ∆(fv−−fv+)SOI(MHz) θ=0° 7.5 10 12.5 15 7.5 10 12.5 15 Ez (MVm-1) E Z( M V / m ) θ=90° 15 15 10 10 5 5 −5 -5 −10 -10 −15 -15 5 5 (MHz) A ∆(fv−−fv+)SOI(MHz) Δ(fv- - fv+)SOI 7 Supplementary Fig. S4. Influence of the vertical electric field Ez on the change in fv− − fv+ due to the gradient magnetic field ∆B, for Bext along [110] θ = 0◦ (bottom panel) and [1¯10] (θ = 90◦) (top panel). Interface condition and parabolic confinement for the dot used in these simulations are the same as that used to match experimental data in Figs. 1 and 2 of the main text. Increasing Ez increases(cid:12)(cid:12)(cid:12)∆(cid:0)fv− − fv+(cid:1)∆B(cid:12)(cid:12)(cid:12). θ=90° θ=0° 7.5 10 12.5 15 7.5 10 12.5 15 Ez (MVm-1) E Z( M V / m ) 5 1 5 5 135 3 135 1 5 3 ∆(fv−−fv+)∆B(MHz) (MHz) Δ(fv- - fv+)ΔB S4. Modeling the stray magnetic field induced by the micro-magnets We calculated the local magnetic field created by the micro-magnets when the external magnetic field is applied along the y0 ([110]) (θ = 0◦) or along the x0 ([1¯10]) axis (θ = 90◦) of the device picture in Fig. S5, assuming that the micro-magnets are fully magnetized5,6. The shape of the micro-magnets is shown in ref. [7]. Fig. S5 and Fig. S6 show the results of the numerical calculation of the total magnetic field gradient when the external magnetic field is applied along the y0 axis and along the x0 axis respectively, and the pink circle shows the estimated dot position. We calculated the stray magnetic field created by the micro-magnets when the external magnetic field is applied between the y0 axis and the x0 axis (0◦ < θ < 90◦) according to the following approximation, 8 i = cos2 θB0◦ Bθ i + sin2 θB90◦ i , (Eq. S22) i i where i =x0,y0,z. We ignore dBθ electron wavefunctions of the two valley states have similar spread along z, so hv− dBθ the effects of dBθ shown in the Figs. 1d and 2b of the main text, the effect of the gradient magnetic field on fv± to that of the homogeneous magnetic fields. dz , with i =x0,y0,z. For small electric field like Ez=6.77 MVm−1 used to match the experiment, the dz z v+i. Thus dz on fv− or fv+ can be larger, but as is negligible compared dz on fv− − fv+ should be small. However, the effects of these dBθ dz z v−i ≈ hv+ dBθ The micro-magnetic field used in our calculations are B0◦ dx0 = (−0.185, 0.056, 0.217) mT/nm, dB0◦ micro = (0.03, 0.03, 0.139) T, B90◦ dy0 = (−0.0653,−0.93,−0.052) mT/nm, dB90◦ micro = (−0.122, 0.03, 0.089) dx0 = (−0.272,−0.185,−0.456) i i i T, dB0◦ mT/nm, dB0◦ dy0 = (−0.184,−0.065, 0.211) mT/nm. 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 9 Supplementary Fig. S5. Numerically computed x0, y0 and z components of the magnetic field and their gradients along the x0 and y0 axes induced by the micro-magnets when the external magnetic field is applied along the y0 axis in the plane of the Si quantum well, for fully magnetized micro-magnets. The black solid lines indicate the edges of the micro-magnet as simulated. 400nm y',[110] θ=0° Bext - x',[110] 10 Supplementary Fig. S6. The same as in Fig. S5 when the external magnetic field is applied along the x0 axis. 400nm y',[110] θ=90° Bext - x',[110] 11 Supplementary Fig. S7. Effect of Bmicro on the total homogeneous magnetic field. A, Anisotropic total magnetic field, due to the presence of the homogeneous magnetic field from the micro-magnet, Bθ micro. Bθ micro adds vectorially to Bext. Also the magnetization of the micro-magnet depends on the direction of Bext or θ. So Bθ micro itself is anisotropic, hence the superscript θ. B, Total magnetic field (shown in green dashed line with square markers) along along [110] (θ = 0◦) (bottom panel) and [1¯10] (θ = 90◦) (top panel) with changing Bext. The black line represents Btot = Bext. B θ=90° 0.5 0.7 0.9 1.1 1.3 1.5 0.5 1.3 θ=0° 0.9 Bext (T) B e x t 1.5 1.5 1 1 0.5 0.5 1.5 1.5 1 1 0.5 0.5 micro −−→Bext+ Btot=Bext+Bmicro (T) −−−→ Bθ Btot= 60 60 90 90 30 30 θ° θ ◦ A 0.85 0.85 0.8 0.8 0.75 0.75 0.7 0.7 1.5 1.5 0 0 micro (T) Btot=Bext+Bθ 102 103 104 S5. Extracting SOI parameters from the experimental measurements The contributions of the SOI due to Bext on fv− − fv+ is given by equation 1 of the main text. The dependence of fv− − fv+ on Bext only comes from this equation. Thus we can calculate d(fv−−fv+ ) dBext , 12 d(fv− − fv+) dBφ ext 4π ehzi h2 ≈ {(β− − β+) sin 2φ − (α− − α−)} Now, for Bext along the [110] and [1¯10] crystal orientations, we get, d(fv− − fv+) dB[110] ext d(fv− − fv+) dB[1¯10] ext 4π ehzi h2 4π ehzi h2 ≈ ≈ {(β− − β+) − (α− − α−)} {− (β− − β+) − (α− − α−)} 105 Then we can extract (β− − β+) and (α− − α−), (β− − β+) ≈ (α− − α+) ≈ − h2 8π ehzi( d(fv− − fv+) 8π ehzi( d(fv− − fv+) dB[110] ext dB[110] ext h2 − d(fv−−fv+ ) dBext + ) d(fv− − fv+) ext dB[1¯10] d(fv− − fv+) dB[1¯10] ext ) (Eq. S23) (Eq. S24) (Eq. S25) (Eq. S26) (Eq. S27) 106 107 108 Thus from the experimentally measured SOI parameters. along the [110] and [1¯10] crystal orientations, we extract the 13 109 110 S6. Steps to calculate Rabi frequency for electric-dipole spin resonance (EDSR) due to SOI and inhomogeneous B-field The time dependent perturbation to the Hamiltonian due to an ac electric field Eac(t) = Eacf (t) is Hp(t) = qEac(t)x. Then the off-diagonal matrix element between up and down spin states for both the valley states is hv± ↓ Hp(t)v± ↑i = qEacf (t)hv± ↓ xv± ↑i. For a sinusoidal perturbation, f (t) = cos(wt), where w = 2πfv± , the Rabi frequency is given by, fRabi(v±) = qEachv± ↓ xv± ↑i h (Eq. S28) 111 112 113 Using atomistic tight-binding wavefunctions we calculate hv± ↓ xv± ↑i and then use Eq. S28 to calculate the Rabi frequency. Here we use9, Eac = 2 kVm−1. Since interface roughness affects hv± ↓ xv± ↑i in a Si QD, these Rabi frequencies will strongly depend on the interface condition. 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 [1] Veldhorst, M. et al. Spin-orbit coupling and operation of multivalley spin qubits. Phys. Rev. B 92, 201401(R) (2015). [2] Nestoklon, M. O., Ivchenko, E. L., Jancu, J. -M., & Voisin, P. Electric field effect on electron spin splitting in SiGe/Si quantum wells. Phys. Rev. B 77, 155328 (2008). [3] Nestoklon, M. O., Golub, L. E., & Ivchenko, E. L. Spin and valley-orbit splittings in SiGe/Si heterostructures. Phys. Rev. B 73, 235334 (2006). [4] Roth, L. g Factor and Donor Spin-Lattice Relaxation for Electrons in Germanium and Silicon. Phys. Rev. 118, 1534 (1960) [5] Goldman, J. R., Ladd, T. D., Yamaguchi, F., & Yamamoto, Y. Magnet designs for a crystal lattice quantum computer. Appl. Phys. A 71, 11 (2000). [6] Pioro-Ladri`ere, M., Tokura, Y., Obata, T., Kubo, T. & Tarucha, S. Micromagnets for coherent control of spin-charge qubit in lateral quantum dots. Applied Physics Letters 90, 024105 (2007). [7] Kawakami, E. et al. Electrical control of a long-lived spin qubit in a Si/SiGe quantum dot. Nature Nanotechnology 9, 666670 (2014). [8] Tokura, Y., van der Wiel, W. G., Obata, T. & Tarucha, S. Coherent single electron spin control in a slanting Zeeman field. Phys. Rev. Lett. 96, 047202 (2006). [9] Huang, W., Veldhorst, M., Zimmerman, N. M., Dzurak, A. S. & Culcer, D. Electrically driven spin qubit based on valley mixing. Phys. Rev. B 95, 075403 (2017). 

1711.00206

1

1711

Steady state current fluctuations and dynamical control in a nonequilibrium single-site Bose-Hubbard system

We investigate nonequilibrium energy transfer in a single-site Bose-Hubbard model coupled to two thermal baths. By including a quantum kinetic equation combined with full counting statistics, we investigate the steady state energy flux and noise power. The influence of the nonlinear Bose-Hubbard interaction on the transfer behaviors is analyzed, and the nonmonotonic features are clearly exhibited. Particularly, in the strong on-site repulsion limit, the results become identical with the nonequilibrium spin-boson model. We also extend the quantum kinetic equation to study the geometric-phase-induced energy pump. An interesting reversal behavior is unraveled by enhancing the Bose-Hubbard repulsion strength.

cond-mat.mes-hall

cond-mat

Steady state current fluctuations and dynamical control in a nonequilibrium single-site Bose-Hubbard system Xu-Min Chen1, Chen Wang1,∗and Ke-Wei Sun1 1 Department of Physics, Hangzhou Dianzi University, Hangzhou, Zhejiang 310018, China Abstract We investigate nonequilibrium energy transfer in a single-site Bose-Hubbard model coupled to two thermal baths. By including a quantum kinetic equation combined with full counting statistics, we investigate the steady state energy flux and noise power. The influence of the nonlinear Bose-Hubbard interaction on the transfer behaviors is analyzed, and the nonmonotonic features are clearly exhibited. Particularly, in the strong on-site repulsion limit, the results become identical with the nonequilibrium spin-boson model. We also extend the quantum kinetic equation to study the geometric-phase-induced energy pump. An interesting reversal behavior is unraveled by enhancing the Bose-Hubbard repulsion strength. Keywords: Nonequilibrium Bose-Hubbard model, Quantum transport, Full counting statistics, Geometric-phase induced energy pump 1. Introduction Far-from-equilibrium transport, out of linear-response and quasi-equilibrium regimes, has been attracting much attention, ranging from molecular electronics, quantum magnets to strongly-correlated materials [1 -- 5], which is of great importance both for fundamental research and practical application. According to the second law of thermodynamics, energy will flow naturally from the hot source to the cold drain, under the thermodynamic bias. Thus, how to control energy transfer in low-dimensional quantum systems becomes a crucial issue, to unravel the nonequilibrium mechanism and improve the design of efficient devices [6]. Many proposals have been carried out to study nonequilibrium transport in fermionic systems [7]. Various interesting phenomena have been unraveled mainly due to the interplay between the volt- age and the temperature bias. In particular, the photovoltaic effect, driven by the nonequilibrium light-electron interaction, provides an efficient way to convert the sunlight to electricity for useful per- formance [8]. And the influence of quantum coherence on improving the photovoltaic efficiency was recently proposed [9 -- 11]. While the thermoelectric effect, one typical kind of heat transfer, describes direct conversion of the thermal bias to electric voltage [7, 12]. The relationship between the ther- moelectric figure of merit and the conversion efficiency has been quantitatively characterized [13, 14]. Moreover, the Kondo effect, an anomalous feature of the conductance in low temperature regime, de- scribes the scattering of the conduction electrons mediated by a magnetic impurity [15, 16]. However, as an intimate analogy, the corresponding bosonic systems are lack of exploitation. Recently, due to fast development of photonics and phononics in quantum transport, the bosonic systems gain significant popularity [17 -- 26]. As a prototype, the nonequilibrium single-site Bose- Hubbard (SSBH) model is introduced to describe the bosonic system-reservoir interaction [27, 28]. The nonlinear Bose-Hubbard coupling is found to be crucial to exhibit novel steady state behaviors. Such an interaction can be realized by Kerr interaction in quantum optics [29], by tuning the qubit to the dispersive regime in circuit quantum electrodynamics [30, 31], and by Fermi-Pasta-Ulam inter- action in phononic lattice [20] and dimer [32]. Though the steady state current in SSBH has been Preprint submitted to Elsevier September 25, 2018 TL l 1-l TR Figure 1: (Color online) The schematic description of a nonequilibrium sing-site Bose-Hubbard system. The red and blue half pearls are two thermal baths, characterized by temperatures TL and TR, respectively; the central green lines shows the bosonic junction, with li the occupation state; two purple arrowed circles demonstrate interactions between the junction and thermal baths. analyzed [28], the corresponding higher cumulants of energy current are much less exploited, which is important to characterize the transport features [33]. In the present work, we apply a quantum kinetic equation (QKE) to study the full counting statistics of energy current at steady state in SSBH, which is weakly coupled to two bosonic baths. The influence of the Bose-Hubbard interaction on the energy current and the noise power is systematically analyzed. Moreover, we extend the QKE to study the geometric-phase-induced energy pump, and make a comparison with the steady state flux. This paper is organized as follows: II, we introduce the SSBH model and the quantum kinetic equation combined with the counting field. The steady state population distribution is analytically expressed. In Sec. III, we study the steady state energy transfer. Energy current, the corresponding rectification and the noise power are investigated. In Sec. IV, we focus on the geometric-phase-induced energy pump. In the final section, we provide a concise conclusion. in Sec. 2. Model and method 2.1. Hamiltonian and quantum kinetic equation The Model consisting of a bosonic junction coupled to two thermal baths at Fig. 1, is expressed as H = Hs + Hb + Vsb. Specifically, the junction demonstrated as a nonlinear oscillator [27, 28], is described as Hs = ω0a†a + U a†aa†a, (1) where a† (a) creates (annihilates) one boson with frequency ω0, and U is the onsite boson-boson repulsion strength. The eigen-solution can be obtained as Hsni = Enni, with the eigenvalue En = (ω0n + U n2) (n≥0), and ni the corresponding eigenvector. Two thermal baths are described as k,v (bk,v) is the creating (annihilating) operator of phonons Hb = Pv=L,R with frequency ωk in the vth bath. The junction-bath interaction is given by Hv = Pv,k ωkb† bk,v, where b† k,v Vsb = X k,v (gk,vb† k,va + g∗ k,va†bk,v), (2) which obeys the particle conservation, with gk,v is the interacting strength. In this paper, we select ω0 as the energy unit for simplicity. Assuming the system-bath coupling strength is weak, we perturb the interaction Vsb up to the second order. Based on the Born-Marov approximation, the density operator of the whole system can 2 be approximately decoupled as ρ(t) = ρs(t)⊗ ρb, with ρs(t) the density operator of the junction. thermal baths are fully thermalized as ρb = e− Pv βv Hv /Trb{e− Pv βv Hv }, with the inverse of temperature βv = 1/kBTv. Hence, the quantum kinetic equation (QKE) of the junction is obtained as [34] d dt Pl(t) = X v +κ− [−(κ+ v,l+1 + κ− v,l)Pl(t) v,l+1Pl+1(t) + κ+ v,lPl−1(t)], where the population is Pl(t) = hlρs(t)li. The transition rates in the vth bath are κ+ l,v = Jv(∆l)nv(∆l)l, κ− l,v = Jv(∆l)(nv(∆l) + 1)l, (3) (4) where the spectral function is Jv(ω) = 2π Pk gk,v2δ(ω − ωk), the energy gap between li and l − 1i is ∆l = El − El−1 = ω0 + (2l − 1)U (l≥1), (5) and the Bose-Einstein distribution function is nv(ω) = 1/[exp(βvω) − 1]. From the dynamical picture, the rate k+ l,v describes that one boson is excited from the state l − 1i to li by absorbing one phonon from the vth bath; whereas the rate k− l,v demonstrates that one bosons is released from li to l − 1i by emitting one phonon to the vth bath. In the present work, we specify the spectral function as the typical Ohmic case Jv(ω) = Γvωe−ω/ωc,v, with the coupling coefficient Γv and the cutoff frequency ωc,v. The Ohmic bath has been widely applied to describe the molecular heat transport [35 -- 37], Kondo physics [23], quantum dissipative dynamics [34, 38 -- 40] and light-harvesting processes [41, 42]. Here, the cutoff frequency is considered very large, i.e. ωc,v≫ω0, U, kBTv. Hence, the bottom part of the bath spectrum mainly contributes to the energy transfer. The transition rates can be simplified as k+ l,v = Γv∆l(nv(∆l) + 1)l, where the factor e−∆l/ωc,v ≈1 . l,v = Γv∆lnv(∆l)l and k− 2.2. Steady state After a long time evolution, the bosonic junction is completely thermalized ( d dt Pl(t) = 0), which l Pl−1. Thus, the population at level li can be analytically results in the balanced relation k− obtained as [43, 44] l Pl = k+ Pl = P0 l Y m=1 Pv Γvnv(∆m) Pv Γv[nv(∆m) + 1] , with the population of the ground state P0 = (1 + ∞ X m=1 m Y m′=1 Pv Γvnv(∆m′ ) Pv Γv[nv(∆m′ ) + 1] )−1. (6) (7) We show the steady state population distribution at Fig. 2. In absence of the nonlinearity (U = 0), the steady state population of the bosonic junction exhibits monotonic decrease. As the onsite repulsion is turned on, the populations at high energy levels are dramatically suppressed. Particularly in the strong repulsion regime (e.g., U = 5), only populations of two lowest energy level states are finite, expressed as P0 = Pv Γv[nv(ω0 + U ) + 1] Pv Γv[2nv(ω0 + U ) + 1] P1 = Pv Γvnv(ω0 + U ) Pv Γv[2nv(ω0 + U ) + 1] , , (8) which becomes identical with the seminal nonequilibrium spin-boson model [45]. 3 n P 1 0.8 0.6 0.4 0.2 0 0 U=0 U=5 U=20 1 n 2 3 Figure 2: (Color online) Steady state population distribution Pn with various onsite boson-boson repulsion strength U . The other parameters are given by TL = 4, TR = 2 and ΓL = ΓR = 0.1. 2.3. QKE combined with counting field To analyze the steady state energy flux in the single-site Bose-Hubbard system under temperature bias, we include the full counting statistics to count the energy flow in the right bath [46]. The Hamiltonian is modified as Hχ = ei HRχ/2 He−i HRχ/2 = Hs + Hb + V χ sb, where the modified junction- bath interaction becomes V χ sb = X k,v (gk,veiωkχ/2δv,Rb† k,va + g∗ k,ve−iωkχ/2δv,R a†bk,v), (9) with δR,R = 1 and δL,R = 0. Similarly, based on the Born-Markov approximation at Eq. (3), we apply the second order perturbation theory to obtain the modified quantum kinetic equation as d dt P χ l = −(κ+ +κ+ l+1 + κ− l (χ)P χ l−1, l )P χ l + κ− l+1(χ)P χ l+1 where P χ l = hlρχ s (t)li, and the modified rates are κ+ l (χ) = X Jv(∆l)nv(∆l)le−i∆lχv δv,R , v l (χ) = X κ− Jv(∆l)(nv(∆l) + 1)lei∆lχv δv,R . v (10) (11) In absence of the counting field (χ = 0), they return back to the standard transition rate (κ± κ± l,v(χ = 0)) at Eq. (4). l,v = Next, for discussion convenience, we re-express the kinetic equation at Eq. (10) as d dt Pχi = LχPχi, 4 (12) 0.4 0.3 J 0.2 0.1 0 0 0.4 0.3 J 0.2 0.1 0 0 a) 10 20 U 30 b) γ=0.2 γ=0.5 γ=0.8 10 20 30 U Figure 3: (Color online) a) Energy flux in Ohmic baths by tuning onsite repulsion strength U with ΓL = ΓR = 0.1, the green dashed line from the result at Eq. (18) and the red dashed-dotted line from the result at Eq. (19); b) energy flux with asymmetric system-bath coupling strength with ΓL = 0.1×(1 − γ) and ΓR = 0.1×(1 + γ). The other parameters are given by TL = 4 and TR = 2. 0 , P χ with the vector Pχi = [P χ 1 , · · · ]T , and the evoluting matrix Lχ composed by the modified transition rates. Then, the generating function can be obtained as Gχ(t) = hIeLχtPχ(0)i [46], with the unit vector hI = [1, 1, · · · ], and Pχ(0)i the initial population state. In the long time limit, the generating function can be approximately given by Gχ(t)≈e−λ0(χ)t, where λ0(χ) is the eigenvalue of Lχ owing the maximal real part, and λ0(χ = 0) = 0. Consequently, the cumulant generating function at steady state is obtained as Z(χ) = limt→∞ Gχ(t)/t = λ0(χ). And the nth cumulant of the energy flux can be expressed as J (n) = ∂nλ0(χ) ∂(iχ)n χ=0. Particularly, the energy flux is the lowest cumulant case and the noise power is the second lowest cumulant case J = ∂λ0(χ) ∂(iχ) χ=0, J (2) = ∂2λ0(χ) ∂(iχ)2 χ=0. (13) (14) (15) 3. Steady state energy transfer In this section, we study the steady state energy flux, the energy rectification and the noise power. 5 a) 0.1 0.05 0 −0.05 −0.1 J R =3,γ=0.5 T 0 U=0.5 U=6 0 0.5 1 ∆T 1.5 2 0.1 0.05 b) J R 0 −0.05 −0.1 0 =3,∆T=1 T 0 U=0.5 U=6 0.2 0.4 γ 0.6 0.8 1 Figure 4: (Color online) Energy rectification of the steady state flux (RJ) by tuning: (a) temperature bias between thermal baths; (b) asymmetric junction-bath coupling factor. The other parameters are given by TL = T0 + ∆T , TR = T0 − ∆T , ΓL = 0.1×(1 − γ) and ΓR = 0.1×(1 + γ). 3.1. Energy flux Based on the eigen solution LχPχi = λ0(χ)Pχi, the energy flux can be alternatively obtained as J = hI ∂Lχ ∂(iχ) χ=0Pχ=0i (16) = ∞ X l=1 ∆l(κ− l,RPl − κ+ l,RPl−1), with κ± l,R and Pl given at Eq. (4) and Eq. (6), respectively. We firstly study the influence of the onsite repulsion on the steady state energy transfer with Ohmic baths at Fig. 3(a). In absence of repulsion strength (U = 0), the energy gap at Eq. (5) becomes level independent as ∆l = ω0, and the transition rates at Eq. (4) are simplified to κ+ l,v = Jv(ω0)nv(ω0)l and κ− l,v = Jv(ω0)(nv(ω0) + 1)l. Hence, the expression of heat current for the ballsitic transfer is given by J = ω2 [nL(ω0) − nR(ω0)]. Then, we tune on the repulsion strength. In the weak repulsion 0 regime, it is found that energy flux is enhanced by increasing the onsite repulsion strength, shown at Eq. 3(a). To give an analytical picture, we include the mean-field approximation to simplify boson- boson repulsion at Eq. (1) as a†aa†a≈2nsa†a−n2 s, with ns = ha†ai≈[ΓLnL(ω0)+ΓRnR(ω0)]/(ΓL +ΓR) obtained in the limit of U = 0. It results in the mean-field version of the system Hamiltonian ΓLΓR ΓL+ΓR H s = (ω0 + 2U ns)a†a − U n2 s. (17) After this treatment, the Hamiltonian becomes quadratic. Hence, the expression of heat current can be directly obtained as J = (ω0 + 2U ns)2 ΓLΓR ΓL + ΓR [nL(ω0 + 2U ns) − nR(ω0 + 2U ns)], (18) which clearly exhibits the monotonic behavior by increasing the repulsion strength U (dahsed green line at Fig. 3(a)). 6 While at the intermediate repulsion regime (e.g., U = 5), it is known that high energy state populations have already been dramatically suppressed, shown at Fig. 2. Hence, we may include two lowest states to approximately analyze the behavior of the flux in the moderate/strong repulsion regimes. The expression of the energy flux is given by J = (ω0 + U )2ΓR[(nR(ω0 + U ) + 1)P1 − nR(ω0 + U )P0]. Combined with Eq. (8), it can be specified as J = (ω0 + U )2ΓLΓR nL(ω0 + U ) − nR(ω0 + U ) Pv Γv[2nv(ω0 + U ) + 1] . (19) The energy flux (red dashed line) is found to exhibits clearly non-monotonic behavior, shown at Fig. 3(a). Hence, we conclude the monotonic behavior of the heat current can be observed under the influence of the onsite boson-boson repulsion. Then, we study the asymmetric effect of the junction-bath coupling strength on the behavior of energy flux at Fig. 3(b) [28]. It is found that by increasing the asymmetric factor γ, the energy flux is monotonically suppressed. We conclude that the choice of identical junction-bath coupling strengthes is helpful to strengthen the steady state energy flux. 3.2. Rectification We analyze the energy flux rectification, defined as RJ = J(∆T, γ) − J(−∆T, γ) J(∆T, γ = 0) , (20) with J(∆T, γ) = ∂λ0(χ) preferred in one dimension over the opposite [47]. ∂(iχ) χ=0, and ∆T = TL − TR, which describes an effect that energy transfer is We firstly study the influence of the temperature bias (∆T ) on the rectification at Fig. 4(a). In the weak repulsion regime (e.g., U = 0.5), the rectification is negatively amplified by enlarging the temperature bias. On the contrary, for strong repulsion strength (e.g., U = 6), the rectification is changed to be positively enhanced. Then, we analyze the asymmetric effect of system-bath interactions on the rectification at Fig. 4(b). It is found that for the weak repulsion strength, the rectification exhibits a global valley with the modulated asymmetric factor. While in the strong repulsion regime, the rectification exhibits a positive peak. Particularly, the rectification can be expressed as RJ = 4ΓLΓR[nL + nR + 1] Γ0[Pv Γv(2nv + 1)][Pv Γv(2n¯v + 1)] (ΓL − ΓR)(nR − nL), × (21) It is clearly exhibited that (ΓL − ΓR)(nR − nL)6=0 results in the emergence of the positive recti- fication [48]. Moreover, as the asymmetric factor approaches 1, the rectification effect disappears regardless of the on-site repulsion strength. It is known that in this limiting case, one system-bath coupling strength becomes zero, e.g., ΓL = 0 in the Fig. 4. Hence, there is no steady state currents as defined in Eq. (20) (J(±∆T, γ = 1) = 0), which results in RJ = 0. 3.3. Noise power We study the influence of the onsite repulsion strength on the noise power, which is defined as J (2) = ∂2λ0(χ)/∂(iχ)2χ=0 from Eq. (15). Under symmetric junction-bath coupling condition (ΓL = ΓR), it is found that noise power becomes robust in the weak and mediate coupling regimes, whereas it is strongly suppressed in the strong repulsion regime. The main reason is quite similar to the energy flux, that the large energy gap significantly blocks the state transition between nearest-neighboring states. Then, we tune on the asymmetric coupling factor γ. The noise power is monotonically suppressed by the factor, clearly shown in Fig. 5. Therefore, the asymmetric coupling condition deteriorates the generation of the noise power. 7 r e w o P e s o N i 4 3 2 1 0 0 γ=0 γ=0.2 γ=0.5 γ=0.8 Γ =0.1 0 10 U 20 30 Figure 5: (Color online) Relation of noise power with the repulsion strength, with various asymmetric junction-bath coupling factor. The other parameters are given by TL = 4, TR = 2, ΓL = Γ0×(1 − γ) and ΓR = Γ0×(1 + γ). m o e g Q 0.02 0 −0.02 −0.04 −0.06 0 γ=0 γ=0.2 γ=0.5 γ=0.8 Γ =0.1 0 =3,∆T=1 T 0 5 U 10 15 Figure 6: (Color online) Influence of the repulsion strength on geometric-phase-induced energy pump Qgeom = Tp×Jg, with various asymmetric junction-bath coupling factor. The other parameters are given by TL = T0 + ∆T cos Ωτ , TR = T0 + ∆T sin Ωτ , ΓL = Γ0×(1 − γ) and ΓR = Γ0×(1 + γ). 8 4. Adiabatic energy pump The time-dependent control of heat transfer in junction systems has attracted dramatic atten- tion [20, 49, 50]. In particular, the nonlinear effect is revealed to be crucial to exploit the adiabatic quantum pump. Therefore, we apply the geoemtric-phase-like modulation to study the influence of repulsion strength on the energy transfer in the single-site Bose-Hubbard model. As the bosonic junction is adiabatically modulated by external fields (e.g., TL(R)(t) in two baths), the evoluting matrix becomes time dependent Lχ(t). Then, the generating function is composed by two components as [49, 51] lim t→∞ = eGχ(t) = exp([Gd(χ) + Gg(χ)]t), Tp ´ Tp (22) where Gχ is the cumulant generating function. Gd(χ) = − 1 0 dtλ0(χ, t) is the dynamical con- tribution, with Tp the modulating period. It straightforwardly results in the dynamical energy flux Jdyn = ∂ ∂(iχ) Gd(χ)χ=0. In absence of the external modulation, it reduces to the steady state flux from Eq. (14). Gg(χ) is the geometric contribution, which is expressed as Gg(χ) = − 1 Tp Tp 0 dthφχ(t) ∂ ∂t ψχ(t)i, (23) where ψχ(t) (hφχ(t)) is the right (left) eigenvector, corresponding to λ0(χ, t). Assuming the driving fields are expressed as u1(t) and u2(t), the geometric phase induced energy flux is specified as [52, 53] Gg(χ) = − 1 Tp (du1hφχ ∂ ∂u1 ψχi + du2hφχ ∂ ∂u2 ψχi). According to the Stocks theorem, it can be re-expressed as [49, 50, 54] Gg(χ) = − 1 Tp u1,u2 du1du2Fχ(u1, u2), with the Berry-like curvature Then, the geometric-phase-induced current is given by Fχ(u1, u2) = h∂u1 φχ∂u2 ψχi − h∂u2 φχ∂u1 ψχi. Jg = ∂Gg(χ) ∂(iχ) χ=0 = − 1 Tp u1,u2 du1du2 ∂ ∂(iχ) Fχ(u1, u2)χ=0. (24) (25) (26) (27) We firstly study the geometric-phase-induced energy pump under the symmetric junction-bath coupling condition (γ = 0) at Fig. 6. In absence of the on-site repulsion (U = 0), there is no geometric- phase-induced pump, due to the fully harmonic behavior of the molecular junction. From the previous works, it is proposed that nonlinearity of the system is crucial to exhibit the geometric contribution of the flux [49, 50]. Hence, the present result is consistent with the proposal. Moreover, in the weak repulsion regime the energy pump shows negative enhancement. It is in sharp contrast to the behavior of the steady state flux at Fig. 3(a), which shows positive robustness. As is known that steady state flux reflects the behavior of quasi-energy spectrum of the Liouvillian matrix at Eq. (12), whereas geometric flux is sensitive to modulations of the corresponding eigen-states. Therefore, due to different information they capture, the influence of the repulsion strength on the geometric-phase- induced energy flux is significantly distinct from the counterpart for the steady state flux. Then, it rises quickly and reaches the globally positive peak with intermediate repulsion strength. As the repulsion strength further increases, the energy pump is suppressed and asymptotically approaches zero. Next, we tune on the junction-bath coupling asymmetric factor. It is found that the energy pump is monotonically suppressed by enlarging the asymmetric factor. Hence, we conclude that the asymmetric factor plays a similar role in both the steady state and geometric energy flux. 9 5. Conclusion In summary, we study the quantum energy transfer in a single-site bosonic junction weakly coupled to two thermal baths with temperature bias. The quantum kinetic equation combined with the full counting statistics, is included to analyze the energy current and corresponding current fluctuations (e.g., noise power). Steady state population distribution is analytically obtained. Particularly, in strong onsite boson-boson repulsion regime, two lowest energy level states mainly contribute to steady state behaviors. Without the adiabatic modulation, in weak onsite boson-boson repulsion regime, steady state energy flux with Ohmic baths is enhanced by increasing the replusion stregnth. It is analytically explained based on the mean-field scheme. Then, the current shows maximal with inter- mediate repulsion strength, whereas it asymptotically approaches to zero in the strong repulsion limit, which is similar to the behavior of the nonequilibrium spin-boson model. Moreover, we study the energy flux rectification by tuning the temperature bias and asymmetric coupling factor. It is found that the behavior of energy rectification with weak repulsion strength is different from the counterpart in the strong repulsion regime. With the adiabatic modulation, the geometric-phase-induced energy flux shows significant distinction from the steady state counterpart. Specifically, the geometric-phase- induced energy flux shows negative enhancement in the weak onsite repulsion regime. Moreover, it shows reversal behavior and becomes positive maximal with intermediate repulsion strength. As the re- pulsion strength further enlarges, the geometric-phase-induced energy flux is suppressed and gradually decays to zero. In previous works [55, 56], the nonequilibrium green function approach was exploited to investigate the quantum heat transfer in the anharmonic boson junction with strong junction-bath interaction, which dramatically affects the behavior of heat current and high order fluctuations. Hence, we may include this novel method to analyze the influence of interplay between strong junction-bath coupling and on-site boson-boson interaction on the heat transfer. 6. Acknowledgements C.W. is supported by the National Natural Science Foundation of China under Grant No.11704093, No. 11547124 and No. 11574052, and K.W.S. is supported by the National Natural Science Foundation of China under Grant No. 11404084. ∗ Corresponding author. Email:wangchenyifang@gmail.com References [1] T. Prosen and I. Pizorn, Phys. Rev. Lett. 101 (2008) 105701. [2] M. A. Ratner, Nat. Nanotechnology 8 (2013) 378. [3] S. Hild, T. Fukuhara, P. Schaub, J. Zeiher, M. Knap, E. Demler, I. Bloch and C. Gross, Phys. Rev. Lett. 113 (2014) 147205. [4] T. Prosen, Phys. Rev. Lett. 112 (2014) 030603. [5] J. Eisert, M. Friesdorf and C. Gogolin, Nature Physics 11 (2015) 124. [6] Y. Dubi and M. Di Ventra, Rev. Mod. Phys. 83 (2011) 131. [7] H. Haug and A. P. Jauho, Quantum Kinetics in Transport and Optics of Semiconductors (Springer, Berlin, 2007). [8] R. Williams, J. Chem. Phys. 32 (1960) 1505. [9] K. E. Dorfman, D. V. Voronine, S. Mukamel and M. O. Scully, Proc. Natl. Aca. Sci. USA 110 (2013) 2746. 10 [10] C. R. Xu and M. G. Vavilov, Phys. Rev. B 87 (2013) 035429. [11] D. Z. Xu, C. Wang, Y. Zhao and J. S. Cao, New. J. Phys. 18 (2016) 023003. [12] G. J. Snyder and E. S. Toberer, Nature Materials 7 (2008) 105. [13] C. V. den Broeck, Phys. Rev. Lett. 95 (2005) 190602. [14] H. S. Kim, W. S. Liu, G. Chen, C. W. Chu and Z. F. Ren, Proc. Natl. Aca. Sci. USA 112 (2015) 8250. [15] J. Kondo, Progress of Theoretical Physics 32 (1964) 37. [16] J. W. Park, A. N. Pasupathy, J. I. Goldsmith, C. Chang, Y. Yaish, J. R. Petta, M. Rinkoski, J. P. Sethna, H. D. Abruna, P. L. McEuen and D. C. Ralph, Nature 417 (2002) 722. [17] J. S. Cao and R. J. Silbey, J. Phys. Chem. A 113 (2009) 13825. [18] J. L. Wu, F. Liu, Y. Shen. J. S. Cao and R. J. Silbey, New J. Phys. 12 (2010) 105012. [19] L. F. Zhang, J. Ren, J. S. Wang and B. W. Li, Phys. Rev. Lett. 105 (2010) 225901. [20] N. B. Li, J. Ren, L. Wang, G. Zhang, P. Hanggi and B. W. Li, Rev. Mod. Phys. 84 (2012) 1045. [21] N. Lambert, Y. N. Chen, Y. C. Cheng, C. M. Li, G. Y. Chen and F. Nori, Nature Physics 9 (2013) 10. [22] D. W. Wang, H. T. Zhou, M. J. Guo, J. X. Zhang, J. Evers and S. Y. Zhu, Phys. Rev. Lett. 110 (2013) 093901. [23] K. Saito and T. Kato, Phys. Rev. Lett. 111 (2013) 214301. [24] C. C. Chien, S. Peotta and M. Di Ventra, Nature Physics 11 (2015) 998. [25] J. P. Lv, G. Chen, Y. J. Deng and Z. Y. Meng, Phys. Rev. Lett 115 (2015) 037202. [26] X. Huo, Y. Y. Cui, D. L. Wang and J. P. Lv, Phys. Rev. A 95 (2017) 0223613. [27] F. Hakke, H. Risken, C. Savage and D. Walls, Phys. Rev. A 34 (1986) 3969. [28] A. Purkayastha, A. Dhar and M. Kulkarni, Phys. Rev. A 94 (2016) 052134. [29] P. Weinberger, Phil. Magazine Lett. 88 (2008) 897. [30] Y. Yin, H. Wang, M. Mariantoni, R. C. Bialczak, R. Barends, Y. Chen, M. Lenander, E. Lucero, M. Neeley, A. D. O'Connell, D. Sank, M. Weides, J. Wenner, T. Yamamoto, J. Zhao, A. N. Cleland and J. Martinis, Phys. Rev. A 85 (2012) 023826. [31] T. Liu, C. P. Yang, Y. Zhang, C. S. Yu and W. N. Zhang, arXiv:1611.06749. [32] J. Thingna, J. L. García-Palacios and J. S. Wang, Phys. Rev. B 85, 195452 (2012). [33] M. Campisi, P. Hanggi and P. Talkner, Rev. Mod. Phys. 83 (2011) 771. [34] U. Weiss, Quantum Dissipative Systems (World Scientific, Singapore, 2008). [35] D. Segal, Phys. Rev. E 90 (2014) 012148. [36] E. Taylor and D. Segal, Phys. Rev. Lett. 114 (2015) 220401. [37] D. Segal and B. K. Agarwalla, Annual Review of Physical Chemistry 67 (2016) 185. 11 [38] A. J. Leggett, S. Chakravarty, A. T. Dorsey, M. P. A. Fisher, A. Garg and W. Zwerger, 1987 Rev. Mod. Phys. 59 1. [39] T. Vorrath and T. Brandes, Phys. Rev. Lett. 95 (2005) 070402. [40] L. W. Duan, H. Wang, Q. H. Chen and Y. Zhao, J. Chem. Phys. 139 (2013) 044115. [41] L. A. Pachon and P. Brumer, Phys. Chem. Chem. Phys. 14 (2012) 10094. [42] M. Mohseni, Y. Omar, G. S. Engel and M. B. Plenio, Quantum effects in biology (Cambridge University Press, 2014). [43] M. Vogl, G. Shcaller and T. Brandes, Ann. Phys. 326 (2011) 2827. [44] M. Vogl, G. shcaller and T. Brandes, Phys. Rev. A 86 (2012) 033820. [45] D. Segal and A. Nitzan, Phys. Rev. Lett. 94 (2005) 034301. [46] M. Esposito, U. Harbola and S. Mukamel, Rev. Mod. Phys. 81 (2009) 4. [47] B. W. Li, L. Wang and G. Casati, Phys. Rev. Lett. 93 (2004) 184301. [48] L. A. Wu and D. Segal, Phys. Rev. Lett. 102 (2009) 095503. [49] J. Ren, P. Hanggi and B. W. Li, Phys. Rev. Lett. 104 (2010) 170601. [50] J. Ren, S. Liu and B. W. Li, Phys. Rev. Lett. 108 (2012) 210603. [51] N. A. Sinitsyn and I. Nemenman, Europhys. Lett. 77 (2007) 58001. [52] M. V. Berry, Proc. R. Soc. London A 392 (1984) 45. [53] D. Xiao, M. C. Chang and Q. Niu, Rev. Mod. Phys. 82 (2010) 1959. [54] C. Wang, J. Ren and J. S. Cao, Phys. Rev. A 95 (2017) 023610. [55] J. S. Wang, J. Wang and J. T. Lü, Eur. Phys. J. B 62, 281 (2008). [56] H. N. Li, B. K. Agarwalla, B. W. Li and J. S. Wang, Eur. Phys. J. B 86, 500 (2013). 12 

1605.00148

1

1605

Plasmons in a Planar Graphene Superlattice

Plasmon collective excitations are studied in a planar graphene superlattice formed by periodically alternating regions of gapless graphene and of its gapped modification. The plasmon dispersion law is determined both for the quasi-one-dimensional case (the Fermi level is located within the minigap) and for the quasi-two-dimensional case (the Fermi level is located within the miniband). The problem concerning the absorption of modulated electromagnetic radiation at the excitation of plasmons is also considered.

cond-mat.mes-hall

cond-mat

ISSN 0021-3640, JETP Letters, 2015, Vol. 102, No. 11, pp. 713 -- 719. c(cid:13) Pleiades Publishing, Inc., 2015. Original Russian Text c(cid:13) P. V. Ratnikov, A. P. Silin, 2015, published in Pis'ma v Zhurnal Eksperimental'noi i Teoreticheskoi Fiziki, 2015, Vol. 102, No. 11, pp. 823 -- 829. CONDENSED MATTER Plasmons in a Planar Graphene Superlattice aLebedev Physical Institute, Russian Academy of Sciences, Leninskii pr. 53, Moscow, 119991 Russia P. V. Ratnikova and A. P. Silina, b bMoscow Institute of Physics and Technology (State University), Institutskii per. 9, Dolgoprudnyi, Moscow region, 141700 Russia e-mail: ratnikov@lpi.ru Received July 3, 2015; in final form, September 16, 2015 Plasmon collective excitations are studied in a planar graphene superlattice formed by periodically al- ternating regions of gapless graphene and of its gapped modification. The plasmon dispersion law is determined both for the quasi-one-dimensional case (the Fermi level is located within the minigap) and for the quasi-two-dimensional case (the Fermi level is located within the miniband). The problem concerning the absorption of modulated electromagnetic radiation at the excitation of plasmons is also considered. DOI: 10.1134/S0021364015230137 1. INTRODUCTION Graphene (a two-dimensional carbon material) has been actively studied both theoretically and experi- mentally for more than ten years. In recent years, graphene nanostructures have become a forefront is- sue. The usage of collective excitations (plasmons) in these systems promises new advantages for the tunable absorption of electromagnetic radiation. The plasmon- induced enhancement of light absorption within the middle infrared range was observed for the heterostruc- ture formed by graphene strips [1]. The plasmon-type oscillations in spatially uniform systems with different dimensionalities having charge carriers with a linear dispersion law were studied in [2], where the tunneling of charge carriers was neglected. Such approximation is similar to the tight-binding ap- proximation in the band structure theory for crystals. In [3], the plasma oscillations of massless Dirac elec- trons in a planar superlattice were studied. The Dirac plasma was assumed to be weakly modulated. This picture is similar to the weak-binding approximation. The spectrum of plasma oscillations and the related absorption intensity for electromagnetic waves were de- termined by the methods of electrodynamics of contin- uous media. In this paper, we present the calculations of the plasmon dispersion law in planar graphene superlat- tices. The superlattices under study are formed by al- ternating strips of gapless graphene and of its gapped modifications. The latter can be produced using the main property of graphene, namely, its two-dimensiona- lity. For this, there exist two possible ways: (i) choos- 1 ing the material of the substrate on which graphene is deposited and (ii) depositing atoms or molecules, e.g., hydrogen atoms [4] or CrO3 molecules [5] on the surface of a graphene sheet. Several gapped modifi- cations of graphene with the band gap ranging from about 10 meV to 1 eV have been already obtained. In the superlattice under study, the charge carriers effectively acquire a nonzero mass. Their dispersion law becomes nonlinear. The system is similar to a rel- ativistic plasma in a low-dimensional space. Plasma waves in the graphene superlattice in the presence of a high dc electric field were recently studied in [6] in the random phase approximation. The same authors [7] studied numerically the plasmon dispersion law in the planar graphene superlattice. In our work, we obtain explicit analytical results for the plasmon dispersion law in the planar graphene superlattice. 2. EFFECTIVE MODEL DESCRIPTION OF THE SUPERLATTICE 2.1. Fundamentals of the Model Description of the Superlattice The main concepts concerning the planar superlat- tices based on gapless graphene and on its gapped mod- ifications were reported in [8]. Let x and y axes be normal and parallel to the interfaces, respectively (Fig. 1). The charge carriers in a superlattice are described by the Dirac equation (vFσ(cid:98)p + σz∆ + V ) Ψ(x, y) = EΨ(x, y), (1) where vF ≈ 108 cm/s is the Fermi velocity; σ = (σx, σy) and σz are the Pauli matrices; and(cid:98)p = −i∇ is the mo- 6 1 0 2 r p A 0 3 ] l l a h - s e m . t a m - d n o c [ 1 v 8 4 1 0 0 . 5 0 6 1 : v i X r a Fig. 1. Example of an array under study: graphene sheet on SiO2 substrate with hydrogen atoms periodically deposited on graphene strips (graphene -- graphane superlattice). mentum operator (we use units with  = 1). The half- width ∆ of the band gap and the work function V are periodically modulated along the x axis (cid:40) (cid:40) ∆ = V = d(n − 1) < x < −dII + dn, 0, ∆0, −dII + dn < x < dn, 0, V0, −dII + dn < x < dn. d(n − 1) < x < −dII + dn, (2) where n is an integer enumerating the superlattice su- percells; dI and dII are the widths of strips of gapless and gapped graphene, respectively; and d = dI + dII is the period of the superlattice (see Fig. 1). The profile of the potential is depicted in Fig. 2. In this work, we assume that vF has the same value over the whole superlattice. In [9], we considered a new type of superlattice with alternating Fermi velocity. The motion of charge carriers along the y axis is free and the wavefunction has the form Ψ(x, y) = ψ(x)eikyy. The dispersion relation for decaying solution (1) within the potential barriers has the form [8] Fk2 v2 2 − v2 Fk2 1 + V 2 2v2 Fk1k2 0 − ∆2 0 sinh(k2dII) sin(k1dI) (3) + cosh(k2dII) cos(k1dI) = cos(kxd), where k1 and k2 are related to the energy E by the formulas (cid:113) E = V0 ±(cid:113) E = ±vF y + k2 k2 1, ∆2 0 + v2 Fk2 y − v2 Fk2 2. (4) For the further analysis, it is difficult to use the exact spectrum of charge carriers determined by find- ing the numerical solution of Eq. (3). We suggest using the effective spectrum (the spectrum of a model narrow-gap semiconductor). 2.2. Effective Theory We should distinguish two cases: (i) the Fermi level falls within one of the minigaps and (ii) the Fermi level is located within one of the minibands. In the former case, all minibands lying below the Fermi level are completely occupied and the oscillations Fig. 2. Model periodic one-dimensional Kronig -- Penney potential for the superlattice under study. of the electron (hole) density occur only in the direc- tion of the free motion of charge carriers (along the normal to the direction of the voltage applied across the superlattice). This is a quasi-one-dimensional mo- tion. In the latter case, the miniband containing the Fermi level is occupied only partially, whereas all lower bands (if such bands exist) are completely occupied. In the partially occupied miniband, the oscillations of elec- tron (hole) density can also occur along the direction of the voltage applied across the superlattice. This is a quasi-two-dimensional motion. Then, for simplicity, we consider the situation with the filling (complete or partial) of only one lowest elec- tron miniband or the highest hole miniband. 2.2.1. Quasi-one-dimensional case (complete- ly occupied miniband). At sufficiently large values of ∆0 and dII, the minibands are rather narrow (we shall specify this condition below). In this case, the energy spectrum of charge carriers is similar to that characteristic of a quasi-one-dimensional narrow-gap semiconductor E ≈ Veff ±(cid:113) ∆2 eff + v2 Fk2 y. (5) The parameters ∆eff and Veff play the role of the effec- tive band gap and the effective work function, respec- tively. The charge carriers have the effective mass m∗ = ∆eff v2 F . (6) Using dispersion relation (3) and assuming that Veff < ∆eff (cid:28) ∆0, we can easily deduce the follow- ing estimates for ∆eff and Veff: (cid:20) (cid:21) 1 − vF dI∆0 ∆eff = Veff = πvF 2dI vF dI∆0 V0. , (7) In the case under study, the minibands have an exponentially small width owing to an exponentially small probability for charge carriers to tunnel through the barriers. In this limit, we obtain the following es- timate for the miniband width: (cid:19) (cid:18) − dII vF ∆0 . (8) δE = 4vF dI exp 2 The condition defining the narrow minibands is δE (cid:28) ∆eff. Comparing the expression for ∆eff in Eqs. (7) with Eq. (8), we find the condition ∆0 (cid:38) 2vF/dII. Let us write the effective Hamiltonian correspond- ing to the approximate dispersion law given by Eq. (5) as the Dirac Hamiltonian in terms of 2×2 matrices (cid:98)H (1D) eff = vFσy(cid:98)py − σz∆eff + Veff. (9) Here, the minus sign in front of the second term is placed for convenience of further calculations. This does not affect the final results since there ∆eff is squared. In the zeroth order approximation, the Green's func- tion describing the free propagation of charge carriers along the gapless graphene strips has the form of the inverse operator [10] (cid:104) ω + µ − (cid:98)H (1D) eff (cid:105)−1 (ky, ω) = , (10) (cid:98)G(1D) 0 where µ is the chemical potential (coincides with the Fermi energy). Substituting Eq. (9) into operator (10), we can ex- plicitly write the Green's function taking into account the rules of path tracing around the poles (ky, ω) = 1 2εky ω +(cid:101)µ − σz∆eff + vFσyky ω +(cid:101)µ − sεky − iδ sgn((cid:101)µ − sεky ) s s=±1 (11) , where (cid:101)µ = µ − Veff and εky = y, δ → +0. The value of is related to the Fermi momentum pF eff + v2 Fk2 ∆2 (cid:113) (cid:98)G(1D) × (cid:88) 0 as follows: ∆2 eff + v2 Fp2 F. (12) (cid:113) (cid:101)µ = The one-dimensional Fermi momentum is expressed in terms of the charge carrier density pF = π g n2Dd, (13) where g = gsgv is the degeneracy order (gs = 2 is the spin degeneracy order and gv = 2 is the valley degeneracy order). 2.2.1. Quasi-two-dimensional case (partially occupied miniband). In the quasi-two-dimensional case, in addition to the free motion along the gapless graphene strips, charge carriers move across the po- tential barriers. These types of motion occur at dif- ferent velocities: at v(cid:107) for the free motion and at a much lower velocity v⊥ (cid:28) v(cid:107) (since the probability of tunneling through the potential barrier is small). This means the quasi-two-dimensional anisotropic motion of charge carriers. The corresponding values of v⊥ and v(cid:107) are selected by fitting the approximate dispersion law E ≈ Vef f ±(cid:113) ∆2 ef f + v2⊥k2 x + v2(cid:107)k2 y. (14) The energy spectrum is similar to that of an anisotropic narrow-band semiconductor with the effective masses ⊥ = ∆eff/v2⊥ (cid:54)= m∗ m∗ (cid:107) = ∆eff/v2(cid:107). The temperature should be sufficiently low, T (cid:28) δE. 3 The effective Hamiltonian with eigenvalues (14) has The Green's function is determined as inverse op- erator (10) with the Hamiltonian the form(cid:98)H (2D) eff = v⊥σx(cid:98)px + v(cid:107)σy(cid:98)py − σz∆eff + Veff. (cid:98)G(2D) × (cid:88) ω +(cid:101)µ − σz∆eff + v⊥σxkx + v(cid:107)σyky ω +(cid:101)µ − sεk − iδ sgn((cid:101)µ − sεk) (cid:113) (k, ω) = 1 2εk s=±1 s 0 where εk = ∆2 eff + v2⊥k2 x + v2(cid:107)k2 y. (15) (16) , 3. PLASMONS 3.1. Coulomb Interaction In the quasi-one-dimensional case, the charge car- riers do not move between the gapless graphene strips. The Coulomb interaction is similar to that for charge carriers in a periodic planar array formed by parallel filaments. In such array, the Coulomb interaction of charges located at two filaments separated by the dis- tance nd reads [11] V (ky, n) = 2(cid:101)e2K0 (dnky) , (17) where d is the distance between the gapless graphene strips (it coincides with the period of the superlattice); n is the number of a strip (it can be considered as that coinciding with the number of a supercell in the superlattice shown in Fig. 2);(cid:101)e2 = e2/εeff, where εeff = (ε1 + ε2)/2 is the effective static dielectric constant determined by the static dielectric constants ε1 and ε2 of the media surrounding the graphene (e.g., vacuum and the substrate material); and K0(x) is the modified Bessel function of the second kind. Now, we can make the transformation from the dis- crete variable n denoting the strip number to the di- mensionless transverse momentum θ = kxd (−π ≤ θ ≤ π), as was done in [11] V (ky, θ) = V (ky, n)einθ ∞(cid:88) (cid:19) + 4(cid:101)e2 n=−∞ ∞(cid:88) n=1 = 2(cid:101)e2K0 (cid:18) dI 2 ky cos(nθ)K0 (ndky) . (18) In the case of the narrow barrier, which is of main interest to us, expression (18) becomes simpler (dII (cid:28) dI) [11] V (ky, θ) = 2(cid:101)e2 ln (cid:20) (cid:18) θ −2C − 2ψ + d πdI + 1 2 2π (cid:19) + π tan (cid:21)(cid:101)e2 + o(kyd), θ 2 (19) where C = 0.577 . . . is the Euler constant and ψ(x) is the Euler ψ function. At the miniband boundaries, we have V (ky, ±π) = 2(cid:101)e2 ln 2π(cid:101)e2 kyd d πdI + + o(kyd). (20) 3.2. Polarization Operator In the calculations of the plasmon frequencies us- ing the diagram technique, we should distinguish two specific cases: (i) the quasi-one-dimensional isotropic case (the corresponding Green's function is determined in Subsection 2.2.1) and (ii) the quasi-two-dimensional anisotropic case (the corresponding Green's function is determined in Subsection 2.2.2). Hence, we have two expressions for the polarization operator needed for finding the plasmon dispersion law. 3.2.1. Quasi-one-dimensional polarization op- erator. The polarization operator is represented by the loop diagram (Fig. 3) and is given by the expres- sion (cid:90) dε 2π × (cid:110)(cid:98)G(1D) 0 Π(1D)(ky, ω) = −ig (py, ε)(cid:98)G(1D) 0 Tr 2π (py + ky, ε + ω) (cid:111) . (21) (cid:90) dpy (cid:12)(cid:12)(cid:12)n2D→0 Similar to the situation in quantum electrodynamics (QED), expression (21) should be renormalized. How- ever, the many-body problem in solids has its specific features. Although the bare electron and hole spec- tra are identical to those of electrons and positrons in QED, the set of parameters and the laws involved in these renormalizations are different [12, 13]. The renormalization of the polarization operator is reduced to the condition [14] Ren (ky, ω) = Π(1D)(ky, ω) − Π(1D)(ky, ω) Π(1D) (22) We are interested in plasmons (the long-wavelength collective excitations); therefore, it is sufficient to de- termine the polarization operator at low ky and ω val- ues: ky (cid:28) ∆eff vF , ω (cid:28) ∆eff. (23) As we can see below, the plasmon frequencies are low because of low ky values (the plasmon dispersion law for low-dimensional systems). In the quasi-one-dimensional case, the real part of the renormalized polarization operator at low crystal momenta (the expansion is performed up to terms of the order of k2 y) and frequencies specified by Eqs. (23) is given by the expression Re Π(1D) (cid:40) Ren (ky, ω) = −Λ1 + (cid:101)µ + ky g 2π 2v2 × where Λ1 = 1 vF Fk2 y v4 Fp3 3(cid:101)µ3∆2 eff FpFk2 y (cid:101)µω2 + (cid:101)µ + vFpF (cid:101)µ − vFpF ln (cid:41) , (24) . (25) Note that Λ1 is positive, is independent of both ky and ω, and appears in Eq. (24) with the negative sign. It easy to see that this term in Eq. (24) results in a pronounced (background) screening. Hence, Λ1 should . Re Π(1D) Ren (ky, ω) = gv2 FpFk2 y π(cid:101)µω2 . (26) The imaginary part of Π(1D) Ren (ky, ω) vanishes within the range vFky < ω < (cid:113) 4∆2 eff + v2 Fk2 y, (27) which is in agreement with the well-known result for relativistic plasma [15]. 3.2.2. Quasi-two-dimensional polarization op- erator. In the quasi-two-dimensional anisotropic case, the polarization operator is represented similarly to Eq. (21) as (cid:90) dε 2π × Π(2D)(k, ω) = −igd (cid:110)(cid:98)G(2D) 0 (p, ε)(cid:98)G(2D) 0 Tr (cid:90) d2p (2π)2 (p + k, ε + ω) (cid:111) (28) . Fig. 3. Loop diagram. be omitted. In addition, in the limit ∆eff → 0, it leads to a logarithmic divergence. The term linear in ky actually turns out to be small in comparison to the third term, which is for- mally of the order of k2 In the denominator of the y. third term, we have ω2, and ω (cid:38) vFky, but it is still of the same order as vFky. Therefore, in spite of the formally higher order of the third term, it turns out to be larger than the second one. Expression (24) is derived under the assumption that vFky (cid:46) ω (cid:28) ∆eff. At the same time, we assume that vFpF (cid:28) ∆eff (the case of a low charge carrier den- sity) or, at least, vFpF (cid:46) ∆eff (the case of a moderate charge carrier density). Therefore, in contrast to the third term, the last term in Eq. (24) is smaller than or of the order of ω2/∆2 eff in the case of a moderate charge carrier density and of a higher order in the case of a low charge carrier density. Therefore, we can neglect the last term in Eq. (24). Finally, we find The renormalization condition in the form of Eq. (22) should also be imposed on polarization operator (28). At low crystal momenta (we retain the terms of the order of k2 y) and low frequencies specified by Eqs. (23), the real part of the renormalized polar- ization operator has the form x and k2 (cid:40) (cid:18) −Λ2 + v2⊥k2 1 − 3∆eff 2(cid:101)µ + (cid:19)(cid:21)(cid:27) x + v2(cid:107)k2 y v⊥v(cid:107) ∆3 eff 2(cid:101)µ3 (29) , Re Π(2D) Ren (k, ω) = (cid:20)(cid:101)µ2 − ∆2 2(cid:101)µω2 + eff gd 2π 1 6∆eff × 4 which includes the positive parameter (cid:101)µ − ∆eff v⊥v(cid:107) Λ2 = . (30) This parameter should also be omitted for the same reasons as in the case of Λ1 in Eq. (24). The second term in the square brackets is smaller than the first one eff) and, therefore, can also by a factor of ω2/((cid:101)µ2 − ∆2 be omitted. Thus, we obtain Re Π(2D) Ren (k, ω) = gd 4π v2⊥k2 x + v2(cid:107)k2 y v⊥v(cid:107) (cid:101)µ2 − ∆2 (cid:101)µω2 eff . (31) The imaginary part of Π(2D) Ren (k, ω) vanishes within the range(cid:113) (cid:113) v2⊥k2 x + v2(cid:107)k2 y < ω < 4∆2 eff + v2⊥k2 x + v2(cid:107)k2 y. (32) 3.3. Dispersion Law for Plasmons In the random phase approximation, the dispersion law for plasmons is determined by the equation 1 − V (k)Π(k, ω) = 0. (33) When the Fermi level falls within the minigap, Eq. (26) for the polarization operator Π(k, ω) and Eq. (18) for the Coulomb interaction should be substituted into Eq. (33). When the Fermi level falls within the miniband, Eq. (31) for the polarization operator Π(k, ω) and Eq. (18) with θ = kxd for the Coulomb interaction should be substituted into Eq. (33). In the former case, we obtain ω(1D) pl (ky, θ) = vFky In the latter case, we have ω(2D) pl (k) = v2⊥k2 x + v2(cid:107)k2 y (cid:113) (cid:114) gpF π(cid:101)µ V (ky, θ). (cid:115) (cid:101)µ2 − ∆2 v⊥v(cid:107)(cid:101)µ V (k). (35) gd 4π (34) eff In the case of closely spaced strips of gapless graphe- ne, expression (34) at the boundary of the plasmon band gives the square-root plasmon dispersion law char- acteristic of two-dimensional systems: ω(1D) pl (ky) = vF ky. (36) (cid:115) 2πn2D(cid:101)e2 (cid:101)µ At low ky values, we retain only the second term in Eq. (20) for the Coulomb interaction. However, it follows from Eq. (34) in this case that the plasmon dispersion law remains acoustic for nearly the whole plasmon band (almost for all θ values), ω(1D) pl (ky, θ) = vFky where f (θ) = ln − C − ψ d πdI (cid:18) θ 2π (cid:115) 2g(cid:101)e2pF π(cid:101)µ f (θ), (cid:19) + tan π 2 + 1 2 (37) θ 2 (38) 5 according to Eq. (19) for the Coulomb interaction. In the case of the linear dependence of the chemical potential on the Fermi momentum, Eq. (36) gives the well-known result for the plasmon dispersion law in gapless graphene [10] (cid:114) g 2 (cid:101)µ(cid:101)e2ky. ωpl(ky) = (39) Here, the plasmon propagates along the y axis. The dispersion law for the two-dimensional plasmon in gapless graphene can also be obtained from Eq. (35) in the isotropic case, where v⊥ = v(cid:107) = vF and (cid:101)µ2 − ∆2 we should take into account the relation F. Here, in the quasi-two-dimensional case, ef f = v2 Fp2 p2 F = 4π g n2D. (40) Formulas (34) and (35) give the well-known expres- sions for the case of nonrelativistic charge carriers [11]. For example, at large distances between the strips of gapless graphene (dII (cid:29) dI), the system behaves as a set of strips. The Coulomb interaction between the charge carriers in one of such strips is given by the first term on the right-hand side of Eq. (18). In the nonrelativistic limit, when vFpF (cid:28) ∆eff and (cid:101)µ ≈ ∆eff, Eq. (34) yields (cid:115) ω(1D) pl (ky) = ky 2g(cid:101)e2pF πm∗ ln 4 kydI . (41) In the nonrelativistic limit for the case of isotropy with respect to velocities, formula (35) gives where Ωp = ω(2D) pl (k) = Ωp (cid:18) 2π(cid:101)e2n2D dm∗ (42) . (43) 3.4. Band Character of Plasmon Excitations Owing to the periodicity of the array under study, not only the spectrum of single-particle excitations but also the plasmon excitation spectrum is separated into minibands. In the momentum space, the boundaries of plasmon bands coincide with the boundaries of the corresponding minibands for the charge carriers. This is a consequence of the Bragg condition: 2kgj = g2 j , j = ±1, ±2, . . ., where gj = (2πj/d, 0) is the recipro- cal lattice vector related to the potential of the super- lattice. Thus, we find kxj = πj/d. A discontinuity appears at the boundaries of plas- mon bands. Similar to [3], we can find boundary values for the plasmon frequencies. The values of plasmon frequencies in the center of the plasmon band ωpl(0) in higher minibands coincide with the minimum energy values for charge carriers in these minibands. Let us estimate these values. Finding an approximate solution of dispersion relation (3) with respect to energy at the point kx = ky = 0, we obtain (cid:112)kd, (cid:19)1/2 the following estimate for the energy of charge carriers in the nth miniband (n = 0, 1, 2, . . .): In the quasi-one-dimensional case, the absorption intensity is Ee, h n = vF dI + πn 1 − vF dI∆0 + V0 ∆0 , (44) Q(1D) (cid:39) σ(1D) 0 E 2 0 (ω2 − ω2 ω2ν2 0)2 + ω2ν2 , (48) (cid:20) ±(cid:16) π 2 (cid:17)(cid:18) (cid:19) (cid:21) 0 where σ(1D) is the value of conductivity (46) in the zero-frequency limit, ω0 = ω(1D) (k0) is the plasmon frequency corresponding to the wave vector k0 = (0, k0), and E0 is the electric field amplitude. pl In the quasi-two-dimensional case, the anisotropy of the conductivity leads to the dependence of absorption on the orientation of the polarization plane in the inci- dent electromagnetic wave. Let its polarization plane be rotated by the angle ϕ with respect to the x axis. Then, the absorption intensity is equal to Q(2D) = Q(2D)⊥ cos2 ϕ + Q(2D) (cid:107) sin2 ϕ. (49) Here, Q(2D)⊥ (cid:39) 1 2 (cid:39) 1 2 Q(2D) (cid:107) 0 xx0 E 2 σ(2D) yy0 E 2 σ(2D) 0 (ω2 − ω2 (ω2 − ω2 ω2ν2 0)2 + ω2ν2 , ω2ν2 0)2 + ω2ν2 , (50) xx0 and σ(2D) where σ(2D) yy0 are the expressions for the diag- onal elements of conductivity tensor (47) in the zero- frequency limit and ω0 = ω(2D) (k0) is the plasmon fre- quency corresponding to wave vector k0 = (k0 cos ϕ, k0 sin ϕ). pl It follows from Eqs. (50) that the spectrum of ab- sorption of electromagnetic waves by plasmons should be strongly anisotropic: the corresponding ratio of in- tensities is Q(2D)⊥ /Q(2D) = (v⊥/v(cid:107))2. Such strong aniso- tropy could be easily revealed in an experiment. (cid:107) 4. CONCLUSIONS In this work, we have explicitly derived the plasmon dispersion law for the planar graphene superlattice. It has been shown that the absorption spectrum for lin- early polarized electromagnetic waves modulated with the period equal to the plasmon wavelength should ex- hibit a pronounced anisotropy. When the Fermi level falls within the minigap, this anisotropy is due to the exclusion of oscillations of the charge carrier density across the superlattice potential. When the Fermi level falls into the miniband, this anisotropy is attributed to an appreciable suppression (∼ (v⊥/v(cid:107))2) of such oscil- lations because of a significant difference between the transverse and longitudinal velocity components. We are grateful to D. N. Sob'yanin for helpful dis- cussions and valuable comments. where the upper and lower signs correspond to elec- trons and holes, respectively. In particular, estimate (44) for n = 0 (the lowest electron or highest hole 0 = ±∆eff + Veff, where ∆eff and Veff bands) gives Ee, h are specified by Eqs. (7). Let us take the characteristic values dI (cid:39) 10 nm, ∆0 (cid:39) 1 eV, and V0 = 0 for sim- 0 (cid:39) 80 meV and plicity of estimates. Then, we have Ee 1 (cid:39) 240 meV. We can see that the energy difference Ee between the neighboring minibands far exceeds room temperature. Hence, we can neglect the thermally ac- tivated filling of higher minibands. Let us now estimate how many additional electrons are needed to completely fill the lowest electron mini- band and to start the filling of the next electron mini- band. The Fermi momentum pF should be such that 1. Us- (owing to the condition V0 = 0, coincides with ∆eff). Then, (13), we can relate the found pF according to Eq. value to the two-dimensional electron density, n2D = 0 /(πvFd) (cid:39) 5 × 1012 cm−2. This value is fairly large as compared to the experimental data for gapless graphene [10]. the chemical potential (cid:101)µ becomes equal to Ee ing Eq. (12), we obtain vFpF = (cid:112)Ee2 1 − Ee2 g(cid:112)Ee2 1 − Ee2 0 3.5. Absorption Intensity for the Modulated Electromagnetic Radiation The intensity of absorption for electromagnetic wa- ves modulated with the period equal to the plasmon wavelength is given by the well-known formula Q = Re where σ is the conductivity of the system, (cid:101)E is the electric field of the plasmon wave, and E is the electric field of the electromagnetic wave varying with the fre- quency ω. (45) The conductivity of the system is easily found from In the the kinetic equation in the τ approximation. quasi-one-dimensional case, it reads σ(1D) = ige2vFpF π(cid:101)µ(ω + iν) . (46) Here and further on, ν = 1/τ . In the quasi-two-dimensional case, the conductivity is a tensor with diagonal elements (cid:16) σ(cid:101)EE∗(cid:17) , 1 2 σ(2D) xx = σ(2D) yy = ige2 4π(ω + iν) ige2 4π(ω + iν) (cid:101)µ2 − ∆2 (cid:101)µ (cid:101)µ2 − ∆2 (cid:101)µ eff eff v⊥ v(cid:107) v(cid:107) v⊥ , . (47) 6 References [1] M. Freitag, T. Low, W. Zhu, H. Yan, F. Xia, and P. Avouris, Nature Commun. 4, 1951 (2013). [8] P. V. Ratnikov, JETP Lett. 90, 469 (2009). [9] P. V. Ratnikov and A. P. Silin, JETP Lett. 100, 311 [2] S. Das Sarma and E. H. Hwang, Phys. Rev. Lett. (2014). 102, 206412 (2009). [3] A. V. Chaplik, JETP Lett. 100, 262 (2014). [4] D. C. Elias, R. R. Nair, T. M. G. Mohiuddin, S. V. Morozov, P. Blake, M. P. Halsall, A. C. Ferrari, D. W. Boukhvalov, M. I. Katsnelson, A. K. Geim, and K. S. Novoselov, Science 323, 610 (2009). [5] I. Zanella, S. Guerini, S. B. Fagan, J. M. Filho, and G. S. Filho, Phys. Rev. B 77, 073404 (2008). [6] S. Yu. Glazov, A. A. Kovalev, and N. E. Meshch- eryakova, Semiconductors 49, 504 (2015). [7] S. Yu. Glazov, A. A. Kovalev, and N. E. Meshch- eryakova, Bull. Russ. Acad. Sci.: Phys. 76, 1323 (2012). [10] V. N. Kotov, B. Uchoa, and V. M. Pereira, F. Guinea, and A. H. Castro Neto, Rev. Mod. Phys. 84, 1067 (2012). [11] E. A. Andryushin and A. P. Silin, Phys. Solid State 35, 164 (1993). [12] W. Zawadzki, Adv. Phys. 23, 435 (1974). [13] B. L. Gel'mont and M. V. Kisin, Sov. Phys. Semi- cond. 17, 791 (1983), Sov. Phys. Semicond. 17, 947 (1983), Sov. Phys. Semicond. 18, 506 (1984). [14] N. V. Markova, A. P. Silin, J. Moscow Phys. Soc. 4, 311 (1994). [15] V. N. Tsytovich, Sov. Phys. JETP 13, 1249 (1961). Translated by K. Kugel 7 

1211.5155

1

1211

Direct Probing of 1/f Noise Origin with Graphene Multilayers: Surface vs. Volume

Low-frequency noise with the spectral density S(f)~1/f^g (f is the frequency and g~1) is a ubiquitous phenomenon, which hampers operation of many devices and circuits. A long-standing question of particular importance for electronics is whether 1/f noise is generated on the surface of electrical conductors or inside their volumes. Using high-quality graphene multilayers we were able to directly address this fundamental problem of the noise origin. Unlike the thickness of metal or semiconductor films, the thickness of graphene multilayers can be continuously and uniformly varied all the way down to a single atomic layer of graphene - the actual surface. We found that 1/f noise becomes dominated by the volume noise when the thickness exceeds ~7 atomic layers (~2.5 nm). The 1/f noise is the surface phenomenon below this thickness. The obtained results are important for continuous downscaling of conventional electronics and for the proposed graphene applications in sensors and communications.

cond-mat.mes-hall

cond-mat

Direct Probing of 1/f Noise Origin with Graphene Multilayers: Surface vs. Volume Guanxiong Liu1, Sergey Rumyantsev2,3, Michael S. Shur2 and Alexander A. Balandin1,4,* 1Nano-Device Laboratory, Department of Electrical Engineering, Bourns College of Engineering, University of California – Riverside, Riverside, California 92521 USA 2Center for Integrated Electronics and Department of Electrical, Computer and Systems Engineering, Rensselaer Polytechnic Institute, Troy, New York 12180 USA 3Ioffe Physical-Technical Institute, The Russian Academy of Sciences, St. Petersburg, 194021 Russia 4Materials Science and Engineering Program, University of California – Riverside, Riverside, California 92521 USA *Correspondence to: balandin@ee.ucr.edu 1 Abstract: Low-frequency noise with the spectral density S(f)~1/f (f is the frequency and ≈1) is a ubiquitous phenomenon, which hampers operation of many devices and circuits. A long- standing question of particular importance for electronics is whether 1/f noise is generated on the surface of electrical conductors or inside their volumes. Using high-quality graphene multilayers we were able to directly address this fundamental problem of the noise origin. Unlike the thickness of metal or semiconductor films, the thickness of graphene multilayers can be continuously and uniformly varied all the way down to a single atomic layer of graphene – the actual surface. We found that 1/f noise becomes dominated by the volume noise when the thickness exceeds ~7 atomic layers (~2.5 nm). The 1/f noise is the surface phenomenon below this thickness. The obtained results are important for continuous downscaling of conventional electronics and for the proposed graphene applications in sensors and communications. Article Text: Low-frequency noise with the spectral density S(f)~1/f (f is the frequency and ≈1) is a ubiquitous phenomenon, first discovered in vacuum tubes (1) and later observed in a wide variety of electronic materials (2-5). The importance of this noise for electronic and communication devices motivated numerous studies of its physical mechanisms and methods for its control (6). However, after almost a century of investigations, the origin of 1/f noise in most of material systems still remains a mystery. A question of particular importance for electronics is whether 1/f noise is generated on the surface of electrical conductors or inside their volumes. Here we show that by using mechanically exfoliated high-quality graphene multilayers one can address directly the fundamental problem of the surface vs. volume origin of 1/f noise. Unlike the thickness of metal or semiconductor films, the thickness of graphene multilayers can be continuously and uniformly varied all the way down to a single atomic layer of graphene – the 2 actual surface. Using a set of samples with the number of atomic planes, n, varying from n=1 to n=15, we found that in moderately-doped samples 1/f noise becomes dominated by the volume noise when the thickness exceeds n≈7. The latter is an unexpected discovery considering that seven atomic layers constitute a film of only ~2.5-nm thickness. Our results reveal a scaling law for the 1/f noise, which is important for nanoscale devices and for proposed graphene applications in sensors, analog circuits, and communications. The problem of 1/f noise origin: It is hard to find another scientifically and practically important problem that has ignited so many debates but remained unsolved for almost a century as the problem of volume vs. surface origin and mechanism of 1/f noise in electrical conductors (7-16). The intensity of discussions can be inferred even from the titles of seminal publications on the subject: “1/f noise is no surface effect” (1969) (9) followed by “1/f noise: still a surface effect” (1972) (10). The direct test of whether measured 1/f noise is dominated by contributions coming from the sample surface or its volume has not been possible because of inability to fabricate continuous metal or semiconductor films with the uniform thickness below ~8 nm (14, 16). The state-of-the-art in noise field is characterized by existence of a large number of ad hoc models each tailored to a specific material system or a device. For example, one of few conventionally accepted theories – McWhorter model (7) – deals with 1/f noise in field-effect transistors. The fundamental understanding of the origin of 1/f noise in homogeneous electrical conductors is still lacking. The recent advent of the mechanically exfoliated graphene, which allowed for investigation of its electronic (17-18) and thermal (19) properties, opens up a possibility of revisiting the old 3 problem of the 1/f noise origin. Owing to their layered crystal structure and presence of the van- der-Waals gaps, the multilayer graphene films can be made nearly defect-free and continuous down to the thickness of a single atomic layer – an ultimate surface. The noise response in graphene multilayers is not affected as much by the grains and roughness like in metals. The thickness of the graphene multilayers can be made uniform along their entire area and determined with better accuracy than that of the metal films or conducting channels in the semiconductor transistors used previously for the noise studies. Observing the evolution of the noise-power spectral density SI(f) as the number of atomic planes gradually decreases from a relatively thick graphite film to a single layer graphene (SLG) one can answer the fundamental question of whether the noise is coming from the volume or surface or both and, if 1/f noise is coming from both the surface and volume – at what thickness of the conductor the volume contribution becomes dominant. The knowledge of the thickness at which the 1/f noise is the surface phenomenon has important implications for downscaling electronics beyond a few nanometers. Sample Preparation and Measurements: The graphene multilayers under study were produced by the mechanical exfoliation on Si/SiO2 substrate (17). The back-gated devices were fabricated by the electron-beam lithography with Ti/Au (6-nm/60-nm) electrodes (20). Micro-Raman spectroscopy is an excellent tool for determining the number of atomic plans for n<5 (21-22). For larger n – essential for this study – the method becomes ambiguous. For this reason, we combined Raman spectroscopy with atomic force microscopy (AFM). The details of n value extraction are given in Supplementary Information. Figure 1a-d shows scanning electron microscopy (SEM) and AFM images of a typical graphene multilayer device under test, 4 evolution of the Raman spectra as n changes from 1to 15 and the sheet resistance, RST, vs. back- gate bias for different multilayers. The absence of the disorder D peak in Raman spectra confirms a high quality of our samples (22). Since the gating is needed for determining the charge neutrality point (CNP) in different samples, we limited the thickness of the films to n=15 where the gating becomes weak. [Figure 1] The noise measurements were performed using conventional instrumentation (Supplementary Information). Figure 2a shows typical noise spectra for graphene multilayers with different n. In all cases, the noise spectral density was close to ~1/f with ≈1. The absence of Lorentzian bulges indicates that no trap with a specific time constant dominated the spectra (6, 20). The SI proportionality to I2 (Figure 2b) implies that the current does not drive the fluctuations but merely makes them visible as in other homogeneous conductors (16). It is informative to correlate the normalized noise spectral density SI(f)/I2 with the sample resistance. For proper comparison, SI(f)/I2 should also be normalized to the graphene device area A (23-24). In our samples, A varied from 1.5 to 70 μm2. Figure 2c shows SI(f)/I2×A vs. RST in graphene multilayers near CNP at f=10 Hz. The experimental data can be fitted with two segments of a different slope. At the low resistance range – corresponding to thicker multilayers – the 1/f noise is proportional to RST while at the higher resistance range – thinner multilayers – the dependence is close to RST 2. For homogeneous metals and semiconductors, SI(f) proportionality to RST or RST 2 was interpreted as indication of the volume or surface noise origin, respectively (16). 5 It was previously established that 1/f noise in the single layer graphene (SLG) devices has larger contribution from the graphene-metal contacts that graphene multilayer devices (25). The deposition of metal contacts to graphene results in graphene doping via the charge transfer to reach the equilibrium conditions, and, correspondingly, leads to the local shift of the Fermi level in graphene (26). The electron density of states in graphene near CNP is low owing to the Dirac- cone linear dispersion. Thus, even a small amount of the charge transfer from or to the metal affects the local Fermi energy of graphene stronger than in graphene multilayers. In Figure 2c we indicated a data point for a sample which had a single atomic plane thickness along the conducting channel region but gradually increased thickness (to n=2 or 3) under the metal contacts. This sample can be viewed as SLG with reduced graphene-metal contact contribution 2 dependence to the noise. One can see that the data point corresponding to this device fits the RST better. Figure 2d presents the dependence of SI(f)/I2×A on the gate bias, VG, for multilayers with different n. As expected the noise amplitude is the highest in SLG (27-28). The noise decrease with n is consistent with the sheet resistance change presented in Figure 2c. The wide range of the back-gate bias allowed us to probe the noise near CNP and in the high-bias regime (VG>40 V) characterized by the large carrier densities. [Figure 2] Noise scaling and its origin: The electrical conductance in uniform metal and semiconductor channels follows the Ohm’s law: R=ρ×(L/W×H)=RST×(L/W), where ρ is the resistivity, L is the length, W is the width and H is thickness of the conductor. However, RST~1/H scaling is not necessarily valid for graphene multilayers as H approaches a single atomic plane – surface. 6 Figure 3a shows measured RST as a function of H. Our data agree well with independent studies of RST(H) (29). The linear fit in the logarithmic plot works well for n decreasing from 15 to 2. The RST value for SLG falls off this dependence suggesting that the resistance of the first atomic plane in contact with the substrate – surface – has to be distinguished from multilayers on top of it. Considering that the screening length inside graphite in c-direction is 0.38-0.5 nm (30) the data in Figure 3a indicating a deviation from RST~1/H scaling for H below n=2 is reasonable. We now turn to the analysis of the noise data for graphene multilayers as H changes from that of a graphite film to an ultimate surface. Our sample can be viewed as a parallel resistor network consisting of the surface – SLG in contact with the substrate – and graphene multilayer above it (see insets to Figure 3). The first atomic plane on the substrate with the sheet resistance RS is connected in parallel with the other n-1 layers on top of it. The (n-1) layers have the sheet resistance of RB. The RB value can be extracted from the measured total resistance RT as RB=(RST×RS)/(RST-RS). Figure 3b shows RB as a function of (n-1) fitted with the power law RB=ρB/(n-1)β, where ρB is 18.2 kΩ-per atomic plane and β=1.7. The area-normalized surface 2 is generated in the first atomic plane – surface – while the area-normalized volume noise SRS/RS noise originates in the other (n-1) layers. The volume noise scales with the thickness as 1/(n-1) (i.e. 1/H). Both the surface noise from the first graphene atomic plane and the volume noise from multilayers above contribute to the noise measured from the total sheet resistance of RST=RB×RS/(RB+RS). [Figure 3] 7 The 1/f noise spectral density includes components from the surface and from the volume. The volume noise scales with (n-1) and can be written as , where the noise amplitude of each layer, , is the fitting parameter to the experimental data (the derivation details are in the Supplementary Information). The value of the surface resistance RS and are the average values for SLG directly measured as 4.43 kΩ and 2.3×10-8 Hz-1×μm2, respectively. The bulk resistance is described as RB=ρB/(n-1)β with the experimentally determined parameters given above. When the thickness reduces down to one atomic layer, the total noise consists of the surface noise only. Figure 4 shows the experimental data for SI(f)/I2×A vs. n together with the functional dependence predicted by our model. The best fit is obtained for =(1/6)× =(1/6)×(2.3×10-8) Hz- 1×μm2, which means the surface noise from the first graphene layer – the surface – is six times larger than the noise from each of the other graphene layers, which constitute the volume, i.e. “bulk”, of our sample. When plotted separately, the two noise contributions – originating on the surface and in the volume – intersect at n~7. This indicates that the surface noise is dominant in graphene multilayers for n≤7. In thicker samples, the 1/f noise is essentially the volume phenomenon. The situation is different in the high-bias regime shown in the inset. The 1/n2 scaling for graphene multilayers (n>2) indicates that the noise is dominated by the surface contributions, which is consistent with the data in the inset to Figure 2c. There is no transition to 1/n scaling at n~7. In multilayer graphene samples the resistance of the surface layer decreases with increasing VG as more charge carriers are induced electrostatically. This leads to the increased contribution of the bottom surface to the overall current conduction and, therefore, to 8 22//STRIRSISARSRRRSRBSBS222/)/(ARSRRRBRBSSB222/)/()1/(/2nRSBRBARSSRS2/ARSSRS2/1/f noise. As a result, the noise in the multilayers increases with increasing VG (Figure 2d) revealing a transition from the bulk to surface noise. [Figure 4] Prospective: Apart from the fundamental science importance of testing directly the origin of 1/f- noise, the obtained results are very important for electronics applications. The progress in information technologies and communications crucially depends on the continuing downscaling of the conventional devices, such as main stream silicon complementary metal-oxide- semiconductor (CMOS) technology. As the feature size of CMOS devices is scaled down to achieve higher speed and packing density, the 1/f noise level strongly increases (31-34) and becomes the crucial factor limiting the ultimate device performance and downsizing (6, 33, 34). The increase in 1/f noise is detrimental not only for transistors and interconnects in digital circuits but also for the high-frequency circuits such as mixers and oscillators, where 1/f noise upconverts into the phase noise (6) and deteriorates the signal-to-noise ratio in the operational amplifiers and analogue-digital – digital-analogue converters. With typical dielectric thicknesses in modern transistors of the order of a nanometer, it is important to understand when 1/f noise is becoming a pure surface noise. Although the exact thickness may vary from material to material, the length scale at which the transistor channel or barrier layer becomes the surface from the noise prospective is valuable for design of the next generation electronics. The obtained results are also crucial for the graphene applications. Graphene does not have a band-gap, which seriously impedes its prospects for digital electronics. However, graphene revealed potential for the high-frequency analog communications, interconnects (35) and sensors (36). These 9 applications require low level of 1/f noise, which is up-converted via unavoidable device and circuit non-linearity. Therefore, the knowledge of the scaling law for the 1/f noise in graphene multilayers is very important for most of realistic graphene applications in electronics. References: [1] J. B. Johnson, Phys. Rev. 26, 71 (1925). [2] I. Flinn, Nature 219, 1356 (1968). [3] R. F. Voss, J. Clarke, Nature 258, 317 (1975). [4] D. L. Gilden, T. Thornton, M. W. Mallon, Science 267, 1837 (1995). [5] R. J. Schoelkopf, P. Wahlgren, A. A. Kozhevnikov, P. Delsing, D. E. Prober, Science 280, 1238 (1998). [6] A. A. Balandin, Noise and Fluctuations Control in Electronic Devices (American Scientific Publishers, Los Angeles, 2002). [7] A. L. McWhorter, in Semiconductor Surface Physics, R. H. Kingston, Ed. (University of Pennsylvania Press, Philadelphia, 1957), pp. 207-228. [8] C. T. Sah, F. H. Hielscher, Phys. Rev. Lett. 17, 956 (1966). [9] F. N. Hooge, Phys. Lett. A 29, 139 (1969). [10] A. Mircea, A. Roussel, A. Mitonneau, Phys. Lett. A 41, 345 (1972). [11] J. W. Eberhard, P. M. Horn, Phys. Rev. B 18, 6681 (1978). [12] M. Celasco, F. Fiorillo, A. Masoero, Phys. Rev. B 19, 1304 (1979). [13] D. M. Fleetwood, J. T. Masden, N. Giordano, Phys. Rev. Lett. 50, 450 (1983). [14] N. M. Zimmerman, J. H. Scofield, J. V. Mantese, W. W. Webb, Phys. Rev. B 34, 773 (1986). 10 [15] N. V. Dyakonova, M. E. Levinshtein, Sov. Phys. Semicond. 23, 175 (1989). [16] P. Dutta, P. M. Horn, Rev. Mod. Phys. 53, 497 (1981). [17] K. S. Novoselov, A. K. Geim, S. V. Morozov, D. Jiang, M. I. Katsnelson, I. V. Grigorieva, S. V. Dubonos, A. A. Firsov, Nature 438, 197 (2005). [18] Y. Zhang, Y.-W. Tan, H. L. Stormer, P. Kim, Nature 438, 201 (2005). [19] A. A. Balandin, Nat. Mat. 10, 569 (2011). [20] Q. Shao, G. Liu, D. Teweldebrhan, A. A. Balandin, S. Rumyantesv, M. Shur, D. Yan, IEEE Electron Device Lett. 30, 288 (2009). [21] A. C. Ferrari, J. C. Meyer, V. Scardaci, C. Casiraghi, M. Lazzeri, F. Mauri, S. Piscanec, D. Jiang, K. S. Novoselov, S. Roth, A. K. Geim, Phys. Rev. Lett. 97, 187401 (2006). [22] I. Calizo, I. Bejenari, M. Rahman, G. Liu, A. A. Balandin, J. Appl. Phys. 106, 043509 (2009). [23] S. Rumyantsev, G. Liu, M. Shur, A. A. Balandin, J. Phys.: Condens. Matter 22, 395302 (2010). [24] Y. Zhang, E. E. Mendez, X. Du, ACS Nano. 5, 8124 (2011). [25] G. Liu, S. Rumyantsev, M. Shur, A. A. Balandin, Appl. Phys. Lett. 100, 033103 (2012). [26] E. J. H. Lee, K. Balasubramanian, R. T. Weitz, M. Burghard, K. Kern, Nature Nanotech. 3, 486 (2008). [27] Y.-M. Lin, P. Avouris, Nano Lett. 8, 2119 (2008). [28] G. Liu, W. Stillman, S. Rumyantsev, Q. Shao, M. Shur, A. A. Balandin, Appl. Phys. Lett. 95, 033103 (2009). [29] K. Nagashio, T. Nishimura, K. Kita, A. Toriumi, Appl. Phys. Exp. 2, 025003 (2009). [30] L. Pietronero, S. Strassler, H. R. Zeller, M. J. Rice, Phys. Rev. Lett. 41, 763 (1978). 11 [31] M.H. Tsai and T.P. Ma, IEEE Trans. Electron Dev., 41, 2061 (1994). [32] E. Simoen and C. Claeys, Solid-State Electron., 43, 865 (1999). [33] K.W. Chew, K.S. Yeo and S.-F. Chu, IEE Proc.-Circuits Devices Syst., 151, 415 (2004). [34] L. B. Kish, D. Abbott, B.R. Davis, J. Deen, A Practical Introduction to Noise (Cambridge University Press, 2004). [35] J. Yu, G. Liu, A.V. Sumant, V. Goyal and A.A. Balandin, Nano Lett., 12, 1603 (2012). [36] S. Rumyantsev, G. Liu, M. Shur, R.A. Potyrailo and A.A. Balandin, Nano Lett., 12, 2294 (2012). Acknowledgements: The work at UCR was supported, in part, by the Semiconductor Research Corporation (SRC) and Defense Advanced Research Project Agency (DARPA) through FCRP Center on Functional Engineered Nano Architectonics (FENA) and by the National Science Foundation (NSF) projects US EECS-1128304, EECS-1124733 and EECS-1102074. The work at RPI was supported by the US NSF under the auspices of I/UCRC “CONNECTION ONE” at RPI and by the NSF EAGER program. SLR acknowledges partial support from the Russian Fund for Basic Research (RFBR) grant 11-02-00013. Author Contributions: A.A.B. coordinated the project, supervised graphene device fabrication, contributed to data analysis and wrote the manuscript; G.L. fabricated graphene devices, carried out Raman, AFM and electrical measurements, and performed data analysis; S.R. performed noise measurements and data analysis; M.S.S. performed data analysis. 12  Figure 1: (a) Pseudo - color SEM image of a device used for noise measurements showing graphene multilayer (blue) and metal electrodes (yellow). The scale bar is 3 μm. (b) AFM image of a graphene multilayer with the scan direction marked by the dash line. The scanning profile indicates the apparent AFM thickness of 3.4 nm, which corresponds to n≈7 layers with the carbon-bond thickness h=0.35 nm. (c) Raman spectra of the graphene multilayers with different number of atomic planes n. The 2D band undergoes noticeable evolution up to n=7-9 layers while for the thicker samples the 2D band becomes indistinguishable from that of bulk graphite. None of the samples reveal D peak around 1350 cm-1 attesting the sample quality and absence of 13 the fabrication induced defects. (d) Sheet resistance, RST, of graphene and graphene multilayers as a function of the back-gate bias, VG-VCNP, referenced to the charge neutrality point. The data is shown for the continuously changing thickness, H=h×n, from n=1 to n=15. The gating becomes weaker as n increases. Figure 2: (a) Noise spectral density, SI/I2, of graphene multilayers as a function of frequency shown for three devices with distinctively different thickness: n=1, n≈7 and n≈12. In all cases, the noise spectral density is SI~1/f with ≈1. (b) Noise density as a function of the drain-source current, I, indicating perfect scaling with I2 expected for conventional 1/f noise. (c) Noise spectral density normalized by the channel area, SI/I2×A, as a function of the sheet resistance of graphene and graphene multilayers. RST values were taken near the charge neutrality point. Note 2 in the high-resistance the noise density scaling with RST in the low-resistance limit and with RST limit. Two data points indicated for SLG corresponds to the sample (A) with the thickness n=1 over the entire channel and under the metal contact and to the sample (B) with the thickness n=1 over the entire channel but increasing to n=2-3 under the metal contacts to minimize the contact 14 noise contribution. Inset shows the noise spectral density as a function of RST in the high-bias regime (VBG-VCNP=-40 V) with the same units. (d) Noise spectral density SI/I2×A as a function of the gate bias for several multilayers. Figure 3: (a) The measured sheet resistance of graphene and graphene multilayers as a function of the number of atomic planes n near CNP. (b) The extracted sheet resistance, RB, of graphene multilayers on top of the first atomic plane – surface – as the function of the number of atomic planes n-1. The insets show the parallel resistor network used to model the resistance of graphene multilayers. RS is the sheet resistance of the first graphene layer in direct contact with the substrate, which represents the resistance of the actual surface. RB is the sheet resistance of graphene multilayers on top of the first atomic plane. The power-law fitting of RB is used for the noise data analysis. 15  Figure 4: The experimental data for the normalized noise spectral density SI(f)/I2×A near CNP is fitted with the model, which explicitly takes into account the surface noise – originating in the graphene layer in direct contact with the substrate – and the volume (i.e. bulk) noise. For clarity, the surface and volume noise components are also plotted separately. The crossover point at n≈7 indicates the thickness (H≈7×0.35 nm≈2.45 nm) at which the 1/f noise becomes essentially the volume phenomenon. In the conducting channels with thickness below this value, the 1/f noise is dominated by the surface. The inset shows the noise spectral density in the high-bias regime where surface contributions are more persistent as revealed by 1/n2 scaling beyond n=7 thickness. The units in the inset are the same as in the main plot. 16 

1901.10488

1

1901

Initialisation of single spin dressed states using shortcuts to adiabaticity

We demonstrate the use of shortcuts to adiabaticity protocols for initialisation, readout, and coherent control of dressed states generated by closed-contour, coherent driving of a single spin. Such dressed states have recently been shown to exhibit efficient coherence protection, beyond what their two-level counterparts can offer. Our state transfer protocols yield a transfer fidelity of ~ 99.4(2) % while accelerating the transfer speed by a factor of 2.6 compared to the adiabatic approach. We show bi-directionality of the accelerated state transfer, which we employ for direct dressed state population readout after coherent manipulation in the dressed state manifold. Our results enable direct and efficient access to coherence-protected dressed states of individual spins and thereby offer attractive avenues for applications in quantum information processing or quantum sensing.

cond-mat.mes-hall

cond-mat

Initialisation of single spin dressed states using shortcuts to adiabaticity J. Kolbl,1 A. Barfuss,1 M. S. Kasperczyk,1 L. Thiel,1 A. A. Clerk,2 H. Ribeiro,3 and P. Maletinsky1, ∗ 1Department of Physics, University of Basel, Basel 4056, Switzerland 2Institute for Molecular Engineering, University of Chicago, Chicago, Illinois 60637, USA 3Max Planck Institute for the Science of Light, Erlangen 91058, Germany (Dated: January 31, 2019) We demonstrate the use of shortcuts to adiabaticity protocols for initialisation, readout, and coherent control of dressed states generated by closed-contour, coherent driving of a single spin. Such dressed states have recently been shown to exhibit efficient coherence protection, beyond what their two-level counterparts can offer. Our state transfer protocols yield a transfer fidelity of ∼ 99.4(2) % while accelerating the transfer speed by a factor of 2.6 compared to the adiabatic approach. We show bi-directionality of the accelerated state transfer, which we employ for direct dressed state population readout after coherent manipulation in the dressed state manifold. Our results enable direct and efficient access to coherence-protected dressed states of individual spins and thereby offer attractive avenues for applications in quantum information processing or quantum sensing. The pursuit of protocols for quantum sensing [1, 2] and quantum information processing [3, 4] builds on es- tablished techniques for initialising, coherently manipu- lating, and reading out quantum states, as extensively demonstrated in, e.g. trapped ions [5, 6], solid state qubits [7] and color centre spins [8]. Importantly, the in- volved quantum states need to be protected from deco- herence [9], which is primarily achieved by pulsed dynam- ical decoupling [10 -- 12], a technique which suffers from drawbacks including experimental complexity and vul- nerability to pulse errors. In contrast, 'dressed states' generated by continuous driving of a quantum system yield efficient coherence protection [13 -- 16], even for com- paratively weak driving fields [17], in a robust, exper- imentally accessible way that is readily combined with quantum gates [18 -- 20]. A major bottleneck for further applications of such is the difficulty in perform- dressed states, however, ing fast, high-fidelity initialisation into individual, well- defined dressed states. Up to now, such initialisation has focused on two-level systems and has mainly used adiabatic state transfer [18, 19], without detailed char- acterisation of the resulting fidelities. Adiabatic state transfer, however, suffers from a tradeoff between speed and fidelity: The initialisation must be slow to main- tain fidelity, but fast enough to avoid decoherence dur- ing state transfer. For experimentally achievable driving field strengths, this tradeoff and the remaining sources of decoherence form a key limitation to further advances in the use of dressed states in quantum information pro- cessing and sensing. Here, we overcome these limitations with a twofold approach, where we employ recently developed proto- cols for 'shortcuts to adiabaticity' (STA) [21 -- 25] and apply them to the initialisation of three-level dressed states that exhibit efficient coherence protection, be- yond what is offered by driven two-level systems [17]. Specifically, we focus on dressed states emerging from 'closed-contour driving' (CCD) of a quantum three-level system [17] [Fig. 1(a)]. These dressed states stand out due to remarkable coherence properties and tunability through the phase of the involved driving fields [17]. While dynamical decoupling by continuous driving has previously been demonstrated for electronic spins in di- amond [14, 17, 19, 26, 27], STA have never been ex- plored on such solid state spins in their ground state [25], nor on the promising three-level dressed states we study here. By combining STA and CCD, we establish an attractive, room-temperature platform for applications, e.g. in quantum sensing of high-frequency magnetic fields [28, 29] on the nanoscale. We implement these concepts on individual Nitrogen- Vacancy (NV) electronic spins, which, due to their room- temperature operation and well-established methods for optical spin initialisation and readout [30], provide an attractive, solid-state platform for quantum technolo- gies. The dressed states we study emerge from the S = 1 electronic spin ground state of the negatively charged NV centre, specifically from the eigenstates ms(cid:105) of the spin projection operator Sz along the NV axis, with ms = 0,±1 being the corresponding spin quantum numbers [Fig. 1(a)] [31]. To dress the NV spin states, we simultaneously and coherently drive all three avail- able spin transitions, using microwave (MW) magnetic fields [32] to drive the 0(cid:105) ↔ ± 1(cid:105) transitions and time- varying strain fields [26, 27] to drive the magnetic dipole- forbidden − 1(cid:105) ↔ + 1(cid:105) transition [Fig. 1(a) and SOM]. The resulting CCD dressed states [17] offer superior co- herence protection compared to alternative approaches, which rely on MW driving alone [19]. Specifically, CCD dressed states offer decoupling from magnetic field noise up to fourth order in field amplitude [17], and for magne- tometry, a more than twofold improvement in sensitivity and a 1000-fold increased sensing range [SOM] over pre- vious work on dressed states [19]. The CCD dressed states are best described in an appro- 9 1 0 2 n a J 9 2 ] l l a h - s e m . t a m - d n o c [ 1 v 8 8 4 0 1 . 1 0 9 1 : v i X r a priate rotating frame [33] where, under resonant driving of all three transitions, the system Hamiltonian reads H0(t) =  2 (cid:0)Ω1(t) − 1(cid:105)(cid:104)0 + Ω2(t) + 1(cid:105)(cid:104)0+ + Ω eiΦ − 1(cid:105)(cid:104)+1 + H. c.(cid:1) , (1) with  being the reduced Planck constant. The Hamil- tonian H0(t) depends on the global driving phase Φ (Φ = ϕ + ϕ1 − ϕ2, where ϕ, ϕ1, and ϕ2 are the phases of the driving fields with Rabi frequecies Ω, Ω1, and Ω2, respectively) [17], which we tune to Φ = π/2. We choose this value of Φ, as it allows for a straight-forward derivation of an analytical, purely real STA correction for our system. However, our method is applicable to arbitrary values of Φ using established numerical meth- ods for determining the ensuing STA ramps [SOM]. For the case Φ = π/2, the eigenstates of H0(t) prior to state for Ω1,2(t) = 0 and Ω (cid:54)= 0, are 0(cid:105) and transfer, i.e. ±(cid:105) ≡ ( − 1(cid:105) ∓ i + 1(cid:105))/ 2 [Fig. 1(b), left]. In contrast, the eigenstates of the final system, i.e. for Ω1,2(tr) = Ω, are given by [17] eiπ(1−4k)/6 − 1(cid:105) + 0(cid:105) + e−iπ(1−4k)/6 + 1(cid:105)(cid:17) (cid:16) √ Ψk(cid:105) = 1√ 3 (2) with k = 0,±1 [Fig. 1(b), right]. Thus our state trans- fer protocol consists of spin initialisation into 0(cid:105) with Ω1,2(t = 0) = 0, after which we apply suitable ramps Ω1,2(t) to transfer into the dressed state basis with sym- metric driving of all three transitions, i.e. Ω1,2(t = tr) = Ω [Fig. 1(b)]. We study state transfer between the initial (0(cid:105), ±(cid:105)) and final states (Ψk(cid:105)) under ambient conditions using an experimental setup described elsewhere [17] and by em- ploying the pulse sequence shown in Fig. 1(c). A green laser pulse prepares the initial system in ψ(t = 0)(cid:105) ≡ 0(cid:105). Then, we individually ramp the MW field amplitudes (with ramp time tr) to transfer 0(cid:105) to the dressed state Ψ+1(cid:105) [Fig. 1(b)]. After letting the system evolve in the presence of all three driving fields, we read out the population in 0(cid:105) at time t, p0(t) = (cid:104)0ψ(t)(cid:105)2, using spin-dependent fluorescence. During the whole pulse se- quence, the amplitude of the mechanical driving field is constant at Ω = 2π · 510 kHz, while the mechanical os- cillator is driven near resonance at 5.868 MHz (implying Bz = 1.82 G). To demonstrated state transfer into a dressed state, we first focus on an adiabatic protocol to benchmark our subsequent studies. Inspired by the 'STIRAP' sequence developed for quantum-optical 'Λ-systems' [34, 35], we choose [36] Ω1,2(t) = Ω · sin (θ(t)) , (3) 2 FIG. 1. State transfer schematics. (a) Level scheme of the NV S = 1 ground state spin with spin states ms(cid:105) (magnetic quantum number ms = 0,±1). All three spin transitions are individually and coherently addressable, either by MW mag- netic fields (Ω1,2(t), light and dark blue) or by a cantilever in- duced strain field (Ω, red), enabling a CCD. (b) Level schemes in the rotating frame for appropriate driving field phases [see text]. The initial system (left) comprises the states 0(cid:105) and ±(cid:105), with the ±(cid:105) being equal admixtures of ± 1(cid:105). Ramp- ing the amplitudes of both MW fields causes a transfer to the final dressed states Ψk(cid:105) (k = 0,±1, right). We initialise the system in 0(cid:105), which is for adiabatic ramping transferred to Ψ+1(cid:105) (orange transition). (c) Pulse sequence employed for state transfer. Note that in general Ω1,2(t) have different time dependencies. , [Fig. 2(a)] with θ(t) = · π 2 1 1 + exp(−ν(t − t0)) being a Fermi function with time-shift ln(cid:0)π/(2 sin−1(ε)) − 1(cid:1) /ν tr = t0 − ln(cid:0)π/(2 sin−1(1 − ε)) − 1(cid:1) /ν, while ε (cid:28) 1 sets t0 = parameters slope of θ(t) t = t0 and is connected to the ramp time and ν controls ε and ν. at free the Here, (4) the amplitude of the ramp's unavoidable discontinuities at t = 0 and t = tr. In all our experiments we use ε = 10−3, as this value is comparable to the estimated amplitude noise of our MW signals [SOM]. Figure 2(b) presents the time evolution of p0 for sev- eral values of tr. For fast ramping, i.e. small tr, p0 oscillates even for t > tr; by increasing tr, the ampli- tude of these oscillations reduces, until p0 becomes time- independent with p0 ∼ 1/3. This marks a change from a non-adiabatic to an adiabatic transition with increasing tr [SOM]. Fast, non-adiabatic ramping results in a super- position of dressed states at the end of the ramp. During the subsequent time evolution each dressed state accu- mulates a dynamical phase, resulting in a beating (with 3 Ω/2) in the measured population p0. Con- frequency versely, for larger tr we adiabatically prepare the single dressed state Ψ+1(cid:105), where no such beating occurs and √ 00.20.40.60.81-1-0.500.510〉+〉−〉Ψ0〉Ψ+1〉Ψ−1〉E0E+1E−1Ω1(t)Ω2(t)Ω0〉+1〉−1〉ReadInitILaserΩΩ2TimeΩ1t = 0tr(a)(b)(c)3 therefore purely real. An imaginary component would re- quire control of the phase and amplitude of driving fields, which or mechanical-oscillator mediated strain drive can- not provide on the relevant timescales. For different val- ues of Φ, however, other STA ramps could be found using the dressed state formalism [37] [SOM]. Applying correc- tion (5) to Hamiltonian (1) results in the modified MW pulse amplitudes Ω1,2(t) = Ω · sin (θ(t)) ± 2 · cos(θ(t)) ∂tθ(t) 2 − cos(2θ(t)) , (6) while keeping the phases of all fields constant. Figure 3(a) shows the resulting MW pulse shapes for ν = Ω/2 and Ω = 510 kHz. Note that the TD approach provides dif- ferent corrections for the two MW fields, such that both field amplitudes are ramped successively with different functional forms. Figure 3(b) depicts the experimental result of the state transfer when applying the TD corrected ramps. Inde- pendent of tr, the time evolution of p0 converges to 1/3, which indicates perfect initialisation of a single dressed state, even for the fastest ramps, in striking agreement with the calculations. For TD driving, there exists a lower bound for tr, below which the TD ramps lead to momentary driving field amplitudes either < 0 or > Ω [SOM]. For a fair comparison with adiabatic state trans- fer, we therefore exclude this parameter range from our study [grayed area in Fig. 3(b)]. The fastest possible state transfer corresponds to ν = Ω/2, resulting in tr = 6.8 µs -- the value at which the data in Fig. 3(c) have been ob- tained. Figure 2(c) and Fig. 3(c) allow for a direct compari- son of adiabatic and STA transfer protocols, since both measurements are recorded with the same set of exper- imental parameters. For the first approach, p0 clearly indicates non-adiabatic errors in dressed state initialisa- tion [Fig. 2(c)]. However, for the TD ramp, almost no oscillations in p0 are visible, indicating excellent state transfer [Fig. 3(c)]. The remaining small oscillations are attributed to residual imperfections in the dressed state initialisation, which we discuss in the next paragraph. To achieve a transfer fidelity as determined by these resid- ual oscillations, our calculations show that an adiabatic ramp of at least tr = 17.6 µs would be required, which de- termines the speedup factor of 2.6 we achieve for TD over adiabatic ramping for our given experimental parameters [SOM]. To demonstrate reversibility and to verify that the TD protocol indeed results in initialisation of a single, pure dressed state, we reverse the state transfer and map from the dressed states back to the initial system. Specifically, we use the TD technique presented in Fig. 3 to prepare the system in a single dressed state, and then use an inverted TD protocol (i.e. t → tr − t) to map back to the bare NV state 0(cid:105) [Fig. 4(a)]. Figure 4(b) shows the FIG. 2. Adiabatic state transfer. (a) Experimentally em- ployed ramps for the MW field amplitudes. Both MW fields are ramped simultaneously in an optimised STIRAP pulse shape [see text]. The slope parameter ν is directly linked to the ramp time tr. (b) 2D plots of p0 = (cid:104)0ψ(t)(cid:105)2 as a func- tion of evolution time t and ramp time tr. The left plot shows experimental data, and the right plot shows theoretical cal- culations based on a fully coherent evolution. For small tr os- cillations in the population indicate a non-adiabatic transfer, whereas for slower ramping of the MW fields the state trans- fer becomes adiabatic. The green line indicates the location of the linecut presented in (c). (c) Linecut of measured pop- ulation p0 (green) as function of evolution time t for ν = Ω/2 with corresponding calculation (black). p0 = (cid:104)0Ψ+1(cid:105)2 = 1/3, as observed in the experiment. We corroborate our experimental findings by calculat- ing the time evolution p0(t) using Hamiltonian (1) and find excellent agreement with our data [Fig. 2(b)]. This agreement is additionally highlighted by the linecut in Fig. 2(c) taken in the non-adiabatic regime at tr = 6.8 µs [green dashed line indicated in Fig. 2(b)]. Having established adiabatic state transfer into the dressed state basis, we investigate STAs to speed up the initialisation procedure. Theoretical proposals pro- vide various techniques for STA, including transitionless driving (TD) [21, 22] or the dressed state approach to STA [37]. All techniques harness non-adiabatic transi- tions by adding theoretically engineered corrections to the state transfer Hamiltonian. Adding a TD control results in the correction H0(t) → H0(t) + i(cid:16) ∂t U†(t) (cid:17) U(t) , (5) with U(t) being the transformation operator from {ms(cid:105)} into the adiabatic eigenstate basis [21]. We note that in our experiment, we can only implement the TD correc- tion of Eq. (5) for Φ = π/2, where time reversal symmetry is maximally broken and the resulting TD correction is 05101500.51010205101520010102000.5100.51(a)(c)(b)CalculationMeasurementtr = 6.8 µst0/tr0〉Ψ+1〉4 Having shown efficient initialisation of a single, pure dressed state and subsequent dressed-state population readout, we next demonstrate coherent manipulation of dressed sates by performing electron spin resonance (ESR) and Rabi nutation measurements in the dressed state basis. For this, we apply an additional MW ma- nipulation field of Rabi frequency Ωman in between the initialisation and remapping procedures [Fig. 4(a)]. By sweeping the frequency of this manipulation field (at a constant pulse duration τ = 45 µs) across the 0(cid:105) ↔ − 1(cid:105) transition of the NV states, we observe two dips [Fig. 4(d)], corresponding to dressed state transitions at positive and negative frequencies in the rotating frame, i.e. at symmetric detunings around the bare 0(cid:105) ↔ − 1(cid:105) transition frequency (note that the two possible transi- tion from Ψ+1(cid:105) to either Ψ0(cid:105) or Ψ−1(cid:105) [light green ar- rows in Fig. 1(b)] occur at the same frequencies and are therefore indistinguishable). Lastly, by resonant driving of the dressed state transitions for varying durations τ , we demonstrate coherent Rabi oscillations [Fig. 4(e)] and therefore coherent dressed state manipulation. We have shown high-fidelity, reversible initialisation of individual dressed states in a CCD scheme using STA state transfer protocols, for which we demonstrated a more than twofold speedup over the adiabatic approach with state transfer fidelities > 99%. This performance is the direct result of our combination of STA (provid- ing fast, high efficiency initialisation) and CCD dressed states (offering close to 100-fold improvement in coher- ence times compared to the other continuous mechanical or MW driving schemes under similar conditions [17]). Our results provide a basis for future exploitation of dressed states, building on the coherent control of dressed states we have demonstrated. In particular, while the ef- ficiency of coherence protection in CCD has been demon- strated recently [17], details of additional dressed state dephasing mechanisms remain unknown and could be explored by employing noise spectroscopy [38] and dy- namical decoupling [39 -- 41], directly in the dressed state basis. Owing to the prolonged dressed state coherence times over the bare spin states [17, 19, 20, 42], our tech- nique could be used to efficiently store particular NV spin states by mapping to the dressed state basis [43, 44] on timescales much longer than the coherence times of the bare NV states. Lastly, owing to its versatility and sta- bility, the experimental system we established here forms an attractive platform to implement and test novel state transfer protocols, which may emerge from future theo- retical work. We thank A. Stark for fruitful discussions and valu- able input. We gratefully acknowledge financial support through the NCCR QSIT, a competence centre funded by the Swiss NSF, through the Swiss Nanoscience Institute, by the EU FP7 project ASTERIQS (grant #820394) and through SNF Project Grant 169321. FIG. 3. STA state transfer protocol. (a) Envelope of the opti- mised MW field amplitudes for ν = Ω/2 and Ω/2π = 510 kHz. Modifications of the adiabatic pulse shape (dashed) lead to al- tered and unequal ramps for the two MW fields (Ω1 light blue, Ω2 dark blue). (b) 2D plots of p0 as a function of evolution time t and ramp time tr. The left plot shows experimen- tal data, and the right plot the theoretical calculations. We achieve state transfer with high fidelity independent of tr. The green line indicates the location of the linecut presented in (c) with ramp parameters at the experimental limit (border of grayed area). (c) Linecut of measured population p0 (green) as function of evolution time t for the same parameters as in Fig. 2(c) with corresponding calculation (black). time-evolution of p0 as we apply the remapping proto- col, where we set ν = Ω/2 (maximal ramping speed) for both directions. Clearly, almost all of the population in the dressed state returns to 0(cid:105), thereby indicating co- herent, reversible population transfer between undressed and dressed states. Additionally, such measurements al- low us to quantify the efficiency of a single state transfer under the fair assumption that mapping in and mapping out yield the same transfer fidelity. We quantify the fi- delity by repeatedly mapping in and out of the dressed state basis, with each set of one mapping in and one mapping out constituting a single 'remapping cycle'. We vary the number of remapping cycles N and read out the population p0 at the end [Fig. 4(c)]. An exponential fit then yields the fidelity F = 99.4(2) % for a single transfer process. This transfer fidelity is experimentally limited by uncertainties in setting the global phase Φ, leakage of the MW signals, non-equal driving field amplitudes, and unwanted detunings of the driving fields. Although we calculate our ramps assuming equal driving amplitudes and zero detunings, violations of these assumptions are experimentally unavoidable, and the errors will generally fluctuate in time [17]. These factors are also responsible for the remaining, small oscillations in p0 visible after state transfer in Fig. 3(c) [SOM]. 010205101520010102005101500.5100.5100.51(a)(c)(b)tr = 6.8 µs0〉Ψ+1〉CalculationMeasurementt0/trExperimentally inaccessibleExperimentally inaccessible5 [7] J. Clarke and F. K. Wilhelm, Nature 453, 1031 (2008). [8] R. Hanson and D. D. Awschalom, Nature 453, 1043 (2008). [9] M. A. Nielsen and I. L. Chuang, Quantum computation and quantum information (Cambridge University Press, Cambridge, 2010). [10] L. Viola, E. Knill, and S. Lloyd, Physical Review Letters 82, 2417 (1999). [11] G. S. Uhrig, Physical Review Letters 98, 100504 (2007). [12] J. Du, X. Rong, N. Zhao, Y. Wang, J. Yang, and R. B. Liu, Nature 461, 1265 (2009). [13] C. M. Wilson, T. Duty, F. Persson, M. Sandberg, G. Jo- hansson, and P. Delsing, Physical Review Letters 98, 257003 (2007). [14] J.-M. Cai, B. Naydenov, R. Pfeiffer, L. P. McGuinness, K. D. Jahnke, F. Jelezko, M. B. Plenio, and A. Retzker, New Journal of Physics 14, 113023 (2012). [15] D. A. Golter, T. K. Baldwin, and H. Wang, Physical Review Letters 113, 237601 (2014). [16] J. Teissier, A. Barfuss, and P. Maletinsky, Journal of Optics 19, 044003 (2017). [17] A. Barfuss, J. Kolbl, L. Thiel, J. Teissier, M. Kasperczyk, and P. Maletinsky, Nature Physics 14, 1087 (2018). [18] N. Timoney, I. Baumgart, M. Johanning, A. F. Var´on, M. B. Plenio, A. Retzker, and C. Wunderlich, Nature 476, 185 (2011). [19] X. Xu, Z. Wang, C. Duan, P. Huang, P. Wang, Y. Wang, N. Xu, X. Kong, F. Shi, X. Rong, and J. Du, Physical Review Letters 109, 070502 (2012). [20] P. London, J. Scheuer, J.-M. Cai, I. Schwarz, A. Ret- zker, M. B. Plenio, M. Katagiri, T. Teraji, S. Koizumi, J. Isoya, R. Fischer, L. P. McGuinness, B. Naydenov, and F. Jelezko, Physical Review Letters 111, 067601 (2013). [21] M. Demirplak and S. A. Rice, The Journal of Physical Chemistry A 107, 9937 (2003). [22] M. V. Berry, Journal of Physics A: Mathematical and Theoretical 42, 365303 (2009). [23] A. del Campo, Phys. Rev. Lett. 111, 100502 (2013). [24] X. Chen, I. Lizuain, A. Ruschhaupt, D. Gu´ery-Odelin, and J. G. Muga, Phys. Rev. Lett. 105, 123003 (2010). [25] B. B. Zhou, A. Baksic, H. Ribeiro, C. G. Yale, F. J. Here- mans, P. C. Jerger, A. Auer, G. Burkard, A. A. Clerk, and D. D. Awschalom, Nature Physics 13, 330 (2016). [26] E. R. MacQuarrie, T. A. Gosavi, A. M. Moehle, N. R. Jungwirth, S. A. Bhave, and G. D. Fuchs, Optica 2, 233 (2015). [27] A. Barfuss, J. Teissier, E. Neu, A. Nunnenkamp, and P. Maletinsky, Nature Physics 11, 820 (2015). [28] T. Joas, A. Waeber, G. G. Braunbeck, and F. Reinhard, Nature Communications 8, 964 (2017). [29] A. Stark, N. Aharon, T. Unden, D. Louzon, A. Huck, and F. Jelezko, Nature A. Retzker, U. L. Andersen, Communications 8, 1105 (2017). [30] A. Gruber, A. Drabenstedt, C. Tietz, L. Fleury, J. Wrachtrup, and C. v. Borczyskowski, Science 276, 2012 (1997). [31] M. W. Doherty, N. B. Manson, P. Delaney, F. Jelezko, J. Wrachtrup, and L. C. Hollenberg, Physics Reports 528, 1 (2013). [32] F. Jelezko, T. Gaebel, and J. Wrachtrup, Physical Review Letters 92, 076401 (2004). I. Popa, A. Gruber, [33] S. J. Buckle, S. M. Barnett, P. L. Knight, M. A. Lauder, and D. T. Pegg, Optica Acta: International Journal of FIG. 4. Transfer fidelity and dressed state characterisation. (a) Pulse sequence employed for bidirectional state transfer and manipulation of the dressed states. (b) Inverted state transfer to the initial system after the dressed state Ψ+1(cid:105) was prepared. Measured population p0 (green) as function of the remapping time t(cid:48) for ν = Ω/2 with corresponding simulation (black). (c) Experimental transfer fidelity for various numbers of completed (re-)mapping cycles. The exponential fit to the data (dashed) yields a statistical transfer fidelity of 99.4(2) %. (d) Transition frequencies of the dressed states after initiali- sation of Ψ+1(cid:105) measured with an additional probe field Ωman, whose frequency is given relative to the 0(cid:105) ↔ − 1(cid:105) transition of 2.8672 GHz. (e) Rabi oscillations on a dressed state tran- sition extracted from (d) with ΩRabi = 2π · 131(1) kHz and Tdec = 24(3) µs. ∗ patrick.maletinsky@unibas.ch [1] J. A. Jones, S. D. Karlen, J. Fitzsimons, A. Ardavan, S. C. Bejamin, G. A. D. Brigg, and J. J. L. Morton, Science 324, 1166 (2009). [2] J. M. Taylor, P. Cappellaro, L. Childress, L. Jiang, D. Budker, P. R. Hemmer, A. Yacoby, R. Walsworth, and M. D. Lukin, Nature Physics 4, 810 (2008). [3] D. Loss and D. P. DiVincenzo, Physical Review A 57, 120 (1998). [4] B. E. Kane, Nature 393, 133 (1998). [5] J. P. Home, D. Hanneke, J. D. Jost, J. M. Amini, D. Leibfried, and D. J. Wineland, Science 325, 1227 (2009). [6] U. G. Poschinger, G. Huber, F. Ziesel, M. Deiss, M. Het- trich, S. A. Schulz, K. Singer, G. Poulsen, M. Drewsen, R. J. Hendricks, and F. Schmidt-Kaler, Journal of Physics B: Atomic, Molecular and Optical Physics 42, 154013 (2009). 1234560.91051000.51tr = 6.8 µs(b)(c)(d)(e)(a)ReadInitILaserΩΩ2TimeΩ1trΩmanτtr(d,e)t' = 0-1-0.500.510.9101020304000.51τ = 0 µs6 Optics 33, 1129 (1986). [34] K. Bergmann, H. Theuer, and B. W. Shore, Rev. Mod. Phys. 70, 1003 (1998). [35] N. V. Vitanov, A. A. Rangelov, B. W. Shore, and K. Bergmann, Review of Modern Physics 89, 015006 (2017). [36] G. S. Vasilev, A. Kuhn, and N. V. Vitanov, Physical Review A 80, 013417 (2009). [37] A. Baksic, H. Ribeiro, and A. A. Clerk, Physical Review Letters 116, 230503 (2016). [38] N. Bar-Gill, L. Pham, C. Belthangady, D. Le Sage, and P. Cappellaro, J. Maze, M. Lukin, A. Yacoby, R. Walsworth, Nature Communications 3, 858 (2012). [39] C. A. Ryan, J. S. Hodges, and D. G. Cory, Physical Review Letters 105, 200402 (2010). [40] G. de Lange, Z. H. Wang, D. Rist`e, V. V. Dobrovitski, and R. Hanson, Science 330, 60 (2010). [41] B. Naydenov, F. Dolde, L. T. Hall, C. Shin, H. Fedder, L. C. L. Hollenberg, F. Jelezko, and J. Wrachtrup, Phys- ical Review B 83, 081201 (2011). [42] E. R. MacQuarrie, T. A. Gosavi, S. A. Bhave, and G. D. Fuchs, Phys. Rev. B 92, 224419 (2015). [43] C. Simon, M. Afzelius, J. Appel, A. Boyer de la Giro- day, S. J. Dewhurst, N. Gisin, C. Y. Hu, F. Jelezko, S. Kroll, J. H. Muller, J. Nunn, E. S. Polzik, J. G. Rarity, H. De Riedmatten, W. Rosenfeld, A. J. Shields, N. Skold, R. M. Stevenson, R. Thew, I. A. Walmsley, M. C. We- ber, H. Weinfurter, J. Wrachtrup, and R. J. Young, The European Physical Journal D 58, 1 (2010). [44] H. P. Specht, C. Nolleke, A. Reiserer, M. Uphoff, E. Figueroa, S. Ritter, and G. Rempe, Nature 473, 190 (2011). [45] P. Jamonneau, M. Lesik, J. P. Tetienne, I. Alvizu, L. Mayer, A. Dr´eau, S. Kosen, J.-F. Roch, S. Pezzagna, J. Meijer, T. Teraji, Y. Kubo, P. Bertet, J. R. Maze, and V. Jacques, Phys. Rev. B 93, 024305 (2016). [46] A. Chaudhry, Physical Review A 90, 042104 (2014). [47] C. Degen, F. Reinhard, and P. Cappellaro, Reviews of Modern Physics 89, 035002 (2017). "Initialisation of single spin dressed states using shortcuts to adiabaticity" Supplementary Material for 7 OVERVIEW In the first part we describe our readout mechanism for the dressed state preparation measurements in detail and discuss the adiabatic criterion for the state transfer. Next, we present a derivation of the transitionless driving (TD) correction. We then generalize our STA approach, where we calculate the required corrections for arbitrary phase values. In the fourth part, we numerically calculate the Quantum Fisher information of our dressed states, from which we estimate their projection-noise-limited sensitivity. In addition, we discuss the level structure of the NV and how we form a CCD scheme. Finally, we provide a detailed description of the MW field generation and control. CHARACTERISATION OF STATE PREPARATION PROCESS Our experimental observations and theoretical calculations in Fig. 2 and Fig. 3 of the main text are based on the readout of the population (7) where ψ(t)(cid:105) is the system's state after the time evolution t [see Fig. 1(c) of the main text]. The Hamiltonian determining the evolution of ψ(t)(cid:105) is H0(tr) [see Eq. (1) of the main text], with Ω1,2(tr) = Ω, which leads to p0 = (cid:104)0ψ(t)(cid:105)2 , ψ(t)(cid:105) = e−i(t−tr) H0(tr)/ψ(tr)(cid:105) . (8) If we prepare our system in a single dressed state, i. e. ψ(tr)(cid:105) = Ψ+1(cid:105), the final state is ψ(t)(cid:105) = e−i(t−tr) H0(tr)/Ψ+1(cid:105) = e−i(t−tr)E+1/Ψ+1(cid:105) , (9) with E+1 being the eigenenergy corresponding to Ψ+1(cid:105), which characterises the dynamical phase the eigenstate √ accumulates during its evolution. As our readout state is 0(cid:105) = (Ψ−1(cid:105) + Ψ0(cid:105) + Ψ+1(cid:105))/ 3, the measured population p0 can be expressed as (cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) (cid:88) (cid:68) k=0,±1 p0 = 1 3 (cid:12)(cid:12)(cid:12)e−i(t−tr)E+1/(cid:12)(cid:12)(cid:12) Ψ+1 Ψk (cid:69)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)2 = 1 3 . (10) Thus, for a perfect state transfer into the state Ψ+1(cid:105), the measured population is time independent with a value of 1/3. The same holds for the other two dressed states, Ψ0(cid:105) and Ψ−1(cid:105). If, however, we do not prepare a single dressed state, but rather a mixture of dressed states, we start the time j cj2 = 1. In this case, the final state of the evolution is evolution in ψ(tr)(cid:105) =(cid:80) given by as each dressed state Ψj(cid:105) accumulates a dynamical phase corresponding to its eigenenergy Ej (j = 0,±1). During readout we then measure cjΨj(cid:105) = cj e−i(t−tr)Ej /Ψj(cid:105) , (11) j=0,±1 cjΨj(cid:105), with cj ∈ C and(cid:80) ψ(t)(cid:105) = e−i(t−tr) H0(tr)/ (cid:88) (cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) (cid:12)(cid:12)(cid:12)cj e−i(t−tr)Ej /(cid:12)(cid:12)(cid:12) Ψj (cid:88) j,k=0,±1 j=0,±1 (cid:68) Ψk j=0,±1 (cid:88) (cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) 1 3 = (cid:69)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)2 p0 = 1 3 (cid:88) j=0,±1 (cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)2 cj e−i(t−tr)Ej / , (12) resulting in a time dependence of p0 characterized by a beating of the transition frequencies of the dressed states with amplitudes depending on the weighting factors cj. 8 With that, we can explain the observed transition from an oscillatory to a constant time evolution with increasing ramp time tr in Fig. 2(b) of the main text. For fast ramping, i. e. small tr, the preparation is non-adiabatic, and we therefore prepare a mixture of dressed states, resulting in an oscillatory p0, which we observe experimentally. Increasing tr decreases the amplitude of these oscillations, indicating that one weighting factor becomes dominant, so that the dressed state mixing is reduced. If we finally reach the adiabatic regime, only a single dressed state is prepared resulting in a time independent evolution of the measured population. To quantify the transition from the non-adiabatic to the adiabatic regime we compare the energy separation of the instantaneous adiabatic eigenstates with their mutual coupling. Therefore, we calculate Hamiltonian H0(t) [see Eq. (1) of the main text] in the adiabatic basis [21]: H0,ad(t) = U(t) H0(t) U†(t) − i U(t) ∂t U†(t) . (13) Here, the time dependent unitary transformation U(t) is the transformation operator from the {ms(cid:105)} basis to the adiabatic basis states (i. e. the instantaneous eigenvectors of H0(t)). We obtain U(t) from its inverse U†(t), whose columns contain the time dependent, normalised adiabatic eigenvectors. Considering the { − 1(cid:105),0(cid:105), + 1(cid:105)} basis we find  0 sin(θ(t)) −i  , sin(θ(t)) i 0 sin(θ(t)) sin(θ(t)) 0 H0(t) =  Ω 2 and U†(t) = √ i 2−cos(2θ(t)) 2 sin(θ(t)) 2−cos(2θ(t)) √ i 2−cos(2θ(t)) 2 √ − sin(θ(t)) 2−cos(2θ(t)) i√ − 2−cos(2θ(t)) √ sin(θ(t)) 2−cos(2θ(t)) 2 1 2−cos(2θ(t)) √ √ 2−cos(2θ(t)) √ 2 − − sin(θ(t)) 1 2 + 2−cos(2θ(t)) 2 i i Taking the Hermitian conjugate of Eq. (15) and inserting into Eq. (13) yields (cid:112)2 − cos(2θ(t)) cos(θ(t)) ∂tθ(t) 2−cos(2θ(t)) 0 H0,ad(t) =  cos(θ(t)) ∂tθ(t) 2−cos(2θ(t)) 0 − cos(θ(t)) ∂tθ(t) 2−cos(2θ(t)) − Ω 2 0 − cos(θ(t)) ∂tθ(t) 2−cos(2θ(t)) (cid:112)2 − cos(2θ(t)) 1 2 + √ 2 − 1   Ω 2  .  . (14) (15) (16) In order to realise an adiabatic state transfer, the coupling between two instantaneous adiabatic eigenstates has to be much smaller than their energy separation, i. e. we have to compare the skew diagonal elements of Eq. (16) with the separation of the diagonal elements. Thus, in our case the adiabatic criterion reads (cid:28) 1 . (17) To estimate the transition from satisfying to violating the adiabatic criterion, we choose an upper limit for the right-hand side of 1/10 for all times t, i. e. 2 cos(θ(t)) ∂tθ(t) Ω(cid:0)2 − cos(2θ(t))(cid:1)3/2 (cid:40) Ω(cid:0)2 − cos(2θ(t))(cid:1)3/2 2 cos(θ(t)) ∂tθ(t) (cid:41) max ≤ 1 10 . (18) Thus in this paper we call a ramp time tr which satisfies this inequality adiabatic, and we call a ramp time that violates this inequality non-adiabatic. With this definition, we find that tr = 12.9 µs is the critical ramp time for an adiabatic transition. This corresponds to a theoretical fidelity F = (cid:104)Ψ+ψ(tr)(cid:105)2 of (1 − 3.2 · 10−3), which is lower than the asymptotic fidelity of (1 − 2 · 10−6) by 0.35 % percent. 9 FIG. S1. Theoretically calculated TD corrected MW ramp shapes for ν > Ω/2. (a) For ν = 3Ω/4 the envelope of the optimised ramp has negative amplitude values, which cannot be implemented experimentally. (b) Increasing ν further additionally leads to overshoots, e. g. here displayed for ν = 5Ω/4. DERIVATION OF THE TD CORRECTION   . To calculate the TD correction of our microwave (MW) pulses stated in Eq. (6) of the main text, we consider the Hamiltonian H0,ad(t) in the adiabatic basis [see Eq. (16)]. In order to get rid of the off-diagonal elements, i. e. to ensure a transfer on a single adiabatic eigenstate, additional control fields expressed by a control Hamiltonian HTD(t) have to be added to the system. As could be anticipated by inspecting Eq. (13), a general expression for the TD control Hamiltonian in the {ms(cid:105)} basis reads [21]: HTD(t) = i(cid:16) ∂t U†(t) (cid:17) U(t) . (19) Here, U†(t) is the unitary operator given in Eq. (15). Inserting the corresponding expressions into Eq. (19) we yield in the { − 1(cid:105),0(cid:105), + 1(cid:105)} basis: HTD(t) =  0 cos(θ(t)) ∂tθ(t) 2−cos(2θ(t)) 0 cos(θ(t)) ∂tθ(t) 2−cos(2θ(t)) 0 − cos(θ(t)) ∂tθ(t) 2−cos(2θ(t)) 0 − cos(θ(t)) ∂tθ(t) 2−cos(2θ(t)) 0 (20) After adding this Hamiltonian to the initial Hamiltonian H0(t) from Eq. (14), we find that the TD corrected MW pulses are Ω1,(2)(t) = Ω · sin(θ(t)) + (−) 2 · cos(θ(t)) ∂tθ(t) 2 − cos(2θ(t)) . (21) In the experiment we can easily realize TD corrected MW pulses with 0 ≤ Ω1,(2)(t) ≤ Ω, which require the parameter ν to be ν ≤ Ω/2. If we choose ν > Ω/2, however, the TD corrected ramps would lead to negative amplitude values, which we cannot implement experimentally, since flipping the phase of the signal generation fields does not influence the final driving field phases (see later section for details on the MW pulse generation). Figure S1(a) displays an example for such a ramp for ν = 3Ω/4. For even higher values of ν some amplitudes additionally overshoot the steady state driving amplitude (see Fig. S1(b) for ν = 5Ω/4). As mentioned in the main text, the theoretically derived TD corrections do not account for the unavoidable ex- perimental uncertainties that cause the residual small oscillations in p0 after the state transfer in Figure 3(c). Most importantly, our measurements are affected by slow fluctuations in the magnetic environment of the NV caused by nearby nuclear spins (14N or 13C) and by uncertainties in setting the driving field parameters. The magnetic fluctuations are characterized by a zero-mean Gaussian distribution of the MW field detunings with a width of 2 = 2.1(1) µs is the coherence time determined through a Ramsey σT ∗ experiment. The detuning is typically constant during a single measurement,but changes between measurements, i.e. the timescale of the fluctuations is long compared to a single measurement, but far shorter than the total mea- surement time. The uncertainties in setting the driving field parameters mostly affect the MW driving strengths. We measure the driving strengths by driving Rabi oscillations on each of the MW transitions of the bare NV and 2 ) = 107 kHz [45], where T ∗ 2 √ /2π = 1/( 2πT ∗ 00.5100.5100.5100.51(a)(b)t0/trt0/tr10 extracting the Rabi frequency by fitting with an exponentially decaying single sinusoid. But due to fluctuations in, for example, the sample-antenna separation, as well as the microwave amplifier, the measured Rabi frequencies show relative deviations of up to 2 % for the same applied microwave power (∼ 4 dBm from the SRS microwave generator in our case). Moreover, setting the global phase value exactly to Φ = π/2 has experimental limitations. We determine the corresponding phase value by sweeping the mechanical phase with finite sampling rate while maintaining the MW fields constant [see later] and fitting the averaged linecuts of the resulting interference pattern [compare to [17]] for evolution times between 1 µs and 1.7 µs. This allows us to determine the value of the global phase with a 2σ uncertainty of 0.9◦. Figure S2 shows the simulated population p0 as a function of evolution time t averaged over Navg = 100 normally distributed MW detunings, while also including other experimental uncertainties in a worst-case scenario (i.e. maximum global phase error, and maximum deviation in drive strengths). The time evolution clearly shows oscillations after the state transfer, similar to those observed in the experiment. We note that there are additional sources for experimental uncertainties that we neglect in our simulation, e. g. the feedthrough of the MW signals (which leads to non-vanishing MW amplitudes at the beginning of the state transfer), amplitude and frequency noise of the driving fields, as well as fluctuations in the zero-field splitting of the NV induced by variations in temperature or environmental strain or electric fields. FIG. S2. Simulated influence of experimental uncertainties on the state preparation process. While the unperturbed system shows no oscillations after the state transfer (dash-dotted line in black), including experimental uncertainties reveals the experimentally observed oscillations (solid line in orange). Parameters: ν = Ω/2, Ω/2π = 510 kHz, Ωmax /2π = 520 kHz, /2π = 500 kHz (i. e. 2 % opposing deviation for both MW fields), Φ = 0.495 · π/2. The perturbed simulation is averaged Ωmax over Navg = 100 times, with each iteration using a different MW detuning drawn from a zero-mean Gaussian distribution characterized by σT ∗ /2π = 107 kHz. 1 2 2 FINDING STAS FOR ARBITRARY PHASE VALUES STAs for arbitrary phase values Φ can be found for our current setup (no time-dependent phase control on the MW fields and on the amplitude of the mechanical drive) using the dressed state approach introduced in Ref. [37]. The basic idea is to modify the control fields of the Hamiltonian H(t) [see Eq. (1) of the main text], so that the state transfer is realized without errors even when the protocol time is not long compared to the instantaneous adiabatic gap. The modification of the control fields can be described as a modification of the initial Hamiltonian by adding an extra control term, i. e. H(t) → Hmod(t) = H(t) + W(t). We can parametrize W(t) as W(t) = W1(t) − 1(cid:105)(cid:104)0 + W2(t) + 1(cid:105)(cid:104)0 + H.c. , (22) where for our experimental setup W1(t) and W2(t) are real valued functions. We, however, note that if time-dependent phase control is available, then W1(t) and W2(t) can be complex valued. We also note that Eq. (22) can be further generalized to take into account possible time-dependent control of the mechanical drive. There are an infinity of possible W(t), each of them being associated to a specific dressing of the instantaneous eigenstate used to realized the adiabatic state preparation. We choose W(t) such that in the dressed adiabatic frame the evolution of the dressed state used for state preparation is trivial, i. e. one remains in this particular dressed state for the whole duration of the protocol. In our case the dressing transformation can be generated by the unitary operator Udress(t) = eiϕj (t)λj , (23) (cid:89) j 05101500.51where λj are the generators of SU(3) and ϕj(t) must obey the conditions ϕj(0) = ϕj(tr) = 0 to ensure that the dressed states correspond to the adiabatic states of Eq. (1) of the main text at the beginning and end of the protocol. This ensures that one starts and ends in the desired state. To make things concrete let us explicitly write the equations that help us determine W(t). We start by finding the adiabatic states (instantaneous eigenstates) of Eq. (1) of the main text. We need to solve the eigenvalue problem 11 We find the instantaneous eigenenergies where Ω(t) = Ω1(t) = Ω2(t) = Ω sin(θ(t)). The associated eigenvectors are given by H(t)ψk(t)(cid:105) = Ek(t)ψk(t)(cid:105) . (cid:18) (cid:112)1 + 2Ω(t)2 cos (cid:20) 1 (cid:21) (cid:2)Ω(t)(cid:0)Ωe−iΦ + 2Ek(t)(cid:1) + 1(cid:105) −(cid:0)Ω2 − 4Ek(t)(cid:1)0(cid:105) cos(Φ) Ω(t)2 (1 + 2Ω(t)2) 3 − 2π 3 √ 3 (cid:19) arccos 1 +Ω(t)(cid:0)ΩeiΦ + 2Ek(t)(cid:1) − 1(cid:105)(cid:3) , Nk(t) k 3 3 2 , (24) (25) (26) Ek(t) = Ω√ 3 ψk(t)(cid:105) = k(t))2 + 2Ω2(t) [Ω2 + 4Ek(t) (Ω cos(Φ) + Ek(t))] the normalization factor. To transform Eq. (1) of the main text to the adiabatic frame (the instantaneous eigenstate basis), we define the frame-change operator given by the time-dependent unitary with Nk(t) = (cid:112)(Ω2 − 4E2 (cid:88) k=±1,0 U(t) = ψk(t)(cid:105)(cid:104)ψk , which diagonalizes H(t) at each instant in time. We can now express the dressed adiabatic Hamiltonian as Hdress(t) = U† (cid:16) (cid:104) U†(t) H(t) U(t) + i (cid:17) Udress(t) dress(t) (cid:16) dress(t) ∂t U† + i (cid:17) U(t) + U†(t) W(t) U(cid:105) Udress(t) ∂t U†(t) and the dressed adiabatic states as ψk(t)(cid:105) = U† dress(t)ψk(cid:105) . (29) Given that the adiabatic state transfer is realized through ψ+1(t)(cid:105), we ask that ψ+1(t)(cid:105) is decoupled from the other dressed states, which ensures the desired trivial dynamics. This condition can be written as where the states ψk(cid:105) are time-independent since they are expressed in the frame defined by U† system of equation for a chosen dressing gives W(t). dress(t). Solving this (cid:104) ψ+1 Hdress(t) ψj(cid:105) = 0 j = 0,−1 , (30) For instance, for Φ = 0 the TD correction only has purely imaginary matrix elements and the SATD correction (see Ref. [37]) requires controlling the detunings of all states and the amplitude of the mechanical drive. None of these corrections can be implemented with our current setup. However, using the dressed state method one can find a STA that respects the constraints of our system. We choose a dressing of the form 2 (ψ0(cid:105)(cid:104)ψ+1+H.c.) . (31) Using Eq. (30), we find that the equation (cid:104) ψ+1 Hdress(t) ψ−1(cid:105) = 0 can be fulfilled by choosing W1(t) = W2(t) while (cid:104) ψ+1 Hdress(t) ψ0(cid:105) = 0 can be fulfilled by solving Udress,Φ=0(t) = e−i β(t) 1 + 8 sin2(θ(t))(cid:2)Ω(cid:0)1 + 8 sin2(θ(t)) + 16W1(t) sin(θ(t))(cid:1)(cid:3) 2 cos(θ(t)) θ(t) √ 4   (cid:113) β(t) = arctan √ (cid:113) 0 = 2 W1(t) 1 + 8 sin2(θ(t)) − β(t) 2 . Inserting Eq. (32) into Eq. (33) one gets a differential equation for W1(t). (27) (28) (32) (33) 12 QUANTUM FISHER INFORMATION OF DRESSED STATES To estimate how suitable our CCI three-level dressed states are for sensing, we calculate their quantum Fisher information (QFI) and compare it to the QFI of three-level dressed states formed with two microwave driving fields [19]. √ Using the QFI we estimate the sensitivity of these states and show that our CCI dressed states can achieve projection- noise-limited sensitivities down to ≈ 2.75 nT/ Hz for the two microwave (2MW) dressed states (see below for formulae giving the driving Hamiltonian and resulting 2MW dressed states). Additionally, we show that our CCI dressed states can sense a ∼ 1000 times broader range of magnetic field strengths. √ Hz, compared to ≈ 6.43 nT/ We consider the case in which we use our states to sense a static magnetic field Bz. We are therefore interested in the QFI J(Bz) of our states as a function of Bz. The QFI is defined as J(Bz) = Tr{(∂Bz ρ) LBz} , (34) where ∂Bz ρ indicates the derivative of the density matrix ρ with respect to Bz [46]. The operator LBz is given by (cid:88) k,(cid:96) LBz = 2 (cid:104)φk∂Bz ρφ(cid:96)(cid:105) ρk + ρ(cid:96) φk(cid:105)(cid:104)φ(cid:96) , (35) where the sum is over all eigenstates φk(cid:105) (with eigenvalues ρk) of ρ, such that ρk + ρ(cid:96) (cid:54)= 0 [46]. The QFI sets a lower bound on the variance δB2 z of our estimate of Bz through the quantum Cram´er-Rao bound δB2 z ≥ 1 J(Bz) (36) for an unbiased estimator of a single measurement of Bz [46]. We assume that we can find the optimal estimator such that the equality obtains. The sensitivity is broadly defined as the weakest magnetic field that can be sensed while still achieving an signal-to-noise ratio (SNR) of 1 after N measurements [47]. We define the SNR as SNR = √ so that an SNR of 1 corresponds to Bmin = δBz/ N . The optimal evolution time for a single measurement of Bz is approximately the coherence time of the sensing state, such that for a total measurement time T we can make N = T /Tcoh measurements (neglecting state preparation and measurement overhead times) [47]. In terms of Tcoh and J(Bz), we therefore have (37) , √ N Bz δBz Bmin = √ √ T(cid:112)J(Bmin; Tcoh) Tcoh √ (cid:112)J(Bmin; Tcoh) Tcoh S = , (38) where we have indicated that we evaluate J(Bz) at the coherence time Tcoh and the minimum field Bmin that can be sensed with an SNR of 1. The sensitivity S (which characterizes the weakest measurable magnetic field, in units of T/ Hz) is thus given by √ . (39) To determine the sensitivity of our states, we therefore need to know their coherence times and their QFI. Since the analytic expression for J(Bz; Tcoh) is unknown, in practice we numerically evaluate Eq. (38) to approximate Bmin, Tcoh}. Once we find Bmin, we can i. e. we take Bmin to be the value of Bz that gives min{1 − Bz evaluate Eq. (39) to find the sensitivity. This constitutes a lower bound on the experimentally achievable sensitivity, as we have neglected the effects of, for example, photon collection efficiency and state preparation fidelity. Because other dressed states exhibit the same overheads, however, this approximation is sufficient. √ (cid:112)J(Bz; Tcoh)/ We find the QFI of our CCI dressed states by first numerically calculating the density matrix ρCCI. After our initialization pulses to prepare our system, the Hamiltonian in Eq. (1) of the main text becomes time independent: (cid:0) − 1(cid:105)(cid:104)0 + + 1(cid:105)(cid:104)0 + eiΦ − 1(cid:105)(cid:104)+1 + H.c.(cid:1) . HCCI = Ω 2 (40) 13 FIG. S3. Quantum Fisher information neglecting decoherence. (a) Plot of JCCI(Bz; t) at Φ = 0 showing that the superposition of Ψ+1(cid:105) and Ψ−1(cid:105) is sensitive to a wide range of magnetic fields, including weak fields. (b) For Φ = π/2 (the value of Φ used to prepare a single dressed state in the main text), JCCI(Bz; t) drops off for low magnetic fields. (c) For Φ = π/4 (the value of Φ at which the superposition state has the longest coherence time), JCCI(Bz; t) appears very similar to the case of Φ = π/2. (d) Plot of J2MW showing that the 2MW dressed states are sensitive to a much narrower range of magnetic field strengths. The general expression for the dressed states for any global phase Φ is given by (cid:16) ei(Φ/3−2πk/3) − 1(cid:105) + 0(cid:105) + e−i(Φ/3−2πk/3) + 1(cid:105)(cid:17) Ψk(cid:105) = 1√ 3 . (41) Following Degen et al. [47], we use sensing states that are superpositions of the dressed states, such that our initial state is of the form ψinit(cid:105) = 1√ (Ψk(cid:105) + Ψ(cid:96)(cid:105)), {k, (cid:96)} ∈ {−1, 0, 1}, k (cid:54)= (cid:96) (i. e. we work within a two-dimensional subspace determined by the two dressed states Ψk(cid:105) and Ψ(cid:96)(cid:105)). We consider the sensing of a static magnetic field, given by the signal Hamiltonian HB/ = 2πγNVBz Sz, where Sz is the S = 1 spin matrix along the NV quantization axis, Bz is the static magnetic field we would like to sense, and γNV = 2.8 MHz/G is the gyromagnetic ratio of the NV. The total Hamiltonian is therefore given by H = HCCI + HB. We numerically solve the equation of motion for the density matrix ρCCI 2 (cid:104) ρCCI, H/(cid:105) ∂ ρCCI ∂t = −i (42) to find ρCCI as a function of evolution time t, magnetic field strength Bz, and global phase Φ. From ρCCI we can calculate the QFI for the initial state ψinit(cid:105) = 1√ (Ψ−1(cid:105) + Ψ+1(cid:105)). Figure S3(a)-(c) shows plots of JCCI(Bz; t) at 2 100806040200Evolution time (μs)100806040200Evolution time (μs)050100Bz (μT)050100Bz (μT)(μT)-2020406080(a)(b)(d)(c)14 FIG. S4. Best sensitivities SCCI for each possible superposition of CCI dressed states. For comparison, the best sensitivity of the 2MW dressed states S2MW is indicated by a dotted line. Φ = 0, π/2, π/4, respectively. We find that our dressed states can sense a wide range of magnetic field strengths Bz, for many values of Φ. In particular, at Φ = 0, JCCI(Bz) remains large even for small magnetic fields. This is a consequence of the fact that Ψ−1(cid:105) and Ψ+1(cid:105) are degenerate at Φ = 0, and a magnetic field lifts this degeneracy. Note that we do not include decoherence in our calculation of ρCCI. We also compare the QFI of our CCI dressed states to the 2MW dressed states, which are created using two microwave driving fields [19], and which form the most closely comparable dressed states compared to our work. The Hamiltonian describing this system is given by (43) (44) (45) H2MW = Ω 2 ( − 1(cid:105)(cid:104)0 + + 1(cid:105)(cid:104)0 + H.c.) and has eigenstates and eigenenergies given by ϕ±(cid:105) = ϕ0(cid:105) = and 1 2 1√ 2 (cid:16) − 1(cid:105) ± 20(cid:105) + + 1(cid:105)(cid:17) √ ( − 1(cid:105) − + 1(cid:105)) √ E±1 = ±Ω/ 2, E0 = 0 . (46) As for the CCI dressed states, we consider a superposition of two of the 2MW dressed states: ψinit(cid:105) = 1√ (ϕ+(cid:105) + ϕ0(cid:105)). We calculate the density matrix ρ2MW in the presence of H2MW and HB, from which we calculate J2MW(Bz; t), the QFI of the 2MW dressed states as a function of magnetic field strength Bz and evolution time t. We find that J2MW(Bz; t) is large for only a narrow range of magnetic field strengths [see Fig. S3(d)]. Defining the sensing bandwidth of the states as the FWHM of J(Bz; Tcoh), we find that our CCI dressed states can have a sensing bandwith ∼ 1000 times larger than that of the 2MW dressed states for a driving strength of Ω/2π = 500 kHz, as can be anticipated from Fig. S3. To find the sensitivity of superpositions of the CCI and 2MW dressed states, we find J(Bmin; Tcoh) as described above. The coherence times of the dressed states have been measured in Barfuss et al. as a function of Φ [17]. Using 2 -1-0.500.51Global phase ()024681012Sensitivity (nT / Hz)+1, 0+1, -1-1, 02MWcoh, we find the sensitivities Sk,(cid:96) 15 CCI(Φ), {k, (cid:96)} ∈ {−1, 0, +1}, k (cid:54)= (cid:96) and plot them in experimentally obtained values of T k,(cid:96) Fig. S4. For example, for the {+1,−1} superposition state, we find a sensitivity of ≈ 2.75 nT/ Hz at Φ = 0. In each case, the superposition states are most sensitive when the two states are degenerate and have their lowest coherence times (i. e. at Φ = −π, 0, π for {k, (cid:96)} = {−1, 0}, {+1,−1}, and {+1, 0}, respectively). We take the coherence time of the 2MW dressed states to be 5.91 µs. The published value is 18.9 µs for a drive strength of 1.6 MHz; for a drive strength of 500 kHz, as we use in our measurements in Barfuss et al., this corresponds to Tcoh = 5.91 µs, assuming √ Hz, which is indicated in Tcoh scales linearly with driving strength Ω [17, 19, 27]. We then find S2MW = 6.43 nT/ Fig. S4 for comparison with the CCI dressed state sensitivities. √ DETAILS OF THE CCD SCHEME The negatively charged NV centre, which we focus on in this work, possesses an S = 1 electronic spin ground state from which the dressed states under study emerge. This spin system is composed of the eigenstates ms(cid:105) of the spin projection operator Sz along the NV axis, with ms = 0,±1 being the corresponding spin quantum numbers [see Fig. 1(a) of the main text]. In absence of symmetry breaking fields, spin-spin interactions split the degenerate ± 1(cid:105) states from 0(cid:105) by an energy hD0 with D0 = 2.87 GHz and the Planck constant h [31]. Applying a static magnetic field Bz along the NV axis lifts the degeneracy of ± 1(cid:105) and causes a Zeeman splitting of an energy 2hγNVBz, with γNV = 2.8 MHz/G [31]. The NV spin can be readily polarised into 0(cid:105) by optical pumping with a green laser, whereas spin-dependent fluorescence allows for optical readout of the spin state [30]. While the NV spin states exhibit a hyperfine splitting due to the nuclear spin of the 14N nucleus, we restrict ourselves to the nuclear spin subspace with quantum number mI = +1 for experimental simplicity. To form our CCD scheme, we apply two MW fields and one ac strain field. While MW fields address the 0(cid:105) ↔ ±1(cid:105) transitions (supplied by a near-field antenna close to the sample), strain can coherently drive the nominally magnetic dipole-forbidden −1(cid:105) ↔ +1(cid:105) transition [see Fig. 1(a) of the main text] [27]. We realise this strain driving by placing a single NV in a mechanical resonator, which we actuate by mechanical excitation with a piezo element. By application of an appropriate external static magnetic field, the splitting of the ± 1(cid:105) states is brought into resonance with the resulting, time-varying strain field. With this, the strengths, relative phases and detunings of all driving fields can be individually controlled, allowing for full, coherent control of our three-level CCD [17]. For the ensuing state transfer protocol, we implemented arbitrary waveform control [see below] of the MW field Rabi frequencies Ω1,2(t), while the Rabi frequency of the mechanical drive Ω remained constant throughout our experiments. GENERATION OF ARBITRARY MW FIELD PULSE SHAPES In our experiment, we create the pulse envelopes of the MW field amplitudes used for driving the 0(cid:105) ↔ ± 1(cid:105) transitions through an I/Q frequency modulation technique [see Fig. S5]. A carrier signal at frequency ωc is modulated with appropriate modulation signals I and Q. Both modulation signals are composed of two individually generated pairs of (I,Q) signals at the frequencies ωIQ,1 for (I1, Q1) and ωIQ,2 for (I2, Q2), respectively. Within each (I,Q) pair, the amplitudes are constant and equal, although the amplitudes of the pair (I1, Q1) differ in general from (I2, Q2). That is, I1 and Q1 differ from each other only by a phase shift. This phase shift, which is φIQ,1 = −π/2 and φIQ,2 = π/2, allows for the suppression of one modulation sideband for each (I,Q) signal pair. After combining both modulation pairs, the resulting modulated frequency spectrum contains two frequency components, namely ω1 = ωc − ωIQ,1 and ω2 = ωc + ωIQ,2. Modifying each (I,Q) pair's amplitude with well-defined envelope functions using an AWG and a voltage multiplier then enables us to arbitrarily shape the amplitudes of both MW driving fields. In order to achieve phase-locking between the driving fields, the MW source (Standford Research Systems, SG384) and the function generators supplying the Piezo actuation and the (I,Q) signals (Keysight, 33522B) are connected to the same 10 MHz reference signal. To set the global driving phase to Φ = π/2, the output of the Piezo function generator is triggered via a software command. After receiving a trigger pulse, a subsequent trigger is forwarded to the (I,Q) signal generators to start their outputs. The (I,Q) signals' envelopes are synthesised by an arbitrary waveform generator (Tektronix, AWG 5014C), whose signals modifies the (I,Q) pairs' amplitudes via four-quadrant multiplication (Analog Devices, AD734) before both (I,Q) pairs are combined (MiniCircuits, ZFSC-2-6+). For (I,Q) modulation we use the in-built (I,Q) modulator of 16 FIG. S5. Setup used for arbitrary waveform control of the MW driving fields. The two MW driving tones with frequencies ω1 = ωc − ωIQ,1 and ω2 = ωc + ωIQ,2 are created via I/Q frequency modulation of a carrier signal with frequency ωc. Therefore, two pairs of (I,Q) signals with frequencies ωIQ,1 = 2.4 MHz and ωIQ,2 = ω3 − ωIQ,1 are combined. Each pair exhibits an appropriate phase shift of the I and Q signal to suppress one modulation sideband. Additionally, the (I,Q) pairs' amplitudes are modified by multiplication with the signal of an arbitrary waveform generator (AWG) using a four-quadrant multiplier (MP). The mechanical driving field is created by function generator actuating the Piezo element near resonant with frequency ω3. We establish phase-locking of the three driving fields to a global driving phase of Φ = π/2 by pulsed output synchronisation and locking of the MW, I/Q and Piezo function generators to the same 10 MHz reference signal. An additional MW probe field (ωp) is added to the MW driving fields for manipulation of the dressed spin states. our MW source. An additional MW tone used for manipulation (Rhode & Schwarz, SMB 100A) is added to the MW driving fields via a MW combiner (MiniCircuits, ZFRSC-42-S+). All MW pulses are controlled via digital pulses from a fast pulse generator card (SpinCore, PBESR-PRO-500), which triggers the AWG and the MW switches (MiniCircuits, ZASWA-2-50DR+). √ Creating the MW pulses in this way is limited by the vertical resolution of the four-quadrant multiplier (MP), which exhibits a noise spectral density of 1 µV/ Hz. Hence, the estimated noise amplitude within the 10 MHz bandwidth of the MP is 3.2 mVrms. The finite jumps at the beginning and the end of our driving field ramps are determined by the factor ε defined in the main text scaled with the maximum output of the AWG, given by 4.5 V. As additional noise is added within the (I,Q) modulation, choosing ε = 10−3 yields a discontinuity step that is comparable to the noise amplitude of our MW signals. Func. gen.IQ (ωIQ,1)Func. gen.Piezo (ω3)MW gen.(ωc)IQ mixerAWGMWswitchFunc. gen.IQ (ωIQ,2)MW gen.(ωp)MWswitchTTLTTLTTLTTL sync.to setupI1Q1I2Q2tI1'Q1'I2'Q2'I1' + I2'Q1' + Q2'tsoftwaretriggerωωcω3ω1ω2ωωcM1M2MPMP

1810.08024

1

1810

Time-resolved magneto-Raman study of carrier dynamics in low Landau levels of graphene

We study the relaxation dynamics of the electron system in graphene flakes under Landau quantization regime using a novel approach of time-resolved Raman scattering. The non-resonant character of the experiment allows us to analyze the field dependence of the relaxation rate. Our results clearly evidence sharp increase in the relaxation rate upon the resonance between the energy of the Landau transition and the G-band and shed new light on relaxation mechanism of the Landau-quantized electrons in graphene beyond the previously studied Auger scattering.

cond-mat.mes-hall

cond-mat

Time-resolved magneto-Raman study of carrier dynamics in low Landau levels of graphene T. Kazimierczuk,1 A. Bogucki,1 T. Smole´nski,1 M. Goryca,1 C. Faugeras,2 P. Machnikowski,3 M. Potemski,1, 2 and P. Kossacki1 1Institute of Experimental Physics, Faculty of Physics, University of Warsaw, ul. Pasteura 5, 02-093 Warsaw, Poland 2Laboratoire National des Champs Magn´etiques Intenses, CNRS-UGA-UPS-INSA-EMFL, 25 rue des Martyrs, 38042 Grenoble, France 3Institute of Physics, Wroc law University of Technology, 50-370 Wroc law, Poland (Dated: April 4, 2021) We study the relaxation dynamics of the electron system in graphene flakes under Landau quanti- zation regime using a novel approach of time-resolved Raman scattering. The non-resonant character of the experiment allows us to analyze the field dependence of the relaxation rate. Our results clearly evidence sharp increase in the relaxation rate upon the resonance between the energy of the Landau transition and the G-band and shed new light on relaxation mechanism of the Landau-quantized electrons in graphene beyond the previously studied Auger scattering. PACS numbers: 68.65.Pq, 78.30.j, 71.70.Di Despite the whole rapidly expanding field of atomi- cally thin semiconductors, graphene is still one of the most important systems with applications already be- ing introduced. Carrier dynamics is one of a relevant issues in development of devices. Although they are di- rectly related to the transport properties, optical tools are often needed to gain better insight into the carrier behavior. In particular, the ultra-fast dynamics of the carrier relaxation can be accessed using optical pump- probe techniques (for review see Ref. 1). Although such an approach has been extensively exploited to study the basic problem of relaxation dynamics in graphene at zero magnetic field, an independent case of carrier relaxation between Landau Levels (LLs) emerging upon application of the magnetic field still remains to be relatively unex- plored. In fact, there were only two time-resolved opti- cal studies dealing with the carrier dynamics in Landau- quantized graphene [2, 3]. In both of these studies the pump and probe pulses were of the same energy. The pump pulse was utilized to initially populate a certain electronic LL, while the intensity of the probe was used to measure the dynamics of subsequent depletion of this LL (see Fig. 1(a)). Such a depletion was evidenced to be due to efficient Auger scattering regardless of the num- ber of the excited LL, being either a high-energetic level (n ∼ 100) in the case of experiment exploiting near in- frared Ti:sapphire laser [2], or a low-lying level (n = 0, 1) when the graphene was excited with a THz radiation pro- duced by a free electron laser [3]. Here, we present a study of qualitatively different, slower carrier relaxation processes in Landau-quantized graphene, which take place after the system reaches its quasi-equilibrium state due to fast Auger scattering (see Fig. 1(b)). Experimentally it is realized by combining pump-probe technique with monitoring the electronic transitions between LLs using Raman scattering spec- troscopy. More specifically, a near infrared Ti:sapphire laser pulse is exploited to pump the carriers into some FIG. 1: Scheme presenting the spectrum of Landau Levels in graphene placed in external magnetic field. Color saturation denotes the relative population of LLs (a) directly after the pump pulse tuned to a transition between certain low-lying hole and electron LLs, (b) after a few hundred of fs follow- ing relatively high-energy (near infrared) laser pulse, during which the system reaches its quasi-equilibrium state due to fast Auger scattering. high-energy LL, from which they Auger-scatter occupy- ing lower LLs, the population of which is finally mea- sured based on the intensity of magneto-Raman peaks corresponding to electronic excitation between different LLs. Feasibility of such a technique was recently proven with respect to phonon transitions at zero magnetic field [4]. It has an advantage of high spatial resolution, in- accessible in previous experiment exploiting THz sources due to the difference in the wavelength scale. In par- ticular, the presented approach allows to achieve sub- micrometer resolution in a standard microphotolumines- 2D 2500 G ) 1 - 2000 m c ( t f i h s n a m a R 1500 1000 500 laser 0 0 1 2 L -2,2 L -1,2 & L -2,1 L -1,1 ) m n ( h t g n e l e v a W 970 925 884 847 813 781 5 6 4 3 7 Magnetic field (T) 8 9 10 FIG. 2: Magnetic field dependence of the Raman scattering spectrum measured at T = 200 K under excitation with CW Ti:sapphire laser at λ = 781 nm. Each spectrum was cor- rected by subtracting 80% of the zero-field spectrum in order to spotlight the field-induced changes. The white dashed lines mark two example Landau level transitions originating from graphene domain. The signal below 400 cm−1 is suppressed due to the long-pass filter placed in the detection path. cence setup. In such a regime, the intensity of various peaks in the Raman scattering spectrum provides di- rect access to carrier dynamics in a given LL even for very small graphene flakes, such as graphene domains on the surface of the natural graphite, which in turn ex- hibit much better optical properties as compared, e.g., to larger epitaxial graphene flakes [5, 6]. Additional advan- tage of this system is its inherent neutrality, which makes the results clear from effects of residual background car- riers, possibly present in the graphene system [7]. In our work we studied graphene-like domains occur- ring on the surface of the natural graphite. The identifi- cation of such domains was performed upon application of strong magnetic field, which reveals qualitative differ- ences in the LL structure between 2D graphene and 3D graphite [8, 9]. Figure 2(a) presents the magnetic field evolution of the Raman scattering spectrum measured for one out of several investigated graphene domains. 2 Apart from the well-known phonon-related resonances -- G-band ≈ 1590 cm−1 and 2D-band ≈ 2690 cm−1 -- the data features a series of field-dependent peaks corre- sponding to electronic excitations between different LLs. The optical selection rules allow transitions between nth hole Landau level and mth electron Landau level provided that n − m ≤ 1 [10]. The energy position of such tran- sitions in the spectrum is described by the square-root dependence characteristic for the Dirac dispersion[11]: E−n,m = √2eBvF (cid:16)pn + pm(cid:17) . (1) The Fermi velocity extracted from the data shown in Fig. 2 yields vF = 1.00·106 m/s, which is comparable with the results reported in previous studies of such a system [6]. In the time-resolved experiments described in following sections, we focused mainly on the strongest Landau level transition from the 1st hole level to the 1st electron level denoted as L−1,1. The pronounced difference between the graphene and graphite magneto-Raman spectra allowed us also to spa- tially map the graphene domain. The lateral exten- sion of the presented domain was found to be about 10 µm × 25 µm. The core results presented in this work were ob- tained using two-color pump-probe Raman scattering spectroscopy technique in Landau quantization regime. The magnetic field needed for such experiments was ap- plied by placing the sample inside a cryostat equipped with a superconductive magnet (B = 0−10T) oriented in Faraday geometry. An aspheric lens, mounted on a piezo- positioners directly in front of the sample, allowed us to obtain spatial resolution of about 1 µm. Such high reso- lution was important due to relatively small dimensions of the graphene domains as well as to achieve high pump laser fluence, which was needed for the time-resolved ex- periments. The probe beam used for Raman scattering spec- troscopy was either femto- or picosecond Ti:sapphire laser at λTi:Sa = 775 nm with 76 MHz repetition rate. The resulting Raman scattering signal was dispersed by a 30 cm-monochromator and recorded with a Si-based CCD camera. Dichroic filters in the excitation and the detection path were employed, respectively, to filter out the amplified spontaneous emission (ASE) and to remove the excess of the Rayleigh-scattered laser light. Simul- taneously, the sample was additionally excited with the second (pump) beam produced by optical parametrical oscillator (OPO) at λOPO = 1200 nm. The Raman sig- nal induced by the OPO laser corresponds to IR range (e.g., λ = 1.48 µm for the G-band), and thus it was not detected in the experiment. The OPO was pumped with the same Ti:sapphire laser that was used as a probe, which assured necessary syn- chronization between both laser pulse trains. The de- lay between pump and probe pulses was adjusted using a mechanical delay line. The overall temporal resolu- tion of the experiments was limited by the duration of G L-1,1 2D ) s t i n u . b r a ( l a n g i s n a m a R 1000 1500 2000 2500 Raman shift (cm -1) ) s t i n u l a n g i s n a m a R . b r a ( ) s p ( y a l e d e b o r P 0 5 10 15 20 0.1 mW 0.4 mW 1.6 mW 3.0 mW 3000 3 (c) 5 0 Probe delay (ps) 10 15 6 5 4 3 2 -5 (b) ) s t i n u . b r a ( 1 , 1 - L f o y t i s n e t n I (a) 1500 2000 2500 Raman shift (cm-1) FIG. 3: A series of Raman scattering spectra measured using a pulsed laser of different intensity. Each spectrum was nor- malized using intensity of the G-band peak as the reference. Inset presents a dependence of the normalized intensity L on the excitation power. −1 1 , laser pulses. For the femtosecond configuration, used in most of the experiments, the FWHM of the pump pulses yielded 0.21 ps, while FWHM of probe pulses was equal to 0.44 ps. The main contribution to the latter value was the effect of the band-pass filter in the excitation path. Cross-correlation measurements of the pulses from both sources revealed no appreciable jitter, which would lead to reduction of the temporal resolution. In the pi- cosecond configuration, the FWHM of the pump pulses yielded 3.7 ps, whereas the FWHM of the probe pulses was equal to 1.8 ps. The LL population was studied by analysis of the in- tensity of the Raman Ln,m peaks. Crucially, such a quantity is known to be proportional to the probabil- ity of the optical transitions between the involved lev- els, which become blocked for increasing occupancy of the LLs. As a result, the intensity of the Raman line starts to be quenched for sufficiently high carrier den- sity, which in our case was controlled by changing the power of exciting laser. The invoked behavior is illus- trated in Fig. 3, which presents a set of Raman spectra measured at B = 10 T using different intensities of the pulsed laser. Each of these spectra features two phonon peaks (G-band, 2D-band) as well as multitude of elec- tronic peaks (L−1,1, L−1,2, . . . ). The intensity of phonon- related Raman peaks scales linearly with the excitation power. The underlying reason for such behavior is the high density of phonon states and their population can be affected significantly only using much stronger pump pulses [4]. In contrast, the electronic peaks exhibit afore- mentioned saturation behavior being due to filling the relevant electron or hole Landau levels, which, in turn, limits the density of states available for further Raman scattering [9, 12]. The same phenomenon was exploited in the pump- probe experiment: the strong pump pulse was utilized FIG. 4: (a) Spectrum of the changes induced by the pump beam in the Raman spectrum at T = 200 K and B = 10 T. The color reflects a difference between the measured Raman signal and the base signal at the same energy determined for the negative delay. The reference Raman spectrum on top of the map marks the position of various peaks in the spectrum. (b) Transient of the integrated intensity of the L 1 peak. −1 , to initially populate the low Landau levels, while the in- tensity of the Raman peaks from the probe pulse was used as a measure of this population at later time. Ex- ample data measured in such an experiment are shown in Fig. 4. The power of the pump and probe beams was set to, respectively, 20 mW and 0.6 mW. The presented re- sults clearly show that directly after the pump pulse the electronic Raman signal is weaker due to reduced den- sity of states. Full spectrum of the changes [shown in Fig. 4(a)] evidences that the pump indeed affects only Ln,m peaks, while the phonon peaks (e.g., the G-band) do not exhibit noticeable variation upon arrival of the pump pulse. In agreement with previous studies [3], the characteristic time scale of the pump-induced perturba- tion is in the range of a few picoseconds. Based on rel- atively fast (sub-picosecond) rise time of the signal, we attribute the decay dynamics directly to the relaxation rate of quasi-thermalized electronic system. The value of the relaxation time was extracted from the data by fit- ting exponential-decay profile to the measured transient. Example data shown in Fig. 4(b) yields decay time of τ = (3.4± 1.0) ps. The employed experimental technique allowed us to follow the dynamics of the population of the Landau level continuously upon changes of the magnetic field, which was inaccessible in previous pump-probe ex- periments [3]. Two systematic data series are presented in Fig. 5(a) for T = 200 K and T = 10 K. As seen, the data ob- tained for L−1,1 and L−2,2 transitions at lower temper- ature overlay each other, which is consistent with our assumption that the measured decay corresponds to a re- laxation of the system remaining in a quasi-equilibrium with respect to Auger scattering. Importantly, we find ) 1 - s p ( e t a r n o i t a x a l e R 0.35 0.30 0.25 0.20 0.15 0.10 0.05 0.00 0 (a) 4 6 2 8 Magnetic field (T) L-1,1 (T = 10 K) L-2,2 (T = 10 K) L-1,1 (T = 200 K) (b) 9.4T 4.8T ) s t i n u . b r a ( y t i s n e t n i 1 , 1 - -10 0 10 20 30 Probe pulse delay (ps) L f o e g n a h C 10 FIG. 5: (a) Electron relaxation times as a function of the magnetic field. A set of square-root functions are marked by dashed lines as a guide to the eye. Two arrows indicate the resonant fields discussed in the text. (b) Pump-induced change in the signal for two magnetic fields demonstrating the difference in relaxation time. The straight lines represent the exponential-decay profiles fitted to experimental data. that the rate of such a relaxation significantly increases around B = 5 T and B = 7 T (better visible for the lower temperature). This is the first observation of the theoretically predicted [13, 14] increase in the electron relaxation rate due to the resonance between the energy of the Landau levels and the E2g phonon. In partic- ular, at B = 5 T the resonance occurs for L−2,1 and L−1,2 transitions, while at B = 7 T it is the L−1,1 tran- sition, which coincides with the energy of the invoked phonon. Surprisingly, in our data the resonance at 7 T is much broader than the one at 5 T. This finding remains in contrast with the previous theoretical predictions, ac- cording to which the resonant increase in the relaxation rate should occur rather for the non-symmetric transi- tions (e.g., −1 → 2) due to their strong mixing with the optical phonons [13, 14]. The reason for this disparity between the theory and the experiment is not clear at the moment. The second observation on top of the resonant behav- in general, the relaxation ior discussed above is that, rate systematically increases with the magnetic field and it is much faster at higher temperature. Such finding might seem to be expected as several processes related to interaction with acoustic phonons exhibit similar in- crease of the rate with magnetic field and temperature. For example in many systems spin relaxation acceler- ates due to increase of the number of phonons accessi- ble for higher relaxation energy [15]. Similarly the in- crease of the temperature results in the increase of pop- ulation of acoustic phonons and higher probability of the relaxation. However the present case of the graphene is qualitatively different. The picosecond-time-scale re- 4 laxation of the LL occupation is related to cooling of the hot-electron system, which has to be mediated by electron-lattice energy transfer. The inter-landau level transition requires energy much higher than thermal en- ergy even at moderate magnetic field (e.g., 1 → 0 tran- sition at B = 5T corresponds to 670K). Therefore the thermal population of active phonons is negligible and no significant variation should be observed for reason- able temperature range. Thus our experimental findings unequivocally show that the process cannot be explained by simple phonon-assisted relaxation between LLs and the relaxation is related to more complex processes in- volving low energy acoustic phonons and other higher energy excitations. The simplest mechanism is a two- phonon process, in which most of the energy is carried out by the optical phonon, while acoustic phonons pro- vide the required continuum. Such processes appear in the second order in the carrier-phonon coupling or via optical phonon anharmonicity [16]. The former relies on the electron-acoustic-phonon coupling. An electron on the Landau level effectively interacts only with phonons of wavelengths not greater than the magnetic length lB, which restricts the available acoustic (in-plane) phonon energy to at most (v/vF)E1, where v is the speed of sound. Since v/vF ∼ 10−2, the energy conservation limits this two-phonon process to a very narrow range of inter- LL separations around the optical phonon energy. Quan- titative estimate is obtained by treating the two-phonon relaxation as an acoustic-phonon-mediated transition be- tween L−1,1 and electronic ground state with one optical phonon, which is enabled by an optical-phonon admix- ture to the LL states. The most resonant optical phonon admixture to the L−1,1 excitation is the phonon-assisted L0,1 or L−1,0 state, since the most resonant single-phonon state on the electronic ground state (the G line in the Raman spectrum in Fig. 2) is decoupled from the L−1,1 electronic excitation, as witnessed by the lack of reso- nant anti-crossing in the spectrum (which suggests that the observed excitation is valley-symmetric [17]). With only this admixture included and using the description of carrier-phonon interaction in graphene [18, 19], the max- imum values of the relaxation rate at 200 K, in the close vicinity of the resonant magnetic field of 7.3 T, are com- parable to those found in our measurements, while at 10K the rates are a few orders of magnitude below the exper- imental values. In addition, the theoretically predicted rate for this relaxation channel falls off exponentially and decreases by many orders of magnitude already 1 T off resonance, in obvious contrast with the measurements. The anharmonicity-induced relaxation channel yields rates smoothly varying with the magnetic field, because the anharmonic decay of the zone-center optical phonon can involve acoustic phonons with arbitrary, mutually op- posite wave vectors. The resulting rate can be estimated as the product of the optical phonon decay rate and the optical phonon admixture to the LL. The former is deter- mined, both experimentally [20] and theoretically [21] to be on the order of a few ps. The effective optical-phonon- assisted coupling V to other LLs, separated by energies on the order of the G-mode energy EG, can be estimated as the typical with of a resonant anticrossing between the G line and the LL excitations, which is on the or- der of a few meV. The admixture is then on the order of the Huang-Rhys factor (V /EG)2 ∼ 10−4, yielding the anharmonicity-induced inter-LL relaxation in graphene ineffective. Quantitative calculations indeed yield relax- ation times on the order of 100 ns at magnetic fields around 7 T. We therefore conclude that the decay of the LL occu- pation has to be attributed to the overall cooling of the hot-electron system via a more complex process, which might explain, in particular, the thermal dependence of the rate, which is compatible neither with the energies of the effectively cupled acoustic phonons nor with the inter-LL separation, nor with the optical phonon energy. As revealed by the data in Fig. 5(a), the decay rate scales roughly as ∝ √B in the low temperature regime. Such a magnetic-field-dependence was theoretically pre- dicted for broadening of LLs due to impurity induced dephasing [22], which, in principle, could be responsi- ble for the observed increase of the relaxation rate as long as it is accompanied by phonon scattering, since the impurity dephasing alone is not expected to exhibit any temperature dependence [22]. Another feature that is difficult to explain in terms of simple relaxation pro- cesses is the pronounced change in the character of the magnetic-field dependence of the relaxation rate upon in- creasing the temperature, which is found to be almost lin- ear at T = 200 K. The detailed discussion of the physical reason for the observed features would require more in- 5 depth knowledge about the nature of the involved relax- ation processes, which demands further theoretical stud- ies. Possible mechanisms may involve relaxation between higher LLs, or combined phonon-Auger processes involv- ing those levels, accompanied by very fast redistribution of occupations between the levels, consistent with the fs- time-scale rise of the Raman signal. To conclude, our results demonstrate the feasibility of the time-resolved Raman scattering technique in stud- ies of the Landau-quantized electrons. The non-resonant character of the Raman experiment enabled us to vary the magnetic field continuously, which is distinct advan- tage over previous approaches [3]. The results of our study qualitatively confirm predictions regarding reso- nant increase in the electronic relaxation rate due to res- onances with optical phonons. However, these results also reveal some deficiency of the existing theoretical de- scription of the carrier dynamics in graphene at strong magnetic field. Further theoretical studies are needed to determine whether the detected discrepancy is related to the non-resonant character of our experiment or perhaps it is an indication of inadequacy of assumptions made in the existing models. Acknowledgments This work was supported by the Polish Na- tional Science Centre as research grant no DEC- 2013/10/M/ST3/00791 and by the ATOMOPTO project carried out within the TEAM programme of the Foun- dation for Polish Science co-financed by the European Union under the European Regional Development Fund. [1] E. Malic and A. Knorr, Graphene and Carbon Nanotubes: Ultrafast Optics and Relaxation Dynamics (Wiley-VCH, 2013), ISBN 978-3-527-41161-0. [2] P. Plochocka, P. Kossacki, A. Golnik, T. Kazimierczuk, C. Berger, W. A. de Heer, and M. Potemski, Phys. Rev. B 80, 245415 (2009). [3] M. Mittendorff, F. Wendler, E. Malic, A. Knorr, M. Or- lita, M. Potemski, C. Berger, W. A. de Heer, H. Schnei- der, M. Helm, et al., Nat. Phys. 11, 75 (2015). [11] K. S. Novoselov, A. K. Geim, S. V. Morozov, D. Jiang, Y. Zhang, S. V. Dubonos, I. V. Grigorieva, and A. A. Firsov, Science 306, 666 (2004). [12] P. Kossacki, C. Faugeras, M. Kuhne, M. Orlita, A. Mah- mood, E. Dujardin, R. R. Nair, A. K. Geim, and M. Potemski, Phys. Rev. B 86, 205431 (2012). [13] Z.-W. Wang, L. Liu, L. Shi, X.-J. Gong, W.-P. Li, and K. Xu, J. Phys. Soc. Jpn. 82, 094606 (2013). [14] F. Wendler, A. Knorr, and E. Malic, Appl. Phys. Lett. [4] J.-A. Yang, S. Parham, D. Dessau, and D. Reznik, S. 103, 253117 (2013). Rep. 7, 40876 (2016). [5] P. Neugebauer, M. Orlita, C. Faugeras, A.-L. Barra, and M. Potemski, Phys. Rev. Lett. 103, 136403 (2009). [6] C. Faugeras, M. Amado, P. Kossacki, M. Orlita, M. Kuhne, A. A. L. Nicolet, Y. I. Latyshev, and M. Potemski, Phys. Rev. Lett. 107, 036807 (2011). [7] D. Sun, C. Divin, C. Berger, W. A. de Heer, P. N. First, and T. B. Norris, Phys. Rev. Lett. 104, 136802 (2010). [8] M. L. Sadowski, G. Martinez, M. Potemski, C. Berger, and W. A. de Heer, Phys. Rev. Lett. 97, 266405 (2006). [9] C. Faugeras, M. Orlita, and M. Potemski, J. Raman [15] K. J. Standley, Electron Spin Relaxation Phenomena in Solids, Monographs on Electron Spin Resonance (Springer, 1969). [16] L. Jacak, P. Machnikowski, J. Krasnyj, and P. Zoller, Eur. Phys. J. D 22, 319 (2003). [17] M. O. Goerbig, J.-N. Fuchs, K. Kechedzhi, and V. I. Fal'ko, Phys. Rev. Lett. 99, 087402 (2007). [18] H. Suzuura and T. Ando, J. Phys. Soc. Japan 77, 044703 (2008). [19] H. Suzuura and T. Ando, J. Phys. Conf. Ser. 150, 022080 (2009). Spectrosc. 49, 146 (2018). [20] N. Bonini, M. Lazzeri, N. Marzari, and F. Mauri, Phys. [10] O. Kashuba and V. I. Falko, Phys. Rev. B 80, 241404 Rev. Lett. 99, 176802 (2007). (2009). [21] M. Kuhne, C. Faugeras, P. Kossacki, A. A. L. Nicolet, M. Orlita, Y. I. Latyshev, and M. Potemski, Phys. Rev. B 85, 195406 (2012). [22] F. Wendler, A. Knorr, and E. Malic, Nanophotonics 4, 224 (2015). 6 

1108.5337

1

1108

Collective spontaneous emission from pairs of quantum dots: long-range vs. short-range couplings

We study the spontaneous emission from a coherently delocalized exciton state in a double quantum dot as a function of the distance between the dots, focusing on the similarities and differences between the cases of radiative (long-range, dipole) and tunnel coupling between the excitons in the dots. We show that there may be no qualitative difference between the collective emission induced by these two coupling types in spite of their essentially different physical properties.

cond-mat.mes-hall

cond-mat

Collective spontaneous emission from pairs of quantum dots: long-range vs. short-range couplings Wildan Abdussalam and Pawe l Machnikowski∗ Institute of Physics, Wroc law University of Technology, 50-370 Wroc law, Poland We study the spontaneous emission from a coherently delocalized exciton state in a double quan- tum dot as a function of the distance between the dots, focusing on the similarities and differences between the cases of radiative (long-range, dipole) and tunnel coupling between the excitons in the dots. We show that there may be no qualitative difference between the collective emission induced by these two coupling types in spite of their essentially different physical properties. PACS numbers: 78.67.Hc, 42.50.Ct, 03.65.Yz I. INTRODUCTION Excitons delocalized in closely spaced quantum dots (QDs) recombine in a different way than in a single QD [1, 2]. This effect is at least partly due to collective in- teraction of the two emitters with the quantum radiation field [3]. For non-interacting dots this collective effect appears only if the interband transition energies in the two QDs differ by no more than the emission line width, which requires the dots to be nearly identical, beyond the current technological possibilities. However, coupling be- tween the dots restores the collective nature of the emis- sion and leads to accelerated or slowed down emission even for dots with different transitions energies, which is manifested in the optical response from these systems [4, 5]. The two major couplings that may appear in a sys- tem of QDs are due to Coulomb interactions and carrier tunneling. The former results from the coupling between the interband dipole moments associated with the exci- tons in the dots (sometimes referred to as the dispersion force) [6]. It has a long-range nature, with the typical 1/R3 behavior at short distances (actually, this singular- ity is removed for charges distributed in a finite volume [7]) and an oscillating tail with an envelope decaying as 1/R at distances larger than the resonant wave length. The tunnel coupling is a short-range interaction, which vanishes exponentially at distances on the order of a few nanometers. In this contribution, we study the spontaneous emis- sion from an exciton confined in a double quantum dot, focusing on the similarities and differences between the cases of radiative (long-range, dipole) and tunnel cou- pling between the excitons in the dots. We show that for strictly identical dots the oscillating nature of the dipole coupling on long distances leads to non-monotonic dependence of the radiative decay rate on the inter-dot separation. However, for a double dot system with a realistic, technologically feasible mismatch of transition energies, the collective effects disappear completely well ∗Electronic address: Pawel.Machnikowski@pwr.wroc.pl before these oscillations become relevant. In both cases, there is no qualitative difference between the emission in- duced by the long-range dipole interaction and that due to short-range tunnel couplings with appropriately cho- sen (but realistic) parameters, in spite of their essentially different physical origin and properties. The paper is organized as follows. In Sec. II, we de- scribe the model of the system. Next, in Sec. III, we present and discuss the results of numerical simulations. Finally, Sec. IV concludes the paper. II. MODEL We consider two QDs placed in the xy plane and shifted by a vector r12. Each QD is modeled as a two- level system (empty dot and one exciton). The Hilbert space of the double-dot system in our model is then spanned by the empty dot state 00i, the two single ex- citon states 10i,01i corresponding to the exciton in the first and second dot, respectively, and the "molecular biexciton" state 11i. The transition energies for the in- terband transitions in the two dots are E1 = E + ǫ, E2 = E − ǫ. The dots are coupled by an interaction V which can be either of dipole -- dipole character (long-range dispersion force) or result from carrier tunneling (short-range, expo- nentially decaying interaction). We introduce the transi- tion ("exciton annihilation") operators for the two dots, σ1 = 00ih10 + 01ih11, σ2 = 00ih01 + 10ih11 and the exciton number operators nα = σ† ασα, α = 1, 2. Using these operators, the Hamiltonian of the double-dot sys- tem is written in the frame rotating with the frequency E/ in the form H0 = ǫ (n1 − n2) + V (cid:16)σ† 1σ2 + σ† 2σ1(cid:17) + EB n1 n2, where last term represents the biexciton shift. The long-range dipole coupling is described by V = Vlr = −Γ0G(k0r12), where Γ0 = d02k3 0 3πε0εr , is the spontaneous emission (radiative recombination) rate for a single dot, ε0 is the vacuum permittivity, εr is the relative dielectric constant of the semiconductor, and k0 = nE c , where c is the speed of light and n = √εr is the refractive index of the semiconductor, and ) 1 - s p ( h- / V 2 1.5 1 0.5 0 Vlr/h- Vsr/h- A (a) 2 Vlr/h- G 12 Vsr/h- C D E B (b) 2 ) 1 - s n ( 1 2 1 G , h- / V 0 0 5 10 15 20 0 200 400 600 800 r12 (nm) r12 (nm) G(x) = 3 4 h−(cid:16)1 − d · r122(cid:17) cos x +(cid:16)1 − 3 d · r122(cid:17)(cid:18) sin x x2 + x cos x x3 (cid:19)(cid:21) , FIG. 1: The interference term of the decay rate Γ12 and the short- and long-range coupling amplitudes Vlr, Vsr as a func- tion of the inter-dot distance. In (a), the small distance sec- tion is shown, while in (b) the oscillating tail at larger dis- tances is visible. Note the different scales in (a) and (b). where r12 = r12/r12 and d = d/d, where d is the in- terband matrix element of the dipole moment operator which is assumed identical for both dots. For a heavy hole exciton, d = (d0/√2)[1,±i, 0]T , so that for a vector r12 in the xy plane one has d · r122 = 1/2. The tunnel coupling is described by V = Vsr = V0e−r12/r0 . The effect of the coupling to the radiation field is ac- counted for by including the dissipative term in the evolu- tion equations, which describes radiative recombination of excitons. The equation of evolution of the density ma- trix is then given by [8] i  ρ = − [H0, ρ] + 2 Xα,β=1 Γαβ (cid:20)σαρσ† β − 1 2 (cid:8)σ† ασβ, ρ(cid:9)+(cid:21) , (1) where Γ11 = Γ22 = Γ0, Γ12 = Γ21 = Γ0F (k0rαβ), with F (x) = 3 2 (cid:20)(cid:16)1 − d · r122(cid:17) sin x +(cid:16)1 − 3 d · r122(cid:17)(cid:18) cos x x2 − x sin x x3 (cid:19)(cid:21) , and {. . . , . . .}+ denotes the anticommutator. The diago- nal decay rates Γαα describe the emission properties from a single dot, while the off-diagonal terms Γαβ, α 6= β, ac- count for the interference of emission amplitudes result- ing from the interaction with a common reservoir and are responsible for the collective effects in the emission. In our simulations, we use the parameters for a typical InAs/GaAs QD system: Γ0 = 1 ns−1, n = 3.3, E = 1.3 eV. For the tunnel coupling we choose the amplitude V0 = 2.19 meV and the range r0 = 2.03 nm, which makes the values for the tunnel and dipole couplings similar for inter-dot distances around 6 nm. The values of the two couplings as well as the interfer- ence term of the decay rate Γ12 are plotted as a function of the distance between the dots in Fig. 1. In this figure we mark the distance values for which the decay will be discussed in the next section. n o i t a p u c c o n o t i c x E n o i t a p u c c o n o t i c x E 1 0.8 0.6 0.4 0.2 0 1 0.8 0.6 0.4 0.2 0 identical dots, LR coupling identical dots, SR coupling A B C D E A B C D E (a) (b) non-identical dots, LR coupling non-identical dots, SR coupling A B C D E A B C D E (c) 0 1 2 t (ns) (d) 0 3 1 2 t (ns) 3 FIG. 2: The exciton occupation (the average number of ex- citons in the system) for an initial single-exciton state corre- sponding to a coherently delocalized superposition. (a) and (b) show the evolution for a pair of identical dots coupled by long-range dipole forces and by short-range tunnel coupling, respectively. (c) and (d) show the evolution for a pair of non- identical dots, for the two kinds of couplings as previously. The labels A,...,E refer to the values of the inter-dot distance marked in Fig. 1. III. RESULTS In Fig. 2 we show the results of the numerical simula- tions based on Eq. (1). On each plot, the average number of excitons in the system is shown as a function of time for identical dots (ǫ = 0) and for slightly non-identical dots with ǫ = 0.01 meV. The initial state in all the cases is chosen to be (01i + 10i)/√2. We study the decay of exciton population for various distances between the dots and compare the evolution for the two kinds of couplings. For identical dots, the exciton decay time for the delo- calized initial state strongly depends on the distance be- tween the dots. This is due to the oscillations and decay of the interference term Γ12. For dots placed at a short distance (case A), Γ12 ∼ Γ0 and the decay has a strongly collective character, which manifested by the faster emis- sion visible in Figs. 2(a,b) [3]. The collective effect gets weaker as the distance between the dots grows and Γ12 decreases (B). For some values of the distance, Γ12 < 0 (C). Then, the amplitudes for photon emission from the two dots interfere destructively and the decay gets slower than the usual exponential decay with the rate Γ0 (the initial state becomes subradiant). Whenever Γ12 = 0, the decay rate is the same as for an individual dot (D). Comparison of Figs. 2(a) and (b) shows that for identi- cal dots, these effects do not depend on the coupling and are therefore the same, irrespective of the presence and physical nature of the interaction between the dots. For dots that differ by the relatively small transition energy mismatch of 2ǫ = 0.02 meV, almost all this non- monotonic dependence of the emission rate on the dis- tance disappears. The reason is that the oscillations of the interference term take place in the distance range where the coupling between the dots is very weak and is dominated already by a small energy mismatch assumed here, which destroys collectivity of the emission process [3]. The only exception is the smallest distance shown in this plot, where the coupling is sufficiently strong. By comparing Figs. 2(c) and 2(d) one can see that also in this case, the evolution of the exciton occupation is nearly the same for both systems. Here the tunnel coupling param- eters have been deliberately chosen to assure the same coupling strength around the 6 nm distance. At larger distances both couplings are negligible compared to the energy mismatch. IV. CONCLUSIONS We have shown that the radiative decay of exciton oc- cupation in a pair of coupled quantum dots depends on 3 the distance between the dots as a result of the spatial dependence of the interference term governing the in- teraction with the quantum electromagnetic field. For non-identical dots, the emission rate depends on the in- terplay of the energy mismatch between the dots and the coupling between them. Although the two kinds of couplings that are present in the system (tunneling and dipole interaction) have essentially different physical na- ture and properties, they may lead to the same dynamics of the observed collective emission. We believe that these findings may shed some light on the interpretation of the experiment [9] in which en- hanced emission was observed in a quantum dot ensem- ble in which the dipole coupling energies on the typical inter-dot distances were much smaller than the average transition energy mismatch between the dots. Indeed, as we have shown, the tunnel coupling leads to the same effect as the long-range dipole interaction but it can be stronger than the latter at short distances. Hence, it seems very likely that short-range tunnel coupling be- tween some pairs of dots can be responsible for the ob- served collective emission effect in QD ensembles. Acknowledgment: This work was supported by the Foundation for Polish Science under the TEAM pro- gramme, co-financed by the European Regional Devel- opment Fund. [1] C. Bardot, M. Schwab, M. Bayer, S. Fafard, Z. Wasilewski, (2009). and P. Hawrylak, Phys. Rev. B 72, 035314 (2005). [2] P. Borri, W. Langbein, U. Woggon, M. Schwab, M. Bayer, S. Fafard, Z. Wasilewski, and P. Hawrylak, Phys. Rev. Lett. 91, 267401 (2003). [3] A. Sitek and P. Machnikowski, Phys. Rev. B 75, 035328 (2007). [4] A. Sitek and P. Machnikowski, Phys. Rev. B 80, 115319 [6] M. J. Stephen, J. Chem. Phys. 40, 669 (1964). [7] P. Machnikowski and E. Rozbcki, Phys. Stat. Sol. (b) 246, 320 (2009). [8] R. H. Lehmberg, Phys. Rev. A 2, 883 (1970). [9] M. Scheibner, T. Schmidt, L. Worschech, A. Forchel, G. Bacher, T. Passow, and D. Hommel, Nature Physics 3, 106 (2007). (2009). [5] A. Sitek and P. Machnikowski, Phys. Rev. B 80, 115301 

1012.4575

1

1012

Charged excitons and biexcitons in laterally coupled InGaAs quantum dots

We present results of atomistic empirical pseudopotential calculations and configuration interaction for excitons, positive and negative trions (X\pm), positive and negative quartons (X2\pm) and biexcitons. The structure investigated are laterally aligned InGaAs quantum dot molecules embedded in GaAs under a lateral electric field. The rather simple energetic of excitons becomes more complex in the case of charged quasiparticles but remains tractable. The negative trion spectrum shows four anticrossings in the presently available range of fields while the positive trion shows two. The magnitude of the anticrossings reveals many-body effects in the carrier tunneling process that should be experimentally accessible.

cond-mat.mes-hall

cond-mat

Charged excitons and biexcitons in laterally coupled InGaAs quantum dots Jie Peng and Gabriel Bester∗ Max-Planck-Institut fur Festkorperforschung, Heisenbergstr. 1, D-70569 Stuttgart, Germany (Dated: December 9, 2010) We present results of atomistic empirical pseudopotential calculations and configuration interaction for excitons, positive and negative trions (X±), positive and negative quartons (X2±) and biexcitons. The structure investigated are laterally aligned InGaAs quantum dot molecules embedded in GaAs under a lateral electric field. The rather simple energetic of excitons becomes more complex in the case of charged quasiparticles but remains tractable. The negative trion spectrum shows four anticrossings in the presently available range of fields while the positive trion shows two. The magnitude of the anticrossings reveals many-body effects in the carrier tunneling process that should be experimentally accessible. PACS numbers: 73.21.La,78.67.Hc,78.55.Cr,71.35.-y I. INTRODUCTION Epitaxial semiconductor quantum dots (QDs) can be designed with an increasing degree of control over their sizes, shapes and alloy compositions1,2. Individual structures can be arranged laterally and vertically using controlled self-assembly or substrate patterning, to generate complex arrangements of QDs. This newly gained ability to deterministically produce architectures of QDs is the prerequisite for the implementation of scalable quantum networks, where each QD holds a quantum bit (qubit). Furthermore, impressive advances were made in the manipulation of the realized quantum states; enabled by a fundamental understanding of the QDʼs electronic and optical properties. For single QDs, different types of quantum states were already manipulated. Early work focussed on the coherent manipulation of exciton qubits3–8, but due to their rather short coherence times, the attention has shifted to the electron spin. The initial step of the single spin initialization was demonstrated9–11 and is now part of more advanced spin manipulation experiments12–16. For structures with more than one QD, the most advanced prototypes are vertically aligned (stacked) quantum dot molecules (QDMs). These structures have appeared more than a decade ago17–19 but have developed into a fertile platform for quantum manipulations only recently. One important step was the fabrication of structures designed to allowing either electron or hole tunneling20. The observation of the coupling through anticrossings under vertical electric field21–24 along with theoretical atomistic modeling25–27 allowed for precise estimates of many relevant coupling energies. Indeed, the understanding of the quantum states in vertically coupled QDMs is rather deep and goes beyond coupling energies. For instance, the existence of an antibonding ground state28,29 could be demonstrated, the influence of a lateral misalignment of the QDs was investigated30 and indirect excitons could be recently observed31. Success in the area of control and manipulation was very recently reported for vertically stacked QDMs. Namely, the optical spin initialization over the fine-structure of the excited trion state11 and the ultrafast control of the entanglement between two electrons spins located in different dots32. This represents the first two-qubit operation with QDs. The history of laterally aligned QDMs is younger, with high quality structures appearing only in the last few year1,2,33–36. Laterally coupled QDMs are certainly good candidates for applications in quantum information science because of the potential to couple several QDs (“scaling”) to form the first building block of a useful device. It is also believed that the degree of external control of individual QDs within a QDM or an array of laterally aligned QDs should be larger than in vertical structures. However, the control until now has been limited due to the difficulty to apply lateral electric fields in order to gate (or tune) the device. Indeed, the geometrical constraints have lead to relatively weak37,38 applied lateral fields until now. Likewise, the charging of the QDM with extra carriers, as achieved regularly in vertical structures through tunneling from a δ-doping layer, has not yet been achieved in lateral structures. These limitations are not believed to be of fundamental nature but certainly represent technical challenges. 

1302.6356

1

1302

CdSe-Au nanorod networks welded by gold domains - a promising structure for nano-optoelectronic components

CdSe-Au networks were synthesized by a colloidal chemistry technique. They entail CdSe nanorods with a diameter of ~10 nm and a length of ~40 nm, which are joined together by Au domains (~5 nm). Individual networks were positioned by AC dielectrophoresis between bow-tie electrodes with a gap of ~100 nm and their conductivity as well as the photoelectrical properties were investigated. Nanorod networks, with multiple Au domains on the nanorod surface, displayed stable conductivity that was not sensitive to blue laser light illumination. Such nanostructures were transformed by thermal annealing to networks with Au domains only at the nanorod tips. In this system the overall conductivity was reduced, but a clear photocurrent signal could be detected, manifesting semiconductor behavior.

cond-mat.mes-hall

cond-mat

CdSe-Au nanorod networks welded by gold domains - a promising structure for nano- optoelectronic components Peigang Li, Alexandros Lappas* Institute of Electronic Structure & Laser, Foundation for Research and Technology-Hellas, Vassilika Vouton, 71110 Heraklion, Greece Romain Lavieville, Yang Zhang, Roman Krahne+ Nano fabrication unit, Instituto Italiano di Technologia, via Morego 30, 16163 Genoa, Italy Abstract: CdSe-Au networks were synthesized by a colloidal chemistry technique. They entail CdSe nanorods with a diameter of ~10 nm and a length of ~40 nm, which are joined together by Au domains (~5 nm). Individual networks were positioned by AC dielectrophoresis between bow-tie electrodes with a gap of ~100 nm and their conductivity as well as the photoelectrical properties was investigated. Nanorod networks, with multiple Au domains on the nanorod surface, displayed stable conductivity that was not sensitive to blue laser light illumination. Such nanostructures were transformed by thermal annealing to networks with Au domains only at the nanorod tips. In this system the overall conductivity was reduced, but a clear photocurrent signal could be detected, manifesting semiconductor behavior. Keywords: CdSe nanorods networks·nanocrystals·dielectrophoresis·nanofabrication ·electrical transport properties·photoconductivity·annealing Introduction In the last two decades the electrical properties of single, shape-controlled semiconductor colloidal nanocrystals have attracted much attention, especially regarding their potential applications in nanoelectronics and nanophotonics (Klein et al. 1997; Cui et al. 2005; Trudeau et al. 2008; Sheldon et al. 2009; Teich-McGoldrick et al. 2009; *email: lappas@iesl.forth.gr +email: Roman.Krahne@iit.it 1 Steinberg et al. 2009; Steinberg et al. 2010). For example, single-electron transistor functionality has been demonstrated based on CdSe nanoparticles and CdTe tetrapods (Klein et al. 1997; Cui et al. 2005). More recently, CdSe nanorods (NRs) have been studied for their applicability in transistor structures and for electrical current switching (Trudeau et al. 2008; Steinberg et al. 2010), demonstrating their usefulness in the development and building of nanoscale components in modern electronics. Regarding small-scale ensembles of nanocrystals, the electrical transport is hindered by the high tunnel barriers imposed by the surfactant molecules (typically phosphoric or oleic acids) that are needed to passivate the nanocrystal surface (Fig. 1a-b). Recently, Figuerola et al. have successfully assembled Au-tipped CdSe NRs end-to-end via a nanowelding approach mediated by Au domains into 1D, 2D or 3D networks (Fig. 1c)( Figuerola et al. 2008). Such all-inorganic networks, in which the NRs are linked by metal domains, promise to be an interesting alternative to overcome the problem of the high tunnel barriers (Fig. 1d). Concerning the photoelectrical properties of NRs, so far mostly systems consisting of a very large number of nanoparticles have been investigated, like in thin films deposited on micrometer spaced electrodes (Steiner et al. 2009; Persano et al. 2004). Little has been done towards the photoconductive properties of single semiconductor nanoparticles or of small-scale ensembles of those. This is due to the difficulties in fabricating electrical contacts, and to the small optical cross section of few nanoparticles that makes detectable generation of photo-induced current hard. In this respect submicron NR networks represent an interesting intermediate system. On one hand the conductive path across many NRs is facilitated by the all-inorganic interfaces within the network. On the other hand such networks provide a sufficient optical cross section that could make the observation of photocurrent possible. In this communication, we report the implementation of CdSe-Au networks, synthesized according to the procedure published by Figuerola et al. (Figuerola et al. 2009), into bow-tie electrode devices, and the study of their conductive and photo- electrical properties. The networks were assembled in between electrodes with 100 nm gap size by AC dielectrophoresis. Scanning electron microscopy inspection 2 confirmed the trapping of single network structures in our devices with a success rate of ～ 85%. Electrical characterization showed stable and reproducible conductive behavior for the networks, and photo-induced conductivity in the case of networks that have been annealed. Experimental The starting CdSe NRs were prepared by colloidal chemistry via the seeded growth approach (Carbone et al. 2007). The gold domains were then selectively grown on to the tips of CdSe NRs to form a dumbbell structure (Mokari et al. 2004). In order to link the nanorods into a network, a suitable quantity of iodine (I2) solution was added to the nanodumbbell solution (Figuerola et al. 2009). Samples of the network solution deposited on carbon-coated grids were taken for transmission electron microscopy (TEM) inspection (JEM 1011, JEOL Company, Japan). Electron beam lithography (EBL) in combination with metal deposition of 5 nm Ti and 50 nm Au, followed by a standard lift-off process, was used to fabricate the bow- tie electrodes on Si substrate with a 100 nm thick oxidized layer on the surface. A typical device chip consisted of 10 pairs of electrodes with an average gap size of 100 nm. For the network positioning via AC dielectrophoresis we used a cavity with dimensions of 1 cm × 1 cm × 0.3 cm (L×W×H) that was manufactured into a Teflon block. Prior to the trapping, the chip was cleaned successively first with acetone, then with methanol and finally it was placed in ethanol for 12 hours to remove organic contaminants from the surface. Before the trapping experiments, the chip was removed from the ethanol bath, immersed into toluene and dried with nitrogen flow. After this cleaning procedure the device was transferred into the chemical cell that was filled with a toluene solution containing the networks in low concentration (NR concentration around 10-9 M). A Suess probe station was employed to contact the two electrodes and a lock-in amplifier (SR7265, AMETEK Inc. USA) provided the AC voltage. The parameters of the trapping process (voltage amplitude, frequency, and trapping time) were computer controlled via a LABVIEW-based 3 software interface. Fig. 2 shows a schematic drawing of the trapping mechanism. Before trapping, the gaps were checked to ensure that no electrical shorts or leakage currents were detectable. The conductive properties of the trapped nanorod networks were measured with a Keithley 2612 source-meter in a two-probe configuration. All I-V curves were taken in air and at room temperature. For the photoelectrical properties, a diode laser emitting at 473 nm was used to irradiate the device with a power of 10 mW on a spot size with approximately 5 mm diameter. After the first set of measurements, the chip was annealed at 250 °C for 15 minutes under N2 atmosphere, and the electrical properties were recorded again. All the scanning electron microscopy (SEM) imaging for this work was carried out with the Raith 150/2 system, which is a field-emission instrument with fairly high resolution. Results and discussion A typical TEM image of the as-made networks is reported in Fig. 3, clearly showing chain-like nanostructures. In order to control the quantity of trapped networks, we regulated the voltage amplitude, frequency, and trapping time, as they proved to be the crucial parameters in that respect. Best results were obtained with a frequency of 1 kHz, both significantly higher and lower frequency values resulted in less efficient trapping. Increased trapping time resulted in a higher density of networks in-between the electrodes gap area. The most sensitive parameter was the voltage amplitude. Below a voltage amplitude of 3 VRMS, no trapping of networks was observed. Above this threshold, the quantity of trapped networks increased with increasing voltage, as demonstrated in Fig. 4 a-b. Voltage amplitudes exceeding 8 VRMS led to damage of the electrodes. Our best set of parameters for trapping the CdSe-Au networks was a frequency of 1 kHz, a voltage amplitude of 3 VRMS and a trapping time of 60 s. With these parameters we obtained ～85% of our 30 pairs of electrode linked via small-scale networks. A representative result is shown in Fig. 4b, where the networks bridged the electrode gap and formed a conductive electrical pathway. 4 We note that the approach of electrode fabrication followed by dielectrophoresis for nanoparticle positioning has a number of advantages over other methods where the nanoparticles are deposited before electrode fabrication and then selectively contacted by an overlayer EBL process. In the dielectrophoresis approach, (i) the lithography process is much easier because it can be done on a clean wafer surface and there is no need for alignment (like when pre-deposited nanoparticles are in use), therefore resulting in a much higher throughput. Then, (ii) the nanoparticles do not suffer a hard post-deposition treatment, like PMMA coating, exposure to acetone etc., like it is the case in overlayer EBL processing. Typical I-V characteristics of trapped networks prior to annealing are plotted in Fig. 5, showing reproducible and stable non-linear conductivity, but no photo-induced current. The I-V curves measured before annealing can be well described by a simple exponential model, where I /V I~ exp( V 0 0 ) (see Fig. 5), yielding a characteristic voltage of V0 = 0.76 V. After annealing the conductivity of the networks was significantly reduced, as shown in Fig. 6, but interestingly illumination of the device with blue laser light resulted in photo-induced current. To understand the effect of the thermal annealing on the CdSe-Au network structure in more detail, we have applied a similar annealing process to networks that were deposited onto carbon-coated TEM grids. TEM images before and after thermal annealing are shown in Fig. 7, where we clearly observe not only Au domains that link the nanorod tips, but also many additional Au domains that decorated the lateral facets of the nanorods. The conductance of such networks was most likely mediated by thermally activated carriers and hopping charge transport in between localized state. Consequently, transport appears not to be sensitive to the semiconductor material and light irradiation. After annealing (Fig. 7b) the Au had migrated to the nanorod tips by ripening process as reported by Figuerola et al. (Figuerola et al. 2010), which is clearly evidenced by the larger Au domains that link the nanorod tips. Current transport in networks as in Fig. 7b should be mediated by chains of a sequence of Au domain/CdSe nanorod/Au domain and so on. 5 In this case the semiconductor nanorod material has a clear impact on the conductivity, which should lead to photo-induced current as observed in Fig. 6. Furthermore, the dark current in such a network could be limited by a series of Schottky barriers which should reduce significantly the conductivity in comparison to the hopping transport between the metal domains (Fig. 5) ( Subannajui et al. 2008), in good agreement with our experimental observations. Conclusion We have demonstrated that CdSe-Au networks fabricated by colloidal synthesis can be controllably assembled in between bow-tie electrodes with ~100 nm separation by using a dielectrophoresis process. We measured the conductance of small-scale networks composed of CdSe nanorods linked via Au domains. Thermal annealing of our devices resulted in Au migration to the nanorod tips. This in turn led to a conductive path composed of quasi-linear chains of semiconductor nanorods and Au domains, showing photo-induced current as compared to the very low conductivity in the dark. Acknowledgement This research was supported by the European Commission through the Marie-Curie ToK project NANOTAIL (MTKD-CT-2006-042459). References Carbone L, Nobile C, De Giorgi M, Sala FD, Morello G, Pompa P, Hytch M, Snoeck E, Fiore A, Franchini IR, Nadasan M, Silvestre AF, Chiodo L, Kudera S, Cingolani R, Krahne R and Manna L (2007) Synthesis and micrometer-scale assembly of colloidal CdSe/CdS nanorods prepared by a seeded growth approach. NanoLett. 7: 2942-2950 Cui Y, Banin U, Bjoerk MT and Alivisatos AP (2005) Electrical transport through a single nanoscale semiconductor branch point. Nano Lett. 5: 1519-1523 Figuerola A, Franchini IR, Fiore A, Mastria R, Falqui A, Bertoni G, Bals S, Tendeloo GV, Kudera S, Cingolani R, and Manna L (2009) End-to-End Assembly of Shape-Controlled Nanocrystals via a Nano-welding Approach Mediated by Gold Domains. Advanced Materials 21:550-554 6 Figuerola A, Huis MV, Zanella M, Genovese A, Marras S, Falqui A, Zandbergen HW, Cingolani R and Manna L (2010) Epitaxial CdSe-Au Nanocrystal Heterostructures by Thermal Annealing. NanoLett.10: 3028-3036. Klein DL, Roth R, Lin AKL, Alivisatos AP and McEuen PL (1997) A single-electron transistor made from a cadmium selenide nanocrystal. Nature 389:699-711 Mokari T, Rothenberg E, Popov I, Costi R, Banin U (2004) Selective Growth of Metal Tips onto Semiconductor Rods and Tetrapods. Science 304:1787-1790 Persano A, De Giorgi M, Fiore A, Cingolani R, Manna L, Cola A and Krahne R, (2004) Photoconduction Properties in Aligned Assemblies of Colloidal CdSe/CdS Nanorods. Nano Lett. 4: 1646-1652 Sheldon MT, Trudeau PE, Mokari T, Wang LW and Alivisatos AP (2009) Enhanced semiconductor nanocrystal conductance via solution grown contacts. Nanolett. 9: 3676-3682 Steiner D, Azulay D, Aharoni A, Salant A, Banin U, and Millo O (2009) Photoconductivity in Aligned CdSe Nanorod Arrays. Phys. Rev. B 80: 195308 Steinberg H, Lilach Y, Salant A, Wolf O, Faust A, Millo O and Banin U, (2009) Anomalous Temperature Dependent Transport through Single Colloidal Nanorods Strongly Coupled to Metallic Leads. Nano Lett. 9:3671-3675 Steinberg H, Wolf O, Faust A, Salant A, Lilach Y, Millo O and Banin U (2010) Electrical Current Switching in Single CdSe Nanorods. Nano Lett. 9:2416-2420 Subannajui K, Kim DS, and Zacharias M (2008) In-situ electrical analysis of individual ZnO nanowires. J. Appl. Phys. 104: 014308 Teich-McGoldrick SL, Bellanger M, Caussanel M, Tsetseris L, Pantelides ST, Glotzer SC and Schrimpf RD (2009) Design Considerations for CdTe Nanotetrapods as Electronic Devices. Nano Lett. 9: 3683-3688 Trudeau PE, Sheldon M, Altoe V and Alivisatos AP (2008) Electrical Contacts to Individual Colloidal Semiconductor Nanorods. Nano Lett. 8: 1936-1939 7  Figure Captions: Fig.1 Schematic drawing of nanorods (NRs), (a) individual NRs covered by surfactants, (b) trapped in-between a pair of electrodes, a high resistance state, (c) assembled by a nanowelding process into networks, and (d) networks trapped in a nanoscale gap, where the linking of the NRs via gold domains largely reduces the junction resistance. Fig. 2 Schematic drawing of the electronic circuit for the positioning of CdSe-Au networks in- between electrodes by dielectrophoresis. Fig. 3 TEM image of as-synthesized CdSe-Au networks; inset: a higher TEM magnification from the same ensemble. Fig. 4 SEM images of trapped CdSe-Au networks with trapping parameters of 1 kHz, 60 s, 5 VRMS (a), and 1 kHz, 60 s, and 3 VRMS (b). Fig. 5 Room temperature I-V curves of CdSe-Au networks with (open squares) and without (open circles) laser light illumination before thermal annealing. The solid line shows the result of the exponential fit to the data as described in text. Fig. 6 Room temperature I-V curves of CdSe-Au networks with (open squares) and without (open circles) laser light illumination after thermal annealing. Fig. 7 Detailed TEM images of (a) as-synthesized CdSe-Au nanorod networks showing also that small Au-domains decorate the surface of nanorods, (b) of post-annealed CdSe-Au networks, from the same batch, where metal domains are located only at the nanorod tips. 8 Figure 1 Figure 2 9 Figure 3 10 Figure 4 11 Figure 5 Figure 6 12  13 Figure 7 

1807.10932

2

1807

Simple model for second-order topological insulators and loop-nodal semimetals in Transition Metal Dichalcogenides XTe$_2$ (X=Mo,W)

Transition metal dichalcogenides XTe$_{2}$ (X=Mo,W) have been shown to be second-order topological insulators based on first-principles calculations, while topological hinge states have been shown to emerge based on the associated tight-binding model. The model is equivalent to the one constructed from a loop-nodal semimetal by adding mass terms and spin-orbit interactions. We propose to study a chiral-symmetric model obtained from the original Hamiltonian by simplifying it but keeping almost identical band structures and topological hinge states. A merit is that we are able to derive various analytic formulas because of chiral symmetry, which enables us to reveal basic topological properties of transition metal dichalcogenides. We find a linked loop structure where a higher linking number (even 8) is realized. We construct second-order topological semimetals and two-dimensional second-order topological insulators based on this model. It is interesting that topological phase transitions occur without gap closing between a topological insulator, a topological crystalline insulator and a second-order topological insulator. We propose to characterize them by symmetry detectors discriminating whether the symmetry is preserved or not. They differentiate topological phases although the symmetry indicators yield identical values to them. We also show that topological hinge states are controllable by the direction of magnetization. When the magnetization points the $z$ direction, the hinges states shift, while they are gapped when it points the in-plane direction. Accordingly, the quantized conductance is switched by controlling the magnetization direction. Our results will be a basis of future topological devices based on transition metal dichalcogenides.

cond-mat.mes-hall

cond-mat

Second-order topological insulators and loop-nodal semimetals in Transition Metal Dichalcogenides XTe2 (X=Mo,W) Motohiko Ezawa∗ Department of Applied Physics, University of Tokyo, Hongo 7-3-1, 113-8656, Japan ∗Correspondence to ezawa@ap.t.u-tokyo.ac.jp Abstract Transition metal dichalcogenides XTe2 (X=Mo,W) have been shown to be second-order topological insulators based on first- principles calculations, while topological hinge states have been shown to emerge based on the associated tight-binding model. The model is equivalent to the one constructed from a loop-nodal semimetal by adding mass terms and spin-orbit interactions. We propose to study a chiral-symmetric model obtained from the original Hamiltonian by simplifying it but keeping almost identical band structures and topological hinge states. A merit is that we are able to derive various analytic formulas because of chiral symmetry, which enables us to reveal basic topological properties of transition metal dichalcogenides. We find a linked loop structure where a higher linking number (even 8) is realized. We construct second-order topological semimetals and two-dimensional second-order topological insulators based on this model. It is interesting that topological phase transitions occur without gap closing between a topological insulator, a topological crystalline insulator and a second-order topological insulator. We propose to characterize them by symmetry detectors discriminating whether the symmetry is preserved or not. They differentiate topological phases although the symmetry indicators yield identical values to them. We also show that topological hinge states are controllable by the direction of magnetization. When the magnetization points the z direction, the hinges states shift, while they are gapped when it points the in-plane direction. Accordingly, the quantized conductance is switched by controlling the magnetization direction. Our results will be a basis of future topological devices based on transition metal dichalcogenides. Introduction Higher-order topological insulators (HOTIs) are generalization of topological insulators (TIs). In the second-order topological insulators (SOTIs), for instance, topological corner states emerge though edge states do not in two dimensions, while topological hinge states emerge though surface states do not in three dimensions1 -- 15. The emergence of these modes is protected by symme- tries and topological invariants of the bulk. Hence, an insulator so far considered to be trivial due to the lack of the topological boundary states can actually be a HOTI. Indeed, phosphorene is theoretically shown to be a two-dimensional (2D) SOTI16. A three-dimensional (3D) SOTI is experimentally realized in rhombohedral bismuth17, where topological quantum chemistry is used for the material prediction18. Transition metal dichalcogenides XTe2 (X=Mo,W) are also theoretically shown to be 3D SOTIs19,20. The tight-binding model for transition metal dichalcogenides has already been proposed, which is closely related to a type of loop-nodal semimetals20. A loop-nodal semimetal is a semimetal whose Fermi surfaces form loop nodes21 -- 25. Especially, the Hopf semimetal is a kind of loop-nodal semimetal whose Fermi surfaces are linked and characterized by a nontrivial Hopf number26 -- 30. There is another type of loop nodal-semimetals characterized by the monopole charge21. An intriguing feature is that loop nodes at the zero-energy and another energy form a linked-loop structure. The proposed model20 may be obtained by adding certain mass terms to this type of loop-nodal semimetals. It is intriguing that topological boundary states can be controllable externally. Magnetization is an efficient way to do so. Famous examples are surface states of 3D magnetic TIs32 -- 35, where the gap opens for out-of-plane magnetization, while the Dirac cone shifts for in-plane magnetization. Similar phenomena also occur in 2D TIs, which can be used as a giant magnetic resistor36. Recently, a topological switch between a SOTI and a topological crystalline insulator (TCI) was proposed37, where the emergence of topological corner states is controlled by magnetization direction. We ask if a similar magnetic control works in transition metal dichalcogenides. In this paper, we investigate a chiral-symmetric limit of the original model20 constructed in such a way that the simplified model has almost identical band structures and topological hinge states as the original one. Alternatively, we may consider that the original model is a small perturbation of the chiral symmetric model. A great merit is that we are able to derive various analytic formulas because of chiral symmetry, which enable us to reveal basic topological properties of transition metal dichalco- genides. We find that a linking structure with a higher linking number is realized in the 3D model. We also study 2D SOTIs and 3D second-order topological semimetals (SOTSMs) based on this model. Depending on the way to introduce mass parameters there are three phases, i.e., TIs, TCIs and SOTIs in the 2D model. We find that topological phase transitions occur between these phases without band gap closing. Hence, the transition cannot be described by the change of the symmetry indicators. We propose symmetry detectors discriminating whether the symmetry is preserved or not. They can differentiate these three topological phases. Furthermore, we show that the topological hinge states in the SOTIs are controlled by magnetization. When the magnetization direction is out of plane, the topological hinge states only shift. On the other hand, when the magnetization direction is in plane, the gap opens in the topological hinge states. 9 1 0 2 r p A 1 ] l l a h - s e m . t a m - d n o c [ 2 v 2 3 9 0 1 . 7 0 8 1 : v i X r a 2 FIG. 1: (a) Brillouin zone and high symmetry points. (b) -- (d) Bulk band structures along the Γ-X-S-Y-Γ-Z-U-R-T-Z-Y-T-U-X-S-R-Γ line (b) for loop-nodal semimetal, (c) for SOTSM and (d) for SOTI. There are four bands in each phase. The dashed magenta curves represent the band structure of the chiral-symmetric model, while the dashed cyan curves represent that of the chiral-nonsymmetric model. They are indistinguishable in these figures. We have chosen tx = ty = 1, tz = 2, λx = λy = 1, λz = 1.2, m = −3, m2 = 0.3, m3 = 0.2, mmv1 = −0.4, mmv2 = 0.2, mLoop = 0.3 and mSOTSM = 0.3. Result Hamiltonians. Motivated by the model Hamiltonian20 which describes the topological properties of transition metal dichalco- genides β-(1T'-)MoTe2 and γ-(Td-)XTe2 (X=Mo,W), we propose to study a simplified model Hamiltonian, with H0 = HSOTI = H0 + HSO + VLoop + VSOTSM, m + (cid:88) i=x,y,z ti cos ki  τz +λx sin kxτx + λy sin kyτyµy, HSO = λz sin kzτyµzσz, VLoop = mLoopτzµz, VSOTSM = mSOTSMµx, (1) (2) (3) (4) (5) where σ, τ and µ are Pauli matrices representing spin and two orbital degrees of freedom. It contains three mass parameters, m, mLoop and mSOTSM. The role of the term mLoop is to make the system a loop-nodal semimetal, and that of the term mSOTSM is to make the system a SOTSM. The Brillouin zone and high symmetry points are shown in Fig.1(a). Although the band structure of the transition metal dichalcogenides is chiral nonsymmetric, the topological nature is well described by the above simple tight-binding model. The original Hamiltonian contains two extra mass parameters and given by Loop + V (cid:48) H(cid:48) SOTI = H0 + HSO + V (cid:48) SOTSM with (6) (7) The simplified model HSOTI captures essential band structures of the original model H(cid:48) SOTI. Indeed, the bulk band structures are almost identical, as seen in Fig.1(b) -- (d). The rod band structures are also very similar, as seen in Fig.2(a4) -- (d4) and (a5) -- (d5), where the bulk band parts are found almost identical while the boundary states (depicted in red) are slightly different. V (cid:48) Loop = m2τzµx + m3τzµz, V (cid:48) SOTSM = mmv1µz + mmv2µx. 3 FIG. 2: (a1) -- (d1) Bird's eye's views of the LDOS of the zero-energy states: (a1) for H0 with surface zero-energy states on the four side surfaces; (b1) for HLoop with surface zero-energy states on the two side surfaces; (c1) HSOTSM with hinge-arc states at two pillars; (d1) HSOTI with hinge states at two pillars. (a2) -- (d2) Top view of the LDOS corresponding to (a1) -- (d1). (a3) -- (d3) Bulk band structures of valence bands along kx = 0 plane for these Hamiltonians. (a4) -- (d4) Band structures of the square rod along z direction for these Hamiltonians. (a5) -- (d5) Corresponding rod band structures for the chiral-nonsymmetric Hamiltonian H(cid:48). In these two sets of figures red curves represent topological boundary states. The Parameters are the same as in Fig.1. Moreover, the both models have almost identical hinge states, demonstrating that they describe SOTIs inherent to transition metal dichalcogenides XTe2. A merit of the simplified model is the chiral symmetry, {HSOTI(kx, ky, kz), C} = 0, which is absent in the original model, {H(cid:48) SOTI(kx, ky, kz), C} (cid:54)= 0. Accordingly, the band structure of H is symmetric with respect to the Fermi level. Moreover, the bulk band structure is analytically solved. Here, the chiral symmetry operator is C = τyµzσx or C = τyµzσy. Let us call the original model a chiral-nonsymmetric model and the simplified model a chiral-symmetric model. The common properties of the two Hamiltonians HSOTI and H(cid:48) SOTI read as follows. First, they have inversion symmetry P = τz and time-reversal symmetry T = iτzσyK with K the complex conjugation operator. Inversion symmetry P acts on HSOTI as P −1HSOTI(k)P = HSOTI(−k), while time-reversal symmetry T acts as T −1HSOTI(k)T = HSOTI(−k). Accordingly, the −1 HSOTI(k)P T = HSOTI(k), which implies that H∗ = H. Second, the z-component Hamiltonian has the P T symmetry (P T ) of the spin is a good quantum number σz = sz. Since we may decompose the Hamiltonian into two sectors, HSOTI = H ↑ SOTI ⊕ H ↓ SOTI, (8) it is enough to diagonalize the 4 × 4 Hamiltonians. All these relations hold also for H(cid:48) that the Kane-Mele model is decomposed into the up-spin and down-spin Haldane models on the honeycomb lattice38,39. SOTI. The relation (8) resembles the one A convenient way to reveal topological boundary states is to plot the local density of states (LDOS) at zero energy. First, we show the LDOS for the Hamiltonian H0 in Fig.2(a1). It describes a Dirac semimetal, whose topological surfaces appear on the four side surfaces. Then, we show the LDOS for the Hamiltonian HLoop = H0 + VLoop (9) in Fig.2(b1), where the topological surface states appear only on the two side surfaces parallel to the y-z plane. We will soon see that a loop-nodal semimetal is realized in HLoop. Next, we show the LDOS for the Hamiltonian HSOTSM = H0 + VLoop + VSOTSM (10) in Fig.2(c1), where a SOTSM is realized with two topological hinge-arcs. Finally, by including HSO, we show the LDOS for the Hamiltonian HSOTI in Fig.2(d1), where a SOTI is realized with topological two-hinge state. 4 FIG. 3: (a1) -- (d1) Loop-nodal zero-energy Fermi surfaces for (a1) tz = t1, (b1) t1 < tz < t2, (c1) tz = t2 and (d1) t1 < tz < t2. (a2) -- (d2) Band structures along kx = 0 plane. (a3) -- (d3) Drum-head surface states of the valence band along the y-z plane. tx = ty = 1, λx = λy = 1; m = −3, mLoop = 0.75. In (a2) -- (d3), only the valence bands are shown for clarity. Topological phase diagram. The chiral-symmetric Hamiltonian HSOTI is analytically diagonalizable. The energy dispersion is given by (cid:113) E = ± √ G F ± The topological phase diagram is determined by the energy spectra at the eight high-symmetry points Γ = (0, 0, 0), S = (π, π, 0), X = (π, 0, 0), Y = (0, π, 0), Z = (0, 0, π), R = (π, π, π), U = (π, 0, π) and T = (0, π, π) with respect to time-reversal inversion symmetry. The energies at these high-symmetry points (kx, ky, kz) are analytically given by where ηa = ±1 and ηb = ±1. The phase boundaries are given by solving the zero-energy condition (E = 0), E (ki) = ηaM (ki) + ηb m2 Loop + m2 SOTSM, (m + ηxtx + ηyty + ηztz)2 = m2 (17) where ηx = ±1, ηy = ±1 and ηz = ±1. There are 16 critical points apart from degeneracy. When tx = ty, the critical points are reduced to be 12 since E (X) = E (Y ) and E (U ) = E (T ). Hence, solving E = 0 for tz, there are 6 solutions for tz > 0, which we set as tn, n = 1, 2, 3,··· , 6 with ti < ti+1. Loop-nodal semimetals. We first study the loop nodal phase described by the Hamiltonian HLoop. The energy spectrum is simply given by Loop + m2 SOTSM, (cid:115) E = ± x sin2 kx + λ2 (cid:18)(cid:113) (cid:19)2 y sin2 ky + M 2 ± mLoop λ2 The loop-nodal Fermi surface is obtained by solving E (k) = 0. It follows that kx = 0 and y sin2 ky + M 2 (0, ky, kz) = m2 λ2 Loop. . (18) (19) with and F = M 2 + m2 Loop + m2 x sin2 kx + λ2 SOTSM y sin2 ky + λ2 +λ2 z sin2 kz, G = (mSOTSMλx sin kx − mLoopλy sin ky)2 +2M 2(cid:0)m2 Loop + m2 SOTSM (cid:1) , M = m + ti cos ki. (cid:88) i=x,y,z (cid:113) (11) (12) (13) (14) (15) (16) 5 FIG. 4: Evolution of linking structures for various tz. (a) tz = t1, (b) t1 < tz < t2, (c) tz = t2, (d) t2 < tz < t3, (e) tz = t3, (f) t3 < tz < t4, (g) tz = t4, (h) t4 < tz < t5, (i) tz = t5, (j) t5 < tz < t6, (k) tz = t6 and (l) tz > t6. (a1) -- (l1) Loop-nodal Fermi surfaces at the zero-energy (magenta) and at E = −mLoop (cyan). They are linked, whose linking number N is shown in figures. (a2) -- (l2) Band structure along the Γ-Z line (red), the X-U and Y -T lines (thick blue curves representing double degeneracy) and the S-R line (green). Only the valence bands are shown for 0 ≤ kz ≤ π. Cross section of the loop nodes are marked in circles. The Parameters are the same as in Fig.3. Loop nodes at zero energy exist in the kx = 0 plane. They are protected by the mirror symmetry Mx = τzµzσx with respect to the kx = 0 plane and the P T symmetry31,49. We show the band structure along the kx = 0 plane in Fig.3(a2) -- (d2). We see clearly that the loop node structures are formed at the Fermi energy in Fig.3(b2) -- (d2). These loop nodes are also observed as the drum-head surface states, which are partial flat bands surrounded by the loop nodes as shown in Fig.3(b3) -- (d3). The low energy 2 × 2 Hamiltonian is given by (cid:18)(cid:113) (cid:19) y sin2 ky + M 2 ± mLoop λ2 H = σz + λx sin kxσx, (20) (21) where σ is the Pauli matrix for the reduced two bands. In addition, there are loop nodes on the ky = 0 plane at E = −mLoop, which are determined by Loop. We find the two loops determined by Eq.(19) and Eq.(21) are linked, as shown in Fig.4. x sin2 kx + (M (kx, 0, kz) − mLoop)2 = m2 λ2 The system is a trivial insulator for 0 ≤ tz < t1. One loop emerges for t1 < tz < t2 [Fig.3(b1)], which splits into two loops for t2 < tz < t3, as shown in Fig.3(d1). Correspondingly, drum-head surface states, which are partial flat band within the loop nodes, appear along the [100] surface [see Fig.3(b3), (c3) and (d3)]. The emergence of the loop-nodal Fermi surface is understood in terms of the band inversion20,31, as shown in Fig.4. The number of the loops are identical to the number of circles at the Fermi energy as in Fig.4(a2) -- (l2). When only one band is inverted along the Γ-Z line, a single loop node appears [Fig.4(b1)]. When two bands are inverted along the Γ-Z line, two loop nodes appear [Fig4(d1)]. In the similar way, additional loops appear when additional bands are inverted along the X-U and Y -T lines [Fig.4(f1)], and it is split into two loops [Fig.4(h1)] as tz increases. In the final process, a loop appears along the S-R line [Fig.4(j1)], which splits into two loops [Fig.4(l1)]. It has been argued20,31 that a new topological nature of loop-nodal semimetals becomes manifest when we plot the loop-nodal Fermi surfaces at the band crossing energies, where one is at the Fermi energy and the other is at E = −mLoop in the occupied band. We show them in Fig.4. Along the Γ-Z line, the other band crossing occurs at ±mLoop with kz = arccos [(mLoop − m − 2t) /tz] . (22) 6 FIG. 5: (a1)-(c1) Energy spectrum as a function of m/t for TI, TCI and SOTI phases. (a2)-(c2) Corresponding Z4 index. (a3)-(c3) Corre- sponding mirror-symmetry detector χ. It follows that χ = 1 for the TI and the insulating phase of the TCI, and that χ (cid:54)= 1 for the SOTI since the mirror symmetry is broken. Along the X-U and Y -T lines, the band crossing occurs also at ±mLoop with kz = arccos [−m/tz] . Along the S-R line, the band crossing occurs also at ±mLoop with (23) kz = arccos [(−m + 2t) /tz] . (24) As a result, it is enough to plot the Fermi surfaces at E = 0 and E = −mLoop. The linking number N increases as tz increases, where even the linking number N = 8 is realized as in Fig.4(l1). 2D TI, TCI and SOTI. At this stage it is convenient to study the 2D models by setting tz = λz = 0. It follows from (17) that the 2D topological phase boundaries are given by (25) where ηx = ±1 and ηy = ±1. Depending on the way to introduce the mass parameters there are three phases, i.e., TIs, TCIs and SOTIs. The topological number is known to be the Z4 index protected by the inversion symmetry in three dimensions20,46 -- 48. This is (m + ηxtx + ηyty)2 = m2 Loop + m2 SOTSM, also the case in two dimensions. It is defined by (cid:88) κ1 ≡ 1 4 K∈TRIMs (cid:0)n+ K − n− K (cid:1) , (26) where n± K is the number of occupied band with the parity ±. There is a relation46 -- 48 mod2κ1 = ν, (27) where ν is the Z2 index characterizing the time-reversal invariant TIs. We find from Fig.5(c1) that κ1 = 0, 2 in the TI phase, which implies that it is trivial in the viewpoint of the time-reversal invariant topological insulators. We show the LDOS for TI, TCI and SOTI in Fig.6. (i) When mLoop = mSOTSM = 0 and m < 2t, the system is a TI with κ1 = 2, where topological edge states appear for all edges [See Fig.6(a)]. We show the energy spectrum and the Z4 index in Fig.5(a1) and (a2), respectively. The energy spectrum is two-fold degenerate since there is the symmetry P ¯T = µy such that x HLoop (kx, ky) Mx = HLoop (−kx, ky). (ii) When mLoop (cid:54)= 0 and mSOTSM = 0, the system is a TCI, where topological edge states appear only for two edges [See Fig.6(b)]. The energy spectrum and the Z4 index are shown in Fig.5(b1) and (b2). The symmetry P ¯T is broken for mLoop (cid:54)= 0 and the two-fold degeneracy is resolved. On the other hand, the mirror symmetry Mx remains preserved. (iii) Finally, when mLoop (cid:54)= 0 and mSOTSM (cid:54)= 0, the system is a SOTI, where two corner states emerge [See Fig.6(c)]. The energy H0(k)P ¯T = H0(k). Furthermore, there is the mirror symmetry Mx = iτzµz such that M−1 (cid:0)P ¯T(cid:1)−1 7 FIG. 6: (a1) -- (c1) Eigenvalues of the sample in a square geometry, where the insets show the zero-energy states in red. The vertical axis is the energy. (a2) -- (c2) corresponding LDOS of the zero-energy states. The amplitude is represented by the radius of the circles. We have set tx = ty = m = λ = 1 and mLoop = mSOTSM = 0.3. spectrum and the Z4 index are shown in Fig.5(c1) and (c2). The mirror symmetry is broken in the SOTI phase. In TCI and SOTI phases, there are regions where κ1 = 1, 3. However, in this region, the system is semimetallic and the κ1 index has no meaning. The Z4 index takes the same value for the TI, TCI and SOTI phases, and hence it cannot differentiate them. Indeed, because there is no band gap closing between them40, the symmetry indicator cannot change its value48. A natural question is whether there is another topological index to differentiate them. We propose the symmetry detector discriminating whether the symmetry is present or not. The TI and TCI are differentiated whether the symmetry P ¯T is present or not. The band is two-fold degenerate due to the symmetry P ¯T in the TI phase, where we can define a topological index by (cid:88) K∈TRIMs Pf [w](cid:112)det [w] ζ = Mod4 (28) (29) with wij = (cid:104)ψi (−K) P ¯T ψj (K)(cid:105) . where i and j are the two-fold degenerated band index. It is only defined for the TI phase, where it gives the same result as κ1. On the other hand, it is ill-defined for the TCI and SOTI phases since there is no band degeneracy. The TCI and SOTI are differentiated by the mirror-symmetry detector defined by χ ≡ χ+ π χ− 0 χ− π , 0 χ+ where (cid:90) 2π 0 α ≡ −i χ± 2π (cid:104)ψ Mx ψ(cid:105) dky (cid:12)(cid:12)(cid:12)(cid:12)kx=α (30) (31) is the mirror symmetry indicator37 along the axis kx = α with α = 0, π, and ± indicates the band index under the Fermi energy. It is χ = 1 when there is the mirror symmetry. On the other hand, it is χ (cid:54)= 1 when there is no mirror symmetry since ψ(cid:105) is not the eigenstate of the mirror operator. In addition, it is χ (cid:54)= 1 when the system is metallic since (cid:104)ψ Mx ψ(cid:105) changes its value at band gap closing points. See Figs.5(a3)-(c3). In Fig.5(a3), we find always χ = 1 since the mirror symmetry is preserved, where we cannot differentiate the topological and trivial phases. On the other hand, in Fig.5(b3), there are regions with χ (cid:54)= 1 where the system is metallic. Finally, we find χ (cid:54)= 1 in Fig.5(c3) since the mirror symmetry is broken. SOTSM. A 3D SOTSM is constructed by considering kz dependent mass term in the 2D SOTI model10,12,13. We set tz (cid:54)= 0, while keeping λz = 0 in the 2D SOTI model. The properties of the SOTSM are derived by the sliced Hamiltonian H(kz) along the kz axis, which gives a 2D SOTI model with kz dependent mass term M (kz). The bulk band gap closes at On the other hand, there emerge hinge-arc states connecting the two gap closing points. Accordingly, the topological corner states in the 2D SOTI model evolves into hinge-states, whose dispersion forms flat bands as shown in Fig.2(c4). M 2 (kz) = m2 Loop + m2 SOTSM. (32) 8 FIG. 7: Band structures for hinge states (a1) without magnetic field, (b1) with magnetic field along the z direction and (c1) with magnetic field along the x direction for the chiral-symmetric Hamiltonian HSOTI. Hinge states are depicted in red. (a2) -- (c2) Corresponding ones for the chiral-nonsymmetric Hamiltonian H(cid:48) SOTI. (a2) -- (c2) and (a'2) -- (c'2) The conductance is quantized proportional to the number of bands in various cases. Magnetic control of hinges in SOTI. Hinge states are analogous to edge states in two-dimensional topological insulators. Without applying external field, spin currents flow. On the other hand, once electric field is applied, charge current carrying a quantized conductance flows. We show that the current is controlled by the direction of magnetization as in the case of topological edge states. With the inclusion of the HSO, the system turns into a SOTI, which has topological hinge states. We study the effects of the Zeeman term, where the Hamiltonian is described by HSOTI together with the Zeeman term HZ = Bxσx + Byσy + Bzσz, (33) which will be introduced by magnetic impurities, magnetic proximity effects or applying magnetic field. We show the hinge states in the absence and the presence of magnetization in Fig.7. Helical hinge states appear in its absence [see Fig.7(a1)]. They are shifted in the presence of the Bz term [see Fig.7(b1)]. On the other hand, they are gapped out when the Bx or By term exists [see Fig.7(c1)]. For comparison, we also show the hinge states calculated from the chiral-nonsymmetric Hamiltonian H(cid:48) SOTI [see Fig.7(a2) -- (c2)]. The band structure is almost symmetric with respect to the Fermi energy. By taking into the fact that the σz is a good quantum number, the low energy theory of the hinge states is well described by In the presence of the external magnetic field, it is modified as H = vFkzσz + Bxσx + Byσy + Bzσz, H = vFkzσz. which is easily diagonalized to be E = ±(cid:113) (vFkz + Bz)2 + B2 x + B2 y. (34) (35) (36) It well reproduces the results based on the tight binding model shown in Fig.7. One of the intrinsic features of a topological hinge state is that it conveys a quantized conductance in the unit of e2/h. We have calculated the conductance of the hinge states in Fig.7 based on the Landauer formalism41 -- 45. In terms of single-particle Green's functions, the conductance σ(E) at the energy E is given by41 where ΓR(L)(E) = i[ΣR(L)(E) − Σ † R(L)(E)] with the self-energies ΣL(E) and ΣR(E), and σ(E) = (e2/h)Tr[ΓL(E)G † D(E)ΓR(E)GD(E)], GD(E) = [E − HD − ΣL(E) − ΣR(E)]−1, (37) (38) with the Hamiltonian HD for the device region. The self energies ΣL(E) and ΣR(E) are numerically obtained by using the recursive method41 -- 45. The conductance is quantized, which is proportional to the number of bands. When there is no magnetization or the magneti- zation is along the z axis, the conductance is 2 since there are two topological hinges. On the other hand, once there is in-plane magnetization, the conductance is switched off since the hinge states are gapped. It is a giant magnetic resistor36, where the conductance is controlled by the magnetization direction. Conclusion We have studied chiral-symmetric models to describe SOTIs and loop-nodal semimetals in transition metal dichalcogenides. The Hamiltonian is analytically diagonalized due to the chiral symmetry. We have obtained analytic formulas for various phases including loop-nodal semimetals, 2D SOTIs, 3D SOTSMs and 3D SOTIs. We have proposed the symmetry detector discriminating whether the symmetry is present or not. It can differentiate topological phases to which the symmetry indicator yields an identical value. Furthermore, we have proposed a topological device, where the conductance is switched by the direction of magnetization. Our results will open a way to topological devices based on transition metal dichalcogenides. 9 1 Zhang, F., Kane, C. L. & Mele, E.J. Surface State Magnetization and Chiral Edge States on Topological Insulators. Phys. Rev. Lett. 110, 2 Benalcazar, W. A., Bernevig, B. A. & Hughes, T. L. Quantized electric multipole insulators. science 357, 61 (2017). 3 Schindler, F., Cook, A., Vergniory, M. G. & Neupert, T. Higher-order Topological Insulators and Superconductors. in APS March Meeting 4 Peng, Y., Bao, Y. & von Oppen, F. Boundary Green functions of topological insulators and superconductors. Phys. Rev. B 95, 235143 046404 (2013). (2017). (2017). 5 Langbehn, J., Peng, Y., Trifunovic, L., von Oppen, F. & Brouwer, P. W. Reflection-Symmetric Second-Order Topological Insulators and Superconductors. Phys. Rev. Lett. 119, 246401 (2017). 6 Song, Z., Fang, Z. & Fang, C. (d − 2)-Dimensional Edge States of Rotation Symmetry Protected Topological States. Phys. Rev. Lett. 119, 246402 (2017). 7 Benalcazar, W. A., Bernevig, B. A. & Hughes, T. L. Electric multipole moments, topological multipole moment pumping, and chiral hinge states in crystalline insulators. Phys. Rev. B 96, 245115 (2017). 8 Schindler, F., Cook, A. M., Vergniory, M. G., Wang, Z., Parkin, S. S. P., Bernevig, B. A., & Neupert, T. Higher-order topological insulators. Science Advances 4, eaat0346 (2018). 026801 (2018). 9 Fang, C., Fu, L. Rotation Anomaly and Topological Crystalline Insulators. arXiv:1709.01929. 10 Ezawa, M. Higher-Order Topological Insulators and Semimetals on the Breathing Kagome and Pyrochlore Lattices. Phys. Rev. Lett. 120, 11 Geier, M., Trifunovic, L., Hoskam, M., & Brouwer, P. W. Second-order topological insulators and superconductors with an order-two crystalline symmetry. Phys. Rev. B 97, 205135 (2018). 12 Lin, M. & Hughes, T. L. Topological Quadrupolar Semimetals. Phys. Rev. B 98, 241103 (2018). 13 Ezawa, M. Magnetic second-order topological insulators and semimetals. Phys. Rev. B 97, 155305 (2018). 14 Khalaf, E. Higher-order topological insulators and superconductors protected by inversion symmetry. Phys. Rev. B 97, 205136 (2018). 15 Ezawa, M. Strong and weak second-order topological insulators with hexagonal symmetry and Z3 index. Phys. Rev. B 97, 241402(R) 16 Ezawa, M. Minimal models for Wannier-type higher-order topological insulators and phosphorene. Phys. Rev. B 98, 045125 (2018). 17 Schindler, F., Wang, Z., Vergniory, M. G., Cook, A. M., Murani, A., Sengupta, S., Kasumov, A. Y., Deblock, R., Jeon, S., Drozdov, I., Bouchiat, H., Gueron, S., Yazdani, A., Bernevig, B. A., & Neupert, T. Higher-order topology in bismuth. Nature Physics 14, 918 (2018). 18 Bradlyn B., Elcoro, L., Cano, J., Vergniory, M. G., Wang, Z., Felser, C., Aroyo, M. I. & Bernevig, B. A. Topological quantum chemistry. Nature 547, 298 (2017). 19 Tang, F., Po, H. C., Vishwanath, A. & Wan, X. Efficient Topological Materials Discovery Using Symmetry Indicators. arXiv:1805.07314. 20 Wang, Z., Wieder, B. J., Li, J., Yan, B. & Bernevig, B. A. Higher-Order Topology, Monopole Nodal Lines, and the Origin of Large Fermi Arcs in Transition Metal Dichalcogenides XTe2 (X=Mo,W). arXiv:1806.11116. 21 Fang, C., Chen, Y., Kee, H.-Y. & Fu L. Topological nodal line semimetals with and without spin-orbital coupling. Phys. Rev. B 92, 081201(R) (2018). (2015). 22 Kim, Y., Wieder, B. J., Kane, C. L. & Rappe, A. M. Dirac Line Nodes in Inversion-Symmetric Crystals. Phys. Rev. Lett. 115, 036806 (2015). 23 Yu, R., Weng, H., Fang, Z., Dai, X. & Hu, X. Topological Node-Line Semimetal and Dirac Semimetal State in Antiperovskite Cu3PdN. 24 Chan, Y.-H., Chiu, C.-K., Chou, Y. & Schnyder, A. P. Ca3P2 and other topological semimetals with line nodes and drumhead surface states. 25 Song, Z., Zhang, T. & Fang, C. Diagnosis for Nonmagnetic Topological Semimetals in the Absence of Spin-Orbital Coupling Phys. Rev. X Phys. Rev. Lett. 115, 036807 (2015). Phys. Rev. B 93, 205132 (2016). 8, 031069 (2018). 26 Chen, W., Lu, H.-Z. & Hou, J.-M. Topological semimetals with a double-helix nodal link. Phys. Rev. B 96, 041102 (2017). 27 Yan, Z., Bi, R., Shen, H., Lu, L., Zhang, S.-C. & Wang, Z. Nodal-link semimetals. Phys. Rev. B 96, 041103(R) (2017). 28 Chang, P.-Y. & Yee, C.-H. Weyl-link semimetals. Phys. Rev. B 96, 081114 (2017). 29 Ezawa, M. Topological semimetals carrying arbitrary Hopf numbers: Fermi surface topologies of a Hopf link, Solomon's knot, trefoil knot, and other linked nodal varieties. Phys. Rev. B 96, 041202(R) (2017). 30 Chang, G., Xu, S.-Y., Zhou, X., Huang, S.-M., Singh, B., Wang, B., Belopolski, I., Yin, J., Zhang, S., Bansil, A., Lin, H., Hasan, M. Z. Topological Hopf and Chain Link Semimetal States and Their Application to Co2MnGa. Phys. Rev. Lett. 119, 156401 (2017). 10 31 Ahn, J., Kim, Y. & Yang, B.-J. Band Topology and Linking Structure of Nodal Line Semimetals with Z2 Monopole Charges. Phys. Rev. Lett. 121, 106403 (2018). 32 Chen, Y. L., Chu, J.-H., Analytis, J. G., Liu, Z. K., Igarashi, K., Kuo, H.-H., Qi, X. L., Mo, S. K., Moore, R. G., Lu, D. H., Hashimoto, Sasagawa, T., Zhang, S. C., Fisher, I. R., Hussain, Z. & Shen, Z. X. Massive Dirac Fermion on the Surface of a Magnetically Doped Topological Insulator. Science 329, 659 (2010). 33 J. Zhang, Chang, C.-Z., Tang P., Zhang, Z., Feng, X., Li ,K., Wang, L.-l. , Chen, X., Liu, C., Duan, W., He, K., Xue, Q.-K. , Ma, M. & WangY., Topology-Driven Magnetic Quantum Phase Transition in Topological Insulators. Science 339, 1582 (2013). 34 Chang, C.-Z., et.al., Experimental Observation of the Quantum Anomalous Hall Effect in a Magnetic Topological Insulator. Science 340, 35 Checkelsky, J. G., Yoshimi, R., Tsukazaki, A., Takahashi, K. S., Kozuka, Y., Falson, J. , Kawasaki, M. & Tokura, Y. Trajectory of the anomalous Hall effect towards the quantized state in a ferromagnetic topological insulator. Nat. Phys. 10, 731 (2014). 36 Rachel, S. & Ezawa, M. Giant magnetoresistance and perfect spin filter in silicene, germanene, and stanene. Phys. Rev. Bbf89, 195303 37 Ezawa, M. Topological Switch between Second-Order Topological Insulators and Topological Crystalline Insulators. Phys. Rev. Lett. 121, 38 Kane, C. L. & Mele, E. J., Z2 Topological Order and the Quantum Spin Hall Effect. Phys. Rev. Lett. 95, 146802 (2005): Quantum Spin Hall Effect in Graphene. Phys. Rev. Lett. 95, 226801 (2005). 39 Ezawa, M. Monolayer Topological Insulators: Silicene, Germanene, and Stanene. J. Phys. Soc. Jpn. 84, 121003 (2015). 40 Ezawa, M. Tanaka, Y and Nagaosa, N., Topological Phase Transition without Gap Closing Scientific Reports 3, 2790 (2013) 41 Datta, S. Electronic Transport in Mesoscopic Systems (Cambridge University Press, Cambridge, England, 1995): Quantum transport: atom to transistor (Cambridge University Press, England, 2005). 42 Muñoz-Rojas, F., Jacob, D., Fernández-Rossier, J. & Palacios, J. J. Coherent transport in graphene nanoconstrictions. Phys. Rev. B 74, 43 Zârbo, L. P. & Nikoli´c, B. K. Condensed Matter: Electronic Structure, Electrical, Magnetic and Optical Properties. EPL, 80 47001 (2007): Areshkin, D. A. & Nikoli´c, B. K. I-V curve signatures of nonequilibrium-driven band gap collapse in magnetically ordered zigzag graphene nanoribbon two-terminal devices. Phys. Rev. B 79, 205430 (2009). 44 Li, T. C. & Lu, S.-P. Quantum conductance of graphene nanoribbons with edge defects. Phys. Rev. B 77, 085408 (2008). 45 Ezawa, M. Quantized conductance and field-effect topological quantum transistor in silicene nanoribbons. Appl. Phys. Lett. 102, 172103 167 (2013). (2014). 116801 (2018). 195417 (2006). (2013). 46 Po, H. C., Vishwanath, A. & Watanabe, H. Symmetry-based indicators of band topology in the 230 space groups. Nat. Comm. 8, 50 (2017). 47 Song, Z., Zhang, T., Fang, Z. & Fang, C. Quantitative mappings between symmetry and topology in solids. Nat. Com. 9, 3530 (2018). 48 Khalaf, E., Po, H. C., Vishwanath, A. & Watanabe, H. Symmetry Indicators and Anomalous Surface States of Topological Crystalline 49 Fang, C., Chen, Y., Kee, H.-Y. & Fu, L. Topological nodal line semimetals with and without spin-orbital coupling. Phys. Rev. B 92, Insulators. Phys. Rev. X 8, 031070 (2018). 081201(R) (2015). Acknowledgements The author is very much grateful to N. Nagaosa for helpful discussions on the subject. This work is supported by the Grants- in-Aid for Scientific Research from MEXT KAKENHI (Grant Nos.JP17K05490, JP15H05854 and No. JP18H03676). This work is also supported by CREST, JST (JPMJCR16F1). Author contributions M.E. conceived the idea, performed the analysis, and wrote the manuscript. Additional information Competing financial and non-financial interests: The author declares no competing financial and non-financial interests. 

1307.0726

2

1307

Transport phenomena in helical edge states interferometers. A Green's function approach

We analyze the current and the shot-noise of an electron interferometer made of the helical edge states of a two-dimensional topological insulator within the framework of non-equilibrium Green's functions formalism. We study in detail setups with a single and with two quantum point contacts inducing scattering between the different edge states. We consider processes preserving the spin as well as the effect of spin-flip scattering. In the case of a single quantum point contact, a simple test based on the shot-noise measurement is proposed to quantify the strength of the spin-flip scattering. In the case of two single point contacts with the additional ingredient of gate voltages applied within a finite-size region at the top and bottom edges of the sample, we identify two type of interference processes in the behavior of the currents and the noise. One of such processes is analogous to that taking place in a Fabry-P\'erot interferometer, while the second one corresponds to a configuration similar to a Mach-Zehnder interferometer. In the helical interferometer these two processes compete.

cond-mat.mes-hall

cond-mat

Transport phenomena in helical edge states interferometers. A Green's function approach. Bruno Rizzo,1 Liliana Arrachea,1 and Michael Moskalets2 1Departamento de Fisica and IFIBA, Facultad de Ciencias Exactas y Naturales, Universidad de Buenos Aires, Pab. I, Ciudad Universitaria, 1428 Buenos Aires, Argentina 2Department of Metal and Semiconductor Physics, NTU "Kharkiv Polytechnic Institute", 61002 Kharkiv, Ukraine (Dated: November 1, 2018) We analyze the current and the shot-noise of an electron interferometer made of the helical edge states of a two-dimensional topological insulator within the framework of non-equilibrium Green's functions formalism. We study in detail setups with a single and with two quantum point contacts inducing scattering between the different edge states. We consider processes preserving the spin as well as the effect of spin-flip scattering. In the case of a single quantum point contact, a simple test based on the shot-noise measurement is proposed to quantify the strength of the spin-flip scattering. In the case of two single point contacts with the additional ingredient of gate voltages applied within a finite-size region at the top and bottom edges of the sample, we identify two type of interference processes in the behavior of the currents and the noise. One of such processes is analogous to that taking place in a Fabry-P´erot interferometer, while the second one corresponds to a configuration similar to a Mach-Zehnder interferometer. In the helical interferometer these two processes compete. PACS numbers: 73.23.-b, 73.50.Td, 73.22.Dj I. INTRODUCTION Quantum Spin Hall (QSH) insulators1 -- 3 support heli- cal states at their edges (HES).4 -- 6 These are Kramers' pairs of counter-propagating electron states with oppo- site spin and, therefore, they are topologically protected7 against disorder in the absence of time-reversal symme- try breaking factors such as a magnetic field8 or magnetic impurities9 -- 11. Recent experiments performed on mer- cury telluride quantum wells in the QSH regime clearly demonstrated a dissipationless charge transport through the helical edge states.12 -- 14 Due to the fact that they ap- pear in the form of Kramer's pairs, the transport proper- ties of these states is strikingly different15 from the trans- port properties of other topologically protected states, like the chiral edge states of a system16,17 in the Quan- tum Hall state18. As a consequence of their helical nature, the edge states of the QSH insulators allow for the electrical control of spin currents. This property makes them promising for spintronic devices for quantum informa- tion processing.19,20 A prominent feature of the HES is a strong correlation between the propagation direction and the spin of an electron, where neither spin nor the direc- tion of movement are preserved separately. To account for this effect in transport through the HES we develop the corresponding general formalism in the framework of non-equilibrium Green's functions, which is adequate to describe transport away from the linear response regime. As an example, we apply this formalism to analyze the correlation properties of currents flowing through two simple non-trivial helical circuits, corresponding to an in- terferometer comprising two branches with one and two quantum point contacts (QPC), Fig. 1. The scattering at the QPCs connecting two helical states was discussed in FIG. 1. (Color online) Sketch of the topological insulator with two constrictions generating M = 2 quantum point con- tacts at the positions x1, x2. The filling of the different edge states is globally modified by recourse to four bias voltages V1, . . . , V4. In addition, two gate voltages applied to the top Vg,T and bottom Vg,B boundaries of the sample may locally modify the filling of the edge states within a finite region. detail in Refs. 21 -- 25, while the effect of a wide tunneling contact was also recently analyzed.26,27 Quite generally the (local) time-reversal invariance only allows for the scattering between two helical states while the (back-)scattering within the same helical state (the same Kramers' pair) is forbidden even in the pres- ence of a constriction. Therefore, the scattering at the QPC is characterized by two parameters, one describing a spin-preserving scattering and one describing a spin- flip scattering between two Kramers' pairs. This type of a spin-flip scattering does not contradict to the (lo- cal) time-reversal invariance and it can take place even if more than one helical state exists at the same edge (in our case in the vicinity of the constriction).6,13 The possibility of spin-flip processes makes an helical inter- ferometer different from two independent copies of a chi- ral electronic interferometer like those built in the quan- tum Hall regime. Interestingly, the helical interferometer shares28 properties of both the Mach-Zehnder29 and the Fabry-P´erot interferometers30,31. Another feature, which makes a helical interferometer a non-trivial circuit is the possibility of generating an effective back-scattering pro- cesses within the same Kramers' pair. For the latter mechanism to take place the global time-reversal invari- ance has to be broken, for instance, by applying a weak magnetic field (which does not break the local time- reversal invariance).32 The realization of these processes rely on the presence of inter-helical-states spin-flip tun- neling. Therefore, it is crucial to know whether such processes are actually present or no in real setups. In this paper we propose a simple test based on the noise measurement, which enables to check the presence or ab- sence of spin-flip tunneling between helical states at the QPC. The paper is organized as follows. In Sec. II we de- scribe the model of the helical interferometer comprising four metallic contacts as electron sources and sinks, heli- cal edge states as electron waveguides, and some number M of QPCs as wave splitters. Here we also present gen- eral equations for the currents and the current correlation functions. In Sec. III we calculate the transmission coef- ficients for the interferometer under study by using the Green's function approach. In Sec. IV we present and discuss results for the current and the current correla- tion functions in circuits comprising one or two QPCs. We conclude in Sec. V. II. THEORETICAL DESCRIPTION A. Model We consider the setup addressed in Refs. 28, 33 -- 35. We model the TI as a ribbon with infinite length. Each longitudinal edge of the TI hosts a Kramers pair of edge states, which is described by a Hamiltonian of one- dimensional (1D) free electrons with a definite helicity. The sample is assumed to have a number M of constric- tions that define QPCs through which electrons perform inter-pair tunneling processes. Each constriction has two tunneling channels. One of them is spin preserving and the other one involves a spin-flip. The full Hamiltonian reads H = H0 + Ht + Hf + Hg. (1) The Hamiltonian for the HESs is H0 = −ivF Xσ=↑,↓ Z dx[: Ψ† R,σ(x)∂xΨR,σ(x) : − : Ψ† L,σ(x)∂xΨL,σ(x) :], (2) where σ stands for the spin opposite to σ. We have as- sumed that a Kramers pair of right-moving(R) with ↑ 2 spin electron states and left-moving (L) with ↓ spin ones lie along the top edge of the sample, while another pair L, ↑ and R, ↓ lie on the bottom, as shown in Fig 1. The Fermi velocity vF is assumed to be the same for the two pairs and : O : denotes normal ordering of the operator O. The two types of tunneling terms represented by dis- tinct quantum point contacts (QPC) located at xj, j = 1, . . . , M , are Hp = Xσ=↑,↓ Z dxΓp(x)[Ψ† R,σ(x)ΨL,σ(x) + H.c.], (3) for the spin-preserving process and Hf = Xα=L,R sαZ dxΓf (x)[Ψ† α,↑(x)Ψα,↓(x) + H.c.], (4) with sR,(L) = +(−), for the case of spin-flip tunneling. We assume, for simplicity, that the corresponding ampli- tudes are the same for all the QPCs. Hence Γp(f )(x) = 2vF M Xj=1 γp(f )δ(x − xj ). (5) The Hg term refers to gate voltages applied at the top Vg,T and bottom Vg,B edges of the sample allowing for the manipulation of the filling of the helical channels within a finite region between the longitudinal coordinates x1 and xM , Hg = Z xM x1 dx[eVg,T (ρR,↑(x) + ρL,↓(x)) +eVg,B (ρR,↓(x) + ρL,↑(x))], (6) where ρα,σ =: Ψ† erator of the species α, σ at the coordinate x. α,σ(x)Ψα,σ(x) : is the local density op- B. Transport properties In the setup of Fig. 1, the transport is induced by changing the population of the edge states by applying voltages at the four metallic contacts (reservoirs) indi- cated in the corners. The coupling between edge states and metallic reservoirs is in general a subtle issue and the transport properties strongly depend on the details of the contacts.36 Here, we assume that the contact is such that the edges are in equilibrium with the respec- tive reservoir of departure. It is important to notice that this configuration enables the induction of currents even for vanishing tunneling couplings γp(f ). In that case, such currents are due to an imbalanced population of right and left movers and the carriers do not experience any kind of scattering process. For instance, a voltage difference V1 − V4 induces a current only in the terminals 1 and 4, which flows 1 → 4 or 4 → 1 for V1 > V4 and V4 > V1, re- spectively. Similarly, a voltage difference V2 − V3 induces a current in the terminals 2 and 3, which flows 2 → 3 or 3 → 2 for V2 > V3 and V3 > V2, respectively. Interest- ingly, due to the helical nature of the edge states, each current flowing through a given terminal has a net po- larization. For finite values of γp(f ), a voltage difference between any two contacts induces currents through the four terminals, which are the result of scattering and in- terference effects due to the inter-edge tunneling through the QPCs. In this section we derive expressions for the currents flowing through the different terminals. 1. Current The current operator is defined from the conservation of the charge in a given terminal l = 1, . . . , 4 3 where τ = t − t′. We will study the ω = 0 component at T = 0, called the shot-noise. For non-interacting fermions, the mean values entering Eq. (11) can be simply decoupled by recourse to Wick's theorem. The resulting expression can be cast in terms of Fourier transforms of the lesser Green's functions38 G< ασ,α′σ′ (x, x′; t, t′) = Z +∞ −∞ dω 2π e−i ω  (t−t′)G< ασ,α′σ′ (x, x′, ω) (13) as follows Sl,l′ (0) = e2v2 F Z +∞ l′ξ′,lξ(x′, x, ω) + G< G> −∞ dω 2π Xξ,ξ′=+,− l′ξ′,lξ(x′, x, ω)G> lξ,l′ξ′ (x, x′, ω)× ξξ′hG< lξ,l′ξ′ (x, x′, ω)i . (14) (7) III. GREEN'S FUNCTIONS APPROACH A. Dyson's equations It was shown in the previous section that all the ob- servables of interest can be expressed in terms of lesser Green's functions. In order to calculate the latter we use the Schwinger-Keldysh-Kadanoff-Baym technique. We introduce the retarded Green's function, Gr ασ,βσ′ (x, x′; t, t′) = −iθ(t−t′)h{Ψ† β,σ′(x′, t′), Ψα,σ(x, t)}i. (15) Following the standard procedure39, we derive the Dyson's equation for this Green's function from the equa- tion of motion ασ,βσ′ (x, x′; t, t′) = δ(t − t′)δα,βδσ,σ′ −i∂t′Gr −iθ(t − t′)h{[H, Ψ† β,σ′(x′, t′)], Ψα,σ(x, t)}i. (16) Since the Hamiltonian contains spin-preserving and spin- flipping terms, the Dyson's equation reduces to a set ασ,βσ′ (x, x′, ω), where the lat- of linear equations for Gr ter function is the Fourier transform of (15). The corre- sponding equations for the retarded as well as the lesser function are explicitly shown in Appendix A. In what follows, we discuss some formal elaboration of the Dyson's equations, which is useful to evaluate the currents and to set the relation to the scattering matrix approach. Notice that in order to evaluate the current from Eq. (9) we just need the diagonal elements of the Green function in the indices λ ≡ α, σ. We start by fo- cusing on the diagonal elements of the retarded Green's λ,λ(x, x′, ω). After some alge- function Gr bra on the Dyson's Equation based on back-substituting Green's functions with off-diagonal indices (see Appendix A) we obtain the following equation λ(x, x′, ω) ≡ Gr M λ(x, x′, ω) = g0,r Gr Gr λ (x, x′, ω) + Xj,j ′=1 λ(xj , xj ′ , ω)g0,r λ (xj ′ , x′, ω), λ(x, xj , ω) ×Σr (17) Nl = ∂x{vF (cid:2)ρl +(x) − ρl −(x)(cid:3)}, where x is a position within that terminal and ρl ±(x) is the density operator corresponding to the incoming (out- going) fermionic species flowing at that position. Thus, the current operator for electrons flowing into the termi- nal l reads I l(x) = evF (cid:2) ρl +(x) − ρl −(x)(cid:3) . (8) The ensuing mean value I l(x) = h I l(x)i can be expressed as follows I l(x) = −ievF hG< l+,l+(x, x; t, t) − G< l−,l−(x, x; t, t)i . (9) We have introduced the lesser Green's functions G< ασ,α′σ′ (x, x′; t, t′) = ihΨ† α′,σ′ (x′, t′)Ψα,σ(x, t)i, (10) where the couple of indices α, σ, with α = L, R and σ =↑, ↓, labels the incoming (outgoing) state l+ (l−) of the terminal l. Due to the conservation of the charge, the current I l(x) does not depend on x inside a given terminal, i.e. for positions x at the right (left) of the last (first) QPC. 2. Noise To characterize the fluctuations of the currents away from their mean values at the terminals l, l′ we introduce the following correlation function37 Sl,l′ (x, x′; t, t′) = 1 2 hh{ I l(x, t), I l′ −2h I l(x, t)ih I l′ (x′, t′)}i (x′, t′)ii . (11) The spectral power of current fluctuations, the noise power, reads Sl,l′ (x, x′, ω) = Z +∞ −∞ dτ ei ω  τ Sl,l′ (x, x′; t, t + τ ), (12) We have introduced the "self-energy" describing the es- cape of the electrons at edge λ to the other edges. It is also convenient to cast the terms associated to the arguments xj , xj ′ , j = 1, . . . , M as elements of M × M matrices. In this matrix notation, the self-energy reads Σr λ(ω) = Σ0,r λ,λ(ω) + Σ0,r λ,λ (ω) G0,r λ (ω) Σ0,r λ,λ (ω), (18) where the functions entering the matrices Σ0,r λ,λ′ (ω) have been defined in Appendix A. The Green's function G0,r λ (xj , xj ′ , ω), j, j′ = 1, . . . , M entering (18) can also be organized in a matrix form as follows 4 The first term is the equilibrium lesser Green's function corresponding to all the edges with the same chemical potential µλ. The latter corresponds to the reservoir from where the electrons at the λ edge are injected. It reads G<,eq λ (x, x′, ω) = [Ga λ(x, x′, ω) − Gr λ(x, x′, ω)] fλ(ω).(21) It will be useful to define a reference chemical po- tential µ0 λ and a lesser function corresponding to all the edges at equilibrium with that chemical potential G<,eq λ(ω). λ Hence, the lesser function (21) can be, thus, rewritten as λ(x, x′, ω) − Gr λ(x, x′, ω)] f 0 (x, x′, ω)µ0 = [Ga λ λ (ω) = h[g0,r G0,r λ (ω)]−1 − Σ0,r λ,λ(ω)i−1 . (19) G<,eq λ The corresponding lesser Green's functions G< λ (x, x′, ω) can be calculated by using Langreth rules39 in Eq. (17), as discussed in Appendix A. It is convenient to decom- pose this function as follows λ (x, x′, ω) = G<,eq G< λ Gr λ(x, xj , ω) (x, x′, ω) +Xj,j ′ (xj, xj ′ , ω)Ga ×Σ<,neq λ λ(xj ′ , x′, ω). (20) (x, x′, ω) = G<,eq × [Ga λ (x, x′, ω)µ0 λ(x, x′, ω) − Gr +(cid:2)fλ(ω) − f 0 λ(x, x′, ω)] λ λ(ω)(cid:3) (22) where fλ(ω) and f 0 λ(ω) are the Fermi functions cor- responding to the chemical potentials µλ and a refer- ence chemical potential µ0 λ, respectively. The advanced Green's function is related to the retarded one through λ(x′, x, ω)]∗. The non-equilibrium part λ(x, x′, ω) = [Gr Ga of the lesser self-energy can be calculated from Eq. ( A7) in Appendix A and reads Σ<,neq α,σ (ω) = i(cid:2)fασ(ω) − f 0 ασ(ω)(cid:3) Γp ασ(ω) + i(cid:2)fασ(ω) − f 0 ασ(ω)(cid:3) Γf ασ(ω) + i(cid:2)fασ(ω) − f 0 ασ(ω)(cid:3) Γpf ασ(ω), (23) where, as before, we are using the matrix notation to omit explicit reference to the coordinates of the con- tacts xj, xj ′ , j = 1, . . . , M . The "hybridization" matrices Γλ(ω) are Γp ασ(ω) = 42v2 F {γ2 p ρ0 ασ(ω) + 2γpγf sα ρ0 Γf ασ(ω) = 42v2 ασ(ω) = Σ0,r Γpf F {γ2 f ρ0 ασ,ασ(ω)hΛr ασ(ω) + 2γpγf sα ρ0 ασ,ασ(ω) + 1i ρ0 f Λr ασ(ω)RehΛr ασ(ω)RehΛr ασ(ω)h1 + Λa ασ,ασ(ω)i + γ2 ασ,ασ(ω)i + γ2 Λr ασ,ασ(ω)i Σ0,r p ασ,ασ(ω) ασ,ασ(ω)ρ0 ασ(ω)Λa ασ,ασ(ω)}, ασ,ασ(ω)ρ0 ασ(ω)Λa ασ,ασ(ω)}, (24) where ρ0 ασ(ω) = i(cid:2)g0,r ασ (ω) − g0,a ασ,ασ(ω) = Σ0,r Λr ασ (ω)(cid:3), and ασ,ασ(ω) G0,r ασ (ω), with [Λr ασ,ασ(ω)]† = Λa ασ,ασ(ω). B. Transmission functions The charge current flowing through the terminal l de- (9) can be written in terms of the lesser fined in Eq. function defined in (20) as (25) I l(x) = −ievF Z +∞ −∞ dω 2π (cid:2)G< l+(x, x, ω) − G< l−(x, x, ω)(cid:3) . (26) In what follows we will eliminate the explicit reference to the coordinate x in the current and we will simply label this quantity with the terminal index l. The re- tarded Green's functions depend on x and we will take any value of this coordinate within the terminal under consideration. After identifying the pair of indices ασ (ασ) corresponding to the incoming (outgoing) channel of the terminal l and defining the reference chemical po- tential as the one corresponding to the outgoing channel, i.e µ0 α,σ ≡ µl− we substitute the lesser function of Eq. (20) with the representation defined in Eq. (21) and find 5 dω 2π ασ(x, xj , ω){Γp Gr ασ(xj , xj ′ , ω) [fασ(ω) − fl−(ω)] + Γf ασ(xj , xj ′ , ω) [fασ(ω) − fl−(ω)] ασ(xj , xj ′ , ω)i [fασ(ω) − fl−(ω)]}Ga ασ(xj ′ , x, ω) − evF Xjj ′ Z +∞ −∞ dω 2π Gr ασ(x, xj , ω) I l = I l −∞ 0 + evF Xjj ′ Z +∞ −hΓp ×{Γp ασ(xj, xj ′ , ω) + Γf ασ(xj , xj ′ , ω) [fασ(ω) − fl−(ω)] + Γf ασ(xj , xj ′ , ω) [fασ(ω) − fl−(ω)]}Ga ασ(xj ′ , x, ω), (27) where the first term corresponds to the current without tunneling to the other edges I l 0 = −ievF Z +∞ = evF Z +∞ −∞ dω 2π −∞ dω ασ (x, x, ω) − G<,eq 2π (cid:2)G<,eq [fl+(ω) − fl−(ω)] ρ0 ασ (x, x, ω)µl−(cid:3) ασ(x, x, ω). (28) At zero temperature, substituting (A8) into ρ0 results in ασ(x, x, ω) I l 0 = e h (µl+ − µl−). (29) the to the chiral nature of Due electronic mo- tion within the edges, the retarded Green's function Gr R(L),σ(x, xj ′ , ω) for right (left)-moving electrons van- ish for x < x1 (x > xM ), where we assumed that the scattering region containing the point contacts extends within [x1, xM ]. In addition, we must take into account that the terminals placed at the right side of the scat- tering region(x > xM ) correspond to l+ = R ↑, R ↓ for l = 3, 4, respectively, while those at the left side have l+ = L ↑, L ↓ for l = 1, 2, respectively. Thus, the two last lines of Eq. (27) vanish. For the same reason, the term ∝ Γpf of (23) does not contribute and the ex- pression of the current can be cast into the form of the Landauer-Buttiker formula, see, e.g., Ref. 40, I l = e  4 Xl′=1 Z dω 2π {Tl+,l′−(ω) [fl′−(ω) − fl−(ω)]},(30) where Tl+,l′−(ω), l′− = 1, . . . , 4 are transmission func- tions between the incoming channel l+ and the remain- ing four channels through the tunneling contacts. For the model we are considering, these functions explicitly read Tl+,l′−(ω) = vF Xjj ′ ασ(x, xj , ω){δl′,ασΓp Gr ασ(xj , xj ′ , ω) ασ(xj , xj ′ , ω)}Ga +δl′,ασΓf ×{ρ0 ασ(x, x, ω) −Xjj ′ ασ(xj , xj ′ , ω)]Ga +Γf ασ(xj ′ , x, ω) + vF δl′,ασ ασ(x, xj , ω)[Γp Gr ασ(xj, xj ′ , ω) ασ(xj ′ , x, ω)}, (31) where, given l+ ≡ α, σ, the first term corresponds to the transmission between the channel l′− = α, σ and the channel l+ = α, σ, the second term corresponds to the transmission between l′− = α, σ and the channel l+ ≡ α, σ, and the latter is the transmission function between l′− = α, σ and l+. The functions Γ(xj , xj ′ , ω) are the matrix elements of the hybridization matrix Γp ασ(ω), for l′− ≡ ασ or the ones of the matrix Γf ασ(ω), for l′− ≡ ασ. C. Relation to the scattering matrix formalism It is interesting to notice that the transmission func- tions defined in the previous section set an explicit rela- tion between the scattering matrix and the Green's func- tion formalism. In the case of transport through nor- mal tunneling contacts between two reservoirs at different chemical potentials, Fisher and Lee's equation41 provides such an explicit relation. This equation has been gener- alized to harmonically time-dependent problems,42 but so far, it has not been analyzed in the context of trans- port through edge states alone without tunnel coupling to the contacts. To establish such a relation for trans- port of helical edge states in bar geometry we proceed as follows. For simplicity, we start by considering a single QPC connecting the top and bottom edge states. In such case we can set the following identity between elements of the scattering matrix and the retarded diagonal (in the edge indices) elements of the Green's functions Sl+,l′− = −ipvF Gr ασ(x, x1, ω)pΓα′σ′ (x1, x1, ω), (32) for α′σ′ 6= ασ and α′σ′ 6= ασ. We recall the terminal l has an incoming edge state characterized by ασ, which is denoted by l+, and an outgoing one characterized by ασ, which would be denoted by l−, while l′ injects an edge characterized by α′σ′, which is denoted by l′−. Hence, the elements of the scattering matrix defined in Eq. (32) correspond to terminals l, l′ such that l is on the top (bot- tom) and l′ is on the bottom (top) of the sample. We also recall that x lies on the l-th terminal, and x1 is the posi- tion of the QPC where the tunneling contact between the edge connected to the l reservoir and the edge connected to the l′ one . We can easily identify this expression with the one proposed by Fisher and Lee. The hybridization function Γl′−(x1, x1, ω) denotes scattering processes due to the escape of the electrons from the edge injected into the reservoir l and the edge leaving the reservoir l′. In the present case, the contact of the incoming edge state and the reservoir l is assumed to be ideally ballistic, thus the hybridization matrix is just represented by vF . In ασ(x, x′, ω) = 0 for x′ > x within the l termi- addition, Gr nal, which is a consequence of the absence of reflection of the helical edge state into the terminal of departure. In the present system, this is due to time-reversal symmetry which dictates lack of scattering between the two states of a Kramer's pair. Thus, Sl+,l− = 0. (33) In all the cases we define the transmission function as follows Tl+,l′− = Sl+,l′ −2, while the conservation of the charge impose Sl+,l′′ −S ∗ l′+,l′′− = δl,l′ . Xl′′− Thus, Tl+,l′− = 1. Xl′− (34) (35) (36) In the case of M QPC, the scattering between top and bottom edges becomes multichannel. In such case, Fisher and Lee's equation is more suitable represented as Tl+,l′− = Γl+ M Xj,j ′=1 Gr ασ(x, xj , ω)Γα′σ′ (xj , xj ′ , ω) ×Ga ασ(xj ′ , ω), (37) 6= ασ and α′σ′ for α′σ′ 6= ασ. We consider a ballis- tic hybridization parameter Γl+ = vF associated to the ideal connection between the edge and the reservoir l to- wards it travels, while the M × M hybridization matrix Γα′σ′ (xj, xj ′ , ω) represents the escape to the edge α′σ′, which is injected from the reservoir l′, through the M QPCs. Due to the helical nature of the states Tl+,l− = 0, while due to the conservation of the charge (36) 1. Charge-bias configuration This corresponds to V1 = V2 = V , V3 = V4 = 0 . 6 (39) We, thus, assume µ2 = µ1 = µ + eV and µ3 = µ4 = µ. Eq. (30) for the current through l = 3 corresponds to l+ ≡ R, ↑ and l− ≡ L, ↓ e  Z dω 2π {T3,1(ω) [f1(ω) − f3(ω)] +T3,2(ω) [f2(ω) − f3(ω)]}, (40) I 3 = with R↑(x, xj , ω)Γp Gr L,↑(xj , xj ′ , ω)Ga R↑(xj ′ , x, ω), T3,1= vF Xjj ′ T3,2= vF ρ0 R↑(x, x, ω) − vF Xjj ′ Gr R↑(x, xj , ω) × hΓp L↑(xj , xj ′ , ω) + Γf R↓(xj , xj ′ , ω)i Ga R↑(xj ′ , x, ω), (41) where we have simplified the notation and expressed T3,l′, l′ = 1, 2, instead of T3+,l′− as in the previous sec- tion. This configuration generates a net longitudinal flow of charge. Another possibility is to generate net transverse charge flow, V1 = V4 = V , V2 = V3 = 0 . (42) This corresponds to µ2 = µ3 = µ, and µ1 = µ4 = µ + eV . In this case the current through l = 3 reads e  Z dω 2π {T3,1(ω) [f1(ω) − f3(ω)] +T3,4(ω) [f4(ω) − f3(ω)]}, (43) I 3 = with Tl+,l− = 1 − Xl′−6=l+ Tl+,l′−, (38) T3,4 = vF Xjj ′ R↑(x, xj , ω)Γf Gr R,↓(xj , xj ′ , ω)Ga R↑(xj ′ , x, ω). where l− ≡ l+ denotes the reservoir injecting the ασ edge state that incomes into the l reservoir. The two reservoirs l and l are on the same (top or bottom) part of the sample. D. Currents for particular configuration of bias voltages In order to illustrate the use of the previous expres- sions we present more explicitly the expressions for the currents along the terminal l = 3 for three different con- figurations of voltages. (44) (45) 2. Spin-bias configuration Here we have V1 = −V2 = V , V3 = V4 = 0 . We, thus, assume µ1 = µ + eV , µ2 = µ − eV and µ3 = µ4 = µ. In this case the current through l = 3 is given in Eq. (40). Another spin-bias configuration corresponds to V2 = V4 = V , V1 = V3 = 0 , (46) in which case the current through l = 3 reads I 3 = e  Z dω 2π {T3,2(ω) [f2(ω) − f3(ω)] 7 γ p=0.2 γ p=0.5 γ p=0.8 γ p=1 1 0.75 I 0.5 0.25 +T3,4(ω) [f4(ω) − f3(ω)]}. (47) IV. RESULTS We now show results for the behavior of the current and the noise power. We focus on temperature T = 0 and µ = 0, and we analyze separately the case of a single QPC and two QPCs. We have verified that we recover the results presented in Ref. 28 for the first type of charge and spin bias configurations presented in the previous section. In what follows, we discuss the second type of spin- bias configurations shown in the previous section, V2 = V4 = V and V1 = V3 = 0, Eq. (46), which has not been studied in previous works.28,34 In some cases, we extend our study to a more general configuration with V2 6= V4 and V1 = V3 = 0. A. Single point contact In the case of a single QPC, the transport is not af- fected by the top and bottom gate voltages Vg,T and Vg,B. The transmission functions entering the current through the different terminals can be easily evaluated by substi- tuting Eq. (B1) in Eq. (31). The final expressions are summarized for completeness in Appendix D.1. 0 0 0.25 PSfrag replacements 0.5 γf 0.75 1 FIG. 2. (Color online) Current I1 = I3 = −I2 = −I4 in units of e2V /h versus spin-flipping term in a spin-bias configuration V1 = V3 = 0 and V2 = V4 = V . Different plots correspond to different values of the spin-preserving tunneling γp. changes with a finite spin flip tunneling γf 6= 0. This tun- neling process effectively enables the transmission of elec- trons injected from the contact 2 into the contact 4 after flipping the spin ↑→↓ at the QPC. Conversely, electrons injected with ↑ spin from 4 perform a spin flip tunneling at the QPC and enter the contact 2 with spin ↓. The net result of these processes is a decreasing net current through all the terminals as well as through the QPC. In particular, for small γp and γf ∼ 1, the QPC behaves as an approximately ballistic contact for each spin species, allowing for a perfect transfer of particles accompanied by a spin-flip. For this bias configuration, however, the top- to-bottom flow is identical to the bottom-to-top one and they cancel one another, resulting in a vanishing small net current through the QPC, as well as through the four terminals. 2. Noise 1. Currents The corresponding behavior of the noise power is In Fig. 2 we show the behavior of the currents through the terminals l = 3, 4 vs the spin-flipping tunneling am- plitude γf and a finite spin-preserving tunneling γp at the QPC. Notice that because of the symmetry of the setup I1 = I3, I2 = I4 and I3 = −I4. For vanishing γf and γp there is a net ↑ flow from the left to the right in the upper terminals, and another net ↑ flow in the lower terminals and a vanishing current through the QPC. Turning on just the spin preserving tunneling (γp 6= 0, γf = 0) for the present bias configuration does not change this pic- ture. In fact, the contacts 2 and 4 inject an identical flow of ↑ electrons that travel in opposite directions and the spin preserving tunneling does not change the zero value of the net current flowing through the QPC due to the Pauli exclusion principle. This situation, however, shown in Fig. 3. In the upper panel, Fig. 3 a), we show the cross- correlation function for currents flowing through top and bottom terminals. We see that the cross-correlation function for currents flowing through the terminals at the same voltage, see Eq. (46), is exactly zero, S3,1 = S2,4 = 0. Instead, the cross-correlation function for cur- rents flowing through the terminals biased differently, S2,1 = S3,4, vanishes only for γf = 0. The vanishing cross-correlator in the latter case is due to the fact that currents flowing in all the terminals do not fluctuate at γf = 0, see Fig. 3. The relevant components of the correlation function for currents flowing between the top terminals are shown in Fig. 3 b), where we show S3,3 = S2,2 and S2,3. The corresponding ones for the bottom terminals can be in- ferred from the top ones by noticing that S1,4 = S2,3, γ p=0.2 γ p=0.5 γ p=0.8 γ p=1 0.25 0.5 γf 0.75 1 2 1.5 1 V2 0.5 0 0.25 0.5 γf 0.75 1 PSfrag replacements 8 0.4 S 0 -0.4 a) 0 -0.1 -0.2 S -0.3 -0.4 -0.5 0 PSfrag replacements γ p=0.2 γ p=0.5 γ p=0.8 γ p=1 b) 0.4 0.2 S 0 -0.2 -0.4 FIG. 4. (Color online) The cross-correlation function S1,3 (the lower sheet) and the auto-correlation function S3,3 (the upper sheet) both in units of e3V /(2h) are shown as function of both the spin-flip rate γf and the voltage V2. The latter quantity is normalized by V4 = V , hence, V2 = V4 (∆V = 0) corresponds to V2 = 1 in the figure.The spin-preserving rate is γp = 0.5. The voltages V1 = V3 = 0. nels, from the contacts 2 and 4 are fully filled up to the same level, that happens at V2 = V4, then both outgoing channels, into contacts 1 and 3, will be also fully filled to the same level. This is because the Pauli exclusion principle forces the electrons colliding at the QPC to go to different outputs (see, e.g., Refs. 37 and 43 for a more detailed discussion). As a consequence, not only the in- coming currents I2 and I4, but also the outgoing ones, I1 and I3 (if V1 = V3 = 0) will be non-fluctuating. There- fore, for γf = 0 we have S1,1 = S3,3 = 0 as it is clear in Fig. 3 b). The cross-correlator of non-fluctuating cur- rents is trivially zero, S1,3 = 0. If two incoming channels are filled up to different levels, ∆V ≡ V2 − V4 6= 0, then the excess flow from the contact with higher level will be scattered between two outgoing channels. Since one particle cannot be scattered to both outputs simultane- ously, the outgoing streams become fluctuating, S1,1 > 0, S3,3 > 0, and S1,3 < 0, see Fig. 4. These fluctuations are referred to as the shot noise appeared due to indivisibility of carriers.37 The spin-flip processes, γf 6= 0, open additional scat- tering channels. In this case, electrons can be scattered to the terminals 2 and 4, which causes current fluctua- tions even for V2 = V4. The reason for these current fluc- tuations is a smaller number of (completely and equally filled) incoming channels in comparison to the number of outgoing channels. In fact, spin-flip processes opens the total number of four outgoing channels, while the incoming ones are just two. It is, however, remarkable that, even in the presence of spin-flip processes the cross- correlator S13 = 0, see the plot in dashed lines of Fig. 3 a). The vanishing value of the latter cross-correlator for 0 0.25 PSfrag replacements 0.5 γf 0.75 1 FIG. 3. (Color online) The noise power S in units of e3V /(2h) versus spin-flipping term in a spin-bias configuration V1 = V3 = 0 and V2 = V4 = V . a) The cross-correlation func- tions S3,1 = S2,4 (dashed blue line) and S2,1 = S3,4 (solid lines) for currents flowing between the top l = 2, 3 and bot- tom l = 1, 4 terminals. b) The cross-correlation function S2,3 (dashed lines with open symbols) and the auto-correlation functions S2,2 = S3,3 (solid lines with solid symbols) for cur- rents flowing through the top l = 2, 3 terminals. Different colors and symbols correspond to different values of γp. S3,3 = S1,1 and S4,4 = S2,2. Let us begin analyzing S23, which is shown in dashed lines in Fig. 3 b). It is zero at γf = 0, since, as mentioned above, in this case the cur- rents are non-fluctuating. As γf is turned on, the current through the QPC increases, implying an increasing noise power S2,3. 3. The Pauli peak The electron flows emanated by the metallic contacts 2 and 4 are not fluctuating at zero temperature. They can fluctuate only after the electron reflections and trans- missions that take place at the QPC. In the absence of spin-flip tunneling in the bias configuration defined in Eq. (46), electrons with the same spin injected from ter- minals 2 and 4 collide at the QPCs and are backscattered into the terminals 1 and 3. Thus, if both incoming chan- γf 6= 0 and V2 = V4 can be explained as due to the cancellation of positive two-particle contributions, when electrons from 2 an 4 attempt to enter the same ter- minal 1 or 3, and negative single-particle contribution to noise, when each electron is scattered independently. For a more detailed discussion, see Ref. 44. In Fig. 4 we show S1,3 and S3,3 as a function of both the rate of spin-flip tunneling γf and a bias difference ∆V = V2 − V4. Both correlators are suppressed for ∆V = 0. Since this suppression is due to the fermionic correlations arising between colliding electrons, we use the names of the Pauli peak for S1,3 and the Pauli dip for S3,3 as functions of ∆V . This peak/dip structure is clearly visible at a small rate of spin-flip scattering, γf → 0, in Fig. 4 . While the cross-correlator is sup- pressed down to zero, the auto-correlator's dip depends on the rate of spin-flip processes γf . To show this explic- itly we use the scattering matrix approach40,45, and cal- culate S3,3 analytically at V2 = V4 = V and V1 = V3 = 0 : S3,3 = e3 h {T3,1 [V2T3,2 + V4T3,4] + V2 − V4T3,2T3,4} , (48) S3,1 = − e3 h V2 − V4T3,2T3,4 , where the transmission functions Tl,l′ are presented in Appendix D. It becomes apparent from the previous expression that S1,3 = 0 at V2 = V4. For γf → 0 and V2 = V4 = V we find the following behavior of S3,3 in the leading order in the spin-flip rate, S3,3 = e3V h 4γ2 f (1 + γ2 p)2 + O(cid:0)γ4 f(cid:1) . (49) Thus the gap between the maximum of S13 and the mini- mum of S33 would unambiguously demonstrate an actual presence of spin-flip scattering at the QPC allowed by the symmetry of helical states. The numerical calculations confirm that the above re- lations, Eqs. (48) and (49), completely agree with the results of the Green's function approach, see Eqs. (14) and (41), (44). The Pauli peak in the cross-correlation function (as a function of a bias asymmetry ∆V ) is robust and is the di- rect consequence of the fermionic nature of carriers. The gap appearing between the cross- and auto-correlation functions at both V2 = V4 and γf 6= 0 is a direct con- sequence of the helical nature of the edge states. An- other peculiar manifestation of the helical nature of edge states has been recently predicted in Ref. 46 under the name of Z2 peak. This feature manifests itself in the dependence of the current cross-correlator on the exter- nal magnetic field B as a peak at B = 0. The origin is an exact cancellation of two different components of the noise, the partition noise and the exchange noise. Such 9 cancellation is a consequence of the time-reversal sym- metry of the helical state. It is removed as soon as it is broken by a small magnetic field B, in which case the cross-correlation function becomes finite and negative. For γf = 0, our setup of helical states can be effectively decoupled into two overlapping but independent sets of chiral edge states connected in pairs by the QPC. In this case, the Pauli peak and dip discussed above is analo- gous to the one discussed47 and measured48 for electrons emitted with the same rate by two single-electron sources into a chiral edge state of a two-dimensional electron gas in the quantum Hall effect coupled by a QPC. The differ- ence between emission times ∆τ of single-electron sources plays the same role as the voltage difference ∆V = V2−V4 in the present case. For ∆τ = 0 electrons collide at the QPC and due to the Pauli principle they are necessar- ily scattered to different outputs. Hence, the outgoing currents are noiseless. In contrast, if two electrons pass the QPC at different times, ∆τ 6= 0 they are scattered independently. This results in the shot noise (a positive auto-correlator and a negative cross-correlator for out- going currents). Instead, for γf 6= 0, the behavior of the noise power in our setup with helical states is analo- gous to an electronic circuit with chiral states as waveg- uides having the number of outgoing states exceeding the number of populated incoming states. In particular, in Ref. 44 it was shown that for the synchronized emission of electrons in two incoming channels (the analogue of ∆V = 0 in our setup) the cross-correlator vanishes while the auto-correlator remains finite as in our case. B. Two point contacts In the case of an interferometer with two QPC, the cur- rent exhibits oscillations when plotted against the gate potentials. The relevant transmission functions can be calculated from the retarded Green's function and the hy- bridization function presented in Appendix C. For com- pleteness, we present the explicit expression for these functions in Appendix D 2. The different interference processes are characterized by two phases defined in Eq.(D3), the Fabry-P´erot phase φF P , which is symmet- ric in gate voltages, and the Mach-Zehnder phase φM Z , which is antisymmetric in gate voltages. The first phase is associated with the spin-preserving tunneling when a particle traverses the arms of the interferometer in the same direction (clockwise or anticlockwise ) like in the Fabry-P´erot interferometer. Instead, the second phase is associated with the spin-flip tunneling, which forces an electron to traverse the interferometer only once similarly to what happens in the Mach-Zehnder interferometer. At this point, we would like to further justify the choice of the name M Z for the phase that is antisym- metric in the gate voltages. Notice that it differs from the one used in Refs. 28, 33 -- 35, where it is referred to as the "Aharonov-Bohm" (AB) phase. Since the lat- ter suggests the phase induced by the vector potential associated to a magnetic flux penetrating the system,49 we prefer to avoid that denomination in the present sys- tem, which preserves time-reversal invariance. There is literature on the Aharonov-Bohm effect induced by the scalar potential,50 -- 52 which could eventually justify the use of that term in the present context. However, it is not clear in those systems whether the effect is due to the extra phase acquired by electrons according to the Aharonov-Bohm mechanism53,54 or simply due to the change of the electron trajectories50 or the electron density.52 For this reason, we find it more appropriate to use the name Mach-Zehnder with the aim of emphasiz- ing that the electron traverse the interferometer's arms in the same direction.29 Generally both phases φM Z and φF P are involved into the same interference process. 1. Current In Fig. 5, we show the behavior of the current at the Fermi energy in terminal l = 3 versus the two phases for the configuration V1 = V3 = 0 and V2 = V4 = V , and different values of the spin-flipping tunneling. As in the case of a single QPC, the different currents are related as I1 = I3, I2 = I4 and I3 = −I4. Interestingly, for this configuration of voltages, the in- terference is not effective for γf = 0. Thus, for vanishing spin-flip tunneling, the current in terminals l = 1, 3 is just I 0 3 = I 0 1 . As in the case of a single QPC, the reason for this behavior is the Pauli exclusion principle according to which both outgoing channels have to be equally pop- ulated if two incoming channels are equally populated. Therefore, the outgoing currents are not sensitive to gate voltages at this bias configuration. For γf 6= 0, the current through the interferometer becomes sensitive to gate voltages. Moreover, the two interference processes compete. In the upper panel we show that for vanishing φM Z , interference effects sys- tematically decrease the conductance through the ter- minal l = 3 bellow the conductance quantum e2/h. As functions of φF P it presents maxima (minima) at φF P = 2mπ, (φF P = (2m + 1)π), with m integer. In- stead, for φF P = 0, the l = 3 conductance as a function of φM Z achieves maxima equal to the conductance quantum e2/h for φM Z = (2m + 1)π and minima for φM Z = 2mπ which become deeper as γf increases. 2. Noise The corresponding behavior of the noise power at the terminal l = 3 is shown in Fig. 6. The perfect transmission through this terminal ob- served in Fig. 5 for γf = 0 is related to a vanishing noise, S3,3 = 0. As this tunneling parameter is switched on the noise power displays oscillations as a function of φM Z and φF P . In Fig. 6a we show the dependence on φF P of the noise power S3,3 for φM Z = 0. We identify a value γ∗ f 10 -2 -1 0 φF P /π 1 2 3 1 0,75 3 I 0,5 0,25 PSfrag replacements 0 -3 1 0,75 3 I 0,5 0,25 0 -3 PSfrag replacements -2 -1 0 φM Z /π 1 2 3 FIG. 5. (Color online) Current oscillations in terminal 3 ver- sus the Fabry-P´erot phase φF P = eL(VgT +VgB ) and versus the Mach-Zehnder phase φM Z = eL(VgB −VgT ) . The current is given in units of e2V /h. We set γp = 0.2 and γf = 0 (black solid curve), γf = 0.1 (red dashed- double dotted curve), γf = 0.2 (blue dotted curve) and γf = 0.5 (orange dashed- dotted curve). vF vF such that for small γf < γ∗ f , the noise power increases and has maxima (minima) at the phases where the conduc- tance has minima (maxima), i.e. for φF P = (2m + 1)π, (φF P = 2mπ) with m integer. For the parameters of the Fig. γ∗ f ∼ 0.25. This value of γf corresponds to the one for which the conductance of this terminal achieves the value G3 = I3/V = e2/(2h) for some values of φF P . The corresponding noise power at these points is the maxi- mum possible value S3,3 = e3V /(2h) and the noise power turn to have local minima for φF P = (2m + 1)π. For large enough γf , G3 < e2/(2h), ∀φF P and the noise power follows the pattern of maxima (for φF P = 2mπ) and minima (for φF P = (2m + 1)π) of the conductance. Fig. 6b shows the dependence on φM Z of the noise power S3,3 for φF P = 0. In this case, the noise vanishes for φM Z = (2m + 1)π, for which G3 = e2/h. For γf < γ∗ f the noise power displays maxima at φM Z = 2mπ, but this maxima turn to local minima for γf ≥ γ∗ f as a con- sequence of the fact that G3 < e2/(2h) for these values. The cross correlation between the currents through the 0,6 0,5 0,4 3 3 S 0,3 0,2 0,1 0 -3 PSfrag replacements 0,5 3 3 S 0,4 0,3 0,2 0,1 0 -3 PSfrag replacements -2 -1 0 φF P /π 1 2 3 -2 -1 0 φM Z /π 1 2 3 FIG. 6. (Color online) Shot-Noise oscillations of terminal l = 3 versus the Fabry-P´erot phase φF P = eL(VgT +VgB ) (φM Z = 0) and versus the Mach-Zehnder phase φM Z = eL(VgB −VgT ) (φF P = 0), for a spin-bias configuration V1 = V3 = 0 and V2 = V4 = V . The noise power is given in units of e3V /(2h). We set γp = 0.2 and γf = 0 (black solid curve), γf = 0.1 (red dashed- double dotted curve), γf = 0.2 (blue dotted curve) and γf = 0.5 (orange dashed-dotted curve). vF vF 11 0 -0,1 -0,2 3 2 S -0,3 -0,4 -0,5 -3 PSfrag replacements 0 -0,1 -0,2 3 2 S -0,3 -0,4 -0,5 -3 PSfrag replacements -2 -1 0 φF P /π 1 2 3 -2 -1 0 φM Z /π 1 2 3 FIG. 7. (Color online) Shot-Noise oscillations between termi- nals l = 2 and l = 3 versus the Fabry-P´erot phase φF P = eL(VgT +VgB ) and versus the Mach-Zehnder phase φM Z = vF eL(VgB −VgT ) vF , for a spin-bias configuration V1 = V3 = 0 and V2 = V4 = V . The noise power is given in units of e3V /(2h). We set γp = 0.2 and γf = 0 (black solid curve), γf = 0.1 (red dashed- double dotted curve), γf = 0.2 (blue dotted curve) and γf = 0.5 (orange dashed-dotted curve). Note that the cross-correlator of bottom terminals S1,4 = S2,3. top terminals are shown in Fig. 7. V. SUMMARY AND CONCLUSIONS Here we observe that the minima (maxima) of S2,3 always occur at φF P (M Z) = 2mπ. The maxima do not reach the bound e3V /(2h). Finally, in Fig. 8 the top- bottom correlation function S3,4 is showed. This noise is zero just for γf = 0, in contrast of S1,3 (or S2,4) which vanish for any value of γf . The absolute value of the top-bottom noise correlations for φM Z = 0, shown in the upper panel of Fig. 8, has minima (maxima) for φF P = 2mπ (φF P = (2m + 1)π). For φF P = 0 this quantity presents a more complicated structure as a function of φM Z as the tunneling parameter γf varies. For any value of this parameter, it vanishes at φM Z = (2m + 1)π. For γf < γ∗ f the absolute value S2,4 has maxima at φM Z = 2mπ, while for γf ≥ γ∗ f it has local minima for these values of φM Z . Unlike other cases, S2,4 does not reach the upper bound e3V /(2h) for any phase φM Z . We have presented a formal treatment based on non- equilibrium Green's function formalism to study the transport properties of interferometers of helical edge sates in bar geometries. We have derived expressions for the currents and we have defined transmission func- tions. The latter are the building blocks of the scattering matrix approach. In setups consisting of a small system connected to two or more particle reservoirs, the rela- tion between these two formalisms is well known since the work by Fisher and Lee41 for stationary transport. In Ref. 55 the suitable generalization to systems with ac driving was carefully analyzed. The case of currents flow- ing through edge states is special in the sense that the latter constitute reservoirs that support chiral currents. The contacts to metallic terminals just set the proper imbalance between them giving rise to a net current. In 0 -0,1 -0,2 4 3 S -0,3 -0,4 -0,5 -3 PSfrag replacements 0 -0,1 4 3 S -0,2 -2 -1 0 φF P /π 1 2 3 -3 -2 -1 PSfrag replacements 0 φM Z /π 1 2 3 , vF vF FIG. 8. (Color online) Shot-Noise oscillations between top l = 3 and bottom l = 4 terminals versus the Fabry-P´erot phase φF P = eL(VgT +VgB ) , and versus the Mach-Zehnder phase φM Z = eL(VgB −VgT ) for a spin-bias configuration V1 = V3 = 0 and V2 = V4 = V . We show the oscillations for distinct values of spin-flipping term γf = 0 (black solid curve),γf = 0.1 (red dashed- double dotted curve),γf = 0.2 (blue dotted curve),γf = 0.5 (orange dashed-dotted curve). The noise power is given in units of e3V /(2h). We set φM Z = 0 and γp = 0.2. For the rest of top-bottom corre- lators we have that S1,3 = S2,4 = 0 for any value of γf and S1,2 = S3,4. the present work, we have presented the generalization of Fisher and Lee's relation between the two formalisms for edge states currents (see sections III B, III C). We have also presented the expressions for the noise power in the Green's function formalism. We have used these formal tools to analyze the trans- port properties of helical electronic circuits containing one or two quantum point contacts, inducing tunnel- ing and scattering processes between the different edge states. Typically this type of effects are expected to take place preserving or flipping spin. We have focused on identifying the features induced in the transport proper- ties (currents and current-current correlation functions) originated in the spin-flip scattering processes at the QPC allowed by the symmetry of helical states. Impor- tantly, for the case of a single QPC, we have identified 12 a peculiar feature, the Pauli peak-dip structure in the current-current correlation functions, see Fig. 4, which allows to identify the presence or absence of a spin-flip scattering. For two QPC, following Refs. 28, 33 -- 35 we have assumed in the present work that applying gates within a finite region of the top and bottom edges in- duce additional potentials VgT and VgB within the he- lical state. These additional potentials in turn change the electron phase, which yields a rich structure of cur- rents and current-current correlation functions, with pat- terns of maxima and minima. This structure is the con- sequence of competing interference processes and has two important experimental outcomes. On one hand, the necessary condition for these features to exist it a non- vanishing value of the spin-flip tunneling. Thus, the ob- servation of these patterns is an important signature to identify and quantify the relevance of the spin-flip tun- neling. On the other hand, their observation would en- able to verify the idea that the electron phase in helical edge states can be locally controlled by this type of gates. That would be an important breakthrough, which would open a wide research avenue in the usage of solid state linear electronics for efficient quantum computation. In fact, with such phase shifters, quantum point contacts as wave splitters, single electron sources,56,57 and metal- lic contacts as detectors, all the necessary elements for quantum computation would be available in full analogy with quantum optics.58 VI. ACKNOWLEDGEMENTS LA thanks A-P Jauho for useful conversations. LA and BR thank support from CONICET, MINCyT and UBACYT, Argentina. Appendix A: Evaluating the Dyson's equation for the Green's functions The equation of motion (16) leads to the Dyson's equa- tion (DE) for the retarded Green's functions (15). The four elements with edge indices ασ and ασ can be orga- nized in 2 × 2 matrices as follows Gr(x, x′, ω) = g0,r(x, x′, ω) +Xj,j ′ Gr(x, xj , ω) × Σ0,r(xj , xj ′ , ω)g0,r(xj ′ , x′, ω), (A1) with Gr(x, x′, ω) = (cid:18) Gr Gr ασ,ασ(x, x′, ω) Gr ασ,ασ(x, x′, ω) Gr ασ,ασ(x, x′, ω) ασ,ασ(x, x′, ω) (cid:19) , and g0,r(x, x′, ω) = (cid:18) g0,r 0 ασ (x, x′, ω) (A2) 0 ασ (x, x′, ω) (cid:19) . g0,r (A3) The elements of the above matrix are the retarded Green's function for the uncoupled edge where Vg,ασ = Vg,T if ασ = R ↑, L ↓ and Vg,ασ = Vg,B if ασ = R ↓, L ↑. 13 g0,r ασ (x, x′, ω) = g0,r k (ω) = 1  Z +k0 −k0 dkeik(x−x′)g0,r k (ω), 1 ω − (vαk + eVg,ασ) + iη , (A4) which in the limit of k0 → ∞ results g0,r ασ (x, x′, ω) = − i vF  Θ (sα(x − x′)) × exp(cid:20)i 1 vF  (ω − eVg,ασ)(x − x′)(cid:21),(A5) The self energy in (A1) is defined by back substituting the DE for the Green's functions with indices ασ, ασ and ασ,ασ: and Gr ασ,ασ(x, x′, ω) = Xj Gr ασ,ασ(x, x′, ω) = Xj 2vF [γpGr ασ,ασ(x, xj , ω) + sαγf Gr ασ,ασ(x, xj , ω)]g0,r ασ (xj , x′, ω), 2vF [γpGr ασ,ασ(x, xj , ω) + sαγf Gr ασ,ασ(x, xj , ω)]g0,r ασ (xj , x′, ω). The result is F × Σ0,r(xj , xj ′ , ω) = 42v2   pg0,r γ2 γf γphsαg0,r ασ (xj , xj ′ , ω) + sαg0,r ασ (xj, xj ′ , ω) + γ2 f g0,r ασ (xj, xj ′ , ω) ασ (xj , xj ′ , ω)i γf γphsαg0,r f g0,r γ2 ασ (xj , xj ′ , ω) + sαg0,r ασ (xj, xj ′ , ω)i ασ (xj , xj ′ , ω) + γ2 pg0,r ασ (xj , xj ′ , ω)   . (A6) The DEs for the lesser Green's functions can be derived from Eqs. (A1), (A6) and (A6) by recourse to Langreth rules,39 according to which given a product of retarded Green's functions ArBr, then (AB)< = A<Ba + ArB<. In the case of (A1) the result is M G<(x, x′, ω) = Xj ′,j ′′=1(cid:2)Λ0,r(x, xj ′ , ω) + δ(x − xj ′ )(cid:3) g0,<(xj ′ , xj ′′ , ω)[δ(xj ′′ − x′) + Λ0,a(xj ′′ , x′, ω)] Xj,j ′=1 Gr(x, xj , ω)Σ0,<(xj , xj ′ )Ga(xj ′ , x′, ω), + M (A7) with Λ0,r(x, xj ′ , ω) = Pj Gr(x, xj , ω)Σ0,r(xj, xj ′ , ω) = [Λ0,a(xj ′ , x, ω)]†, while ασ (x, x′, ω) = ifα,σ(ω)ρ0 g0,< ασ(x, x′, ω) = i(cid:2)g0,r ρ0 ασ(x, x′, ω), ασ (x, x′, ω) − g0,a ασ (x, x′, ω)(cid:3)  Z dke−ik(x−x′)δ(ω − ǫασ). = 1 (A8) Here ǫασ = sαvF k + eVg,ασ, with sα = ±1, α = R, L and σ =↑, ↓, while fα,σ(ω) = f (ω − µα,σ), and µα,σ is the chemical potential of the reservoir from where the electrons are injected. α,↓(x, x′, ω) = Gr Gr bridization term is ΓL↑(R↓) = 4vF γ2 α,↑(x, x′, ω), p(f ). α = L, R. The hy- 14 Appendix B: Retarded Green's function for a single QPC The Dyson equation (A1) for a single QPC at x1 can be solved resulting, Gr R,↑(x, x1, ω) = G0 R,↑(x, x1, ω) 1 − ΣR R,↑(x1, x1, ω)G0 R,↑(x1, x1, ω) . Using Eq.(A4) and Eq.(A6) we find (B1) Appendix C: Retarded Green's function for two QPCs Gr R,↑(x, x1, ω) = − iΘ(x − x1) vF  vF ei ω 1 + γ2 (x−x1) p + γ2 f . (B2) The other components are calculated from the fol- R,↑(x′, x, ω) and lowing relations Gr L,↑(x, x′, ω) = Gr We consider two point contacts at x1 and x2, with x1 < x2.In the general case (γp(f ) 6= 0), the equations for the two spin species are coupled and the solution for x > x2 of the diagonal for the R, ↑ movers reads Gr R↑(x, x1; ω) = i vF  (x−x1)(ω−eVg,T ) ie vF ∆ h−1 + γ2 h1 + γ2 f (cid:16)1 + 2e p (cid:16)1 + 2e iL vF  (eVg,T −eVg,B )(cid:17)i , vF  (2ω−eVg,T −eVg,B )(cid:17)i , iL (C1) p + γ2 f + γ2 Gr R↑(x, x2; ω) = − ie i vF  (x−x2)(ω−eVg,T ) vF ∆ with ∆ = 4γ2 x < x1, we have pei L vF (2ω−eVgT −eVgB ) + (1 + γ2 p + γ2 f )2. For Gr L↓(x, x1; ω) = − Gr L↓(x, x2; ω) == ie− i vF  (x−x1)(ω−eVg,T ) vF ∆ ie− i vF  (x−x2)(ω−eVg,T ) vF ∆ The other spin diagonal components are calculated changing Vg,T → Vg,B. Appendix D: Transmission functions Since the system we are dealing is time-reversal, the transmission functions fulfill Tα,β = Tβ,α. Hereinafter we present the expressions for all transmission functions in distinct cases (single and double QPC), see Ref. 28. 1. Single QPC In the case where is a single QPC, we have the following transmission functions, f + γ2 h1 + γ2 h−1 + γ2 iL p (cid:16)1 + 2e f (cid:16)1 + 2e vF  (2ω−eVg,T −eVg,B )(cid:17)i , vF  (eVg,T −eVg,B )(cid:17)i . iL p + γ2 T1(2),3(4) = T2(1),3(4) = T4(1),3(2) = 4γ2 f p + γ2 f )2 , f )2 p + γ2 f )2 , p + γ2 (1 + γ2 (−1 + γ2 (1 + γ2 4γ2 p p + γ2 (1 + γ2 f )2 . (C2) (D1) 15 2. Two QPC In the case with two QPC, the transmission functions 8γ2 f [−1 + γ2 p]2[1 + cos φM Z ] f + γ2 ∆2 , are: T1(2),3(4)= T2(1),3(4)= T4(1),3(2)= 1 ∆2 (cid:2)(−1 + γ2 f + γ2 p)4 +16γ4 8γ2 f − 8γ2 p[1 + γ2 f (−1 + γ2 f + γ2 ∆2 f + γ2 p)2 cos φM Z(cid:3) , − φF P )] p]2[1 + cos( 2ωL vF ,(D2) where ∆ = 4γ2 pei L vF and (2ω−eVgT −eVgB ) + (1 + γ2 p + γ2 f )2, φM Z = φF P = eL(Vg,B − Vg,T ) vF  eL(Vg,B + eVg,T ) vF  . (D3) 1 S. Murakami, N. Nagaosa, and S.-C. Zhang, Physical Re- view Letters 93, 156804 (2004). 2 B. A. Bernevig, T. L. Hughes, and S.-C. Zhang, Science 314, 1757 (2006). 3 M. Z. Hasan and C. L. Kane, Rev. Mod. Phys. 82, 3045 Daughton, S. von Moln´ar, M. L. Roukes, A. Y. Chtchelka- nova, and D. M. Treger, Science 294, 1488 (2001). 21 C.-Y. Y. Hou, E.-A. Kim, and C. Chamon, Physical Re- view Letters 102, 076602 (2009). 22 A. Strom and H. Johannesson, Physical Review Letters (2010). 102, 096806 (2009). 4 C. L. Kane and E. J. Mele, Physical Review Letters 95, 23 J. Teo and C. L. Kane, Physical Review B 79, 235321 226801 (2005). 5 C. Wu, B. A. Bernevig, and S.-C. Zhang, Physical Review Letters 96, 106401 (2006). 6 C. Xu and J. Moore, Physical Review B 73, 045322 (2006). 7 C. L. Kane and E. J. Mele, Physical Review Letters 95, 146802 (2005). 8 J. Maciejko, X.-L. Qi, and S.-C. Zhang, Phys. Rev. B 82, 155310 (2010). 9 J. Maciejko, C. Liu, Y. Oreg, X.-L. Qi, C. Wu, and S.-C. Zhang, Physical Review Letters 102, 256803 (2009). 10 Y. Tanaka, A. Furusaki, and K. Matveev, Physical Review Letters 106, 236402 (2011). 11 K. Hattori, J. Phys. Soc. Jpn. 80, 124712 (2011). 12 M. Konig, S. Wiedmann, C. Brune, A. Roth, H. Buhmann, L. W. Molenkamp, X. L. Qi, and S.-C. Zhang, Science 318, 766 (2007). 13 M. Konig, H. Buhmann, L. W. Molenkamp, T. Hughes, C.-X. Liu, X.-L. Qi, and S.-C. Zhang, J. Phys. Soc. Jpn. 77, 031007 (2008). 14 A. Roth, C. Brune, H. Buhmann, L. W. Molenkamp, J. Maciejko, X.L. Qi, and S.-C. Zhang, Science 325, 294 (2009). 15 M. Buttiker, Science 325, 278 (2009). 16 B. I. Halperin, Physical Review B 25, 2185 (1982). 17 M. Buttiker, Physical Review B 38, 9375 (1988). 18 K. Klitzing, G. Dorda, and M. Pepper, Physical Review Letters 45, 494 (1980). 19 G. A. Prinz, Science 282, 1660 (1998). 20 S. A. Wolf, D. D. Awschalom, R. A. Buhrman, J. M. (2009). 24 T. L. Schmidt, Physical Review Letters 107, 096602 (2011). 25 C. P. Orth, G. Strubi, and T. L. Schmidt, arXiv:1305.1875. 26 G. Dolcetto, S. Barbarino, D. Ferraro, N. Magnoli, and M. Sassetti, Physical Review B 85, 195138 (2012). 27 P. Sternativo, and F. Dolcini, arXiv: 1308.2423. 28 F. Dolcini, Physical Review B 83, (2011). 29 Y. Ji, Y. Chung, D. Sprinzak, M. Heiblum, D. Mahalu, and H. Shtrikman, Nature 422, 415 (2003). 30 B. van Wees, L. P. Kouwenhoven, C. J. P. M. Harmans, J. Williamson, C. Timmering, M. Broekaart, C. Foxon, and J. G. E. Harris, Physical Review Letters 62, 2523 (1989). 31 F. Camino, W. Zhou, and V. Goldman, Physical Review B 76, 155305 (2007). 32 P. Delplace, J. Li, and M. Buttiker, Physical Review Let- ters 109, 246803 (2012). 33 R. Citro, F. Romeo, and N. Andrei, Physical Review B 84, 161301(R) (2011). 34 F. Romeo, R. Citro, D. Ferraro and M. Sassetti, Phys. Rev. B 86, 165418 (2012). 35 D. Ferraro, G. Dolcetto, R. Citro, F. Romeo, and M. Sas- setti, Physical Review B 87, 245419 (2013). 36 C. Chamon and E. Fradkin, Phys. Rev. B 56, 2012 (1997). 37 Y. M. Blanter and M. Buttiker, Physics Reports 336, 1 (2000). 38 B. Rizzo, L. Arrachea and J. P. Paz, Phys. Rev. B 85, 045442 (2012). 39 See H. Haug and A. P. Jauho, Quantum Kinetics in Trans- port and Optics of Semiconductors (Springer Series in Solid-State Sciences, Vol. 123, Springer-Verlag, Berlin Hei- delberg, 1996). 40 M. Buttiker, Physical Review B 46, 12485 (1992). 41 D.S. Fisher and P.A. Lee, Phys. Rev. B 23, 6851 (1981). 42 L. Arrachea and M. Moskalets, Phys. Rev. B 74, 245322 (2006). 43 G. F`eve, P. Degiovanni, and T. Jolicoeur, Physical Review B 77, 035308 (2008). 44 M. Moskalets and M. Buttiker, Physical Review B 83, 035316 (2011). 45 M. V. Moskalets, Scattering matrix approach to non- stationary quantum transport (Imperial College Press, Lon- don, 2011). 46 J. M. Edge, J. Li, P. Delplace, and M. Buttiker, Physical Review Letters 110, 246601 (2013). 47 S. Ol'khovskaya, J. Splettstoesser, M. Moskalets, and M. Buttiker, Physical Review Letters 101, 166802 (2008). 48 E. Bocquillon, V. Freulon, J.-M. Berroir, P. Degiovanni, B. Pla¸cais, A. Cavanna, Y. Jin, and G. F`eve, Science 339, 1054 (2013). 16 49 R. Webb, S. Washburn, C. Umbach, and R. Laibowitz, Physical Review Letters 54, 2696 (1985). 50 S. Washburn, H. Schmid, D. Kern, and R. Webb, Physical Review Letters 59, 1791 (1987). 51 A. Van Oudenaarden, M. H. Devoret, Y. V. Nazarov, and J. E. Mooij, Nature 391, 768 (1998). 52 P. de Vegvar, G. Timp, P. Mankiewich, R. Behringer, and J. Cunningham, Physical Review B 40, 3491 (1989). 53 Y. Aharonov and D. Bohm, Phys. Rev. 115, 485 (1959). 54 Y. Aharonov and D. Bohm, Phys. Rev. 123, 1511 (1961). 55 L. Arrachea and M. Moskalets, Phys. Rev. B 74, 245322 (2006). 56 G. F`eve, A. Mah´e, J.-M. Berroir, T. Kontos, B. Pla¸cais, D. C. Glattli, A. Cavanna, B. Etienne, and Y. Jin, Science 316, 1169 (2007). 57 M. D. Blumenthal, B. Kaestner, L. Li, S. P. Giblin, T. J. B. M. Janssen, M. Pepper, D. Anderson, G. A. C. Jones, and D. A. Ritchie, Nature Physics 3, 343 (2007). 58 E. Knill, R. Laflamme, and G. J. Milburn, Nature 409, 46 (2001). 

1504.04790

2

1504

Influence of atomic tip structure on the intensity of inelastic tunneling spectroscopy data analyzed by combined scanning tunneling spectroscopy, force microscopy and density functional theory

Achieving a high intensity in inelastic scanning tunneling spectroscopy (IETS) is important for precise measurements. The intensity of the IETS signal can vary up to a factor three for various tips without an apparent reason accessible by scanning tunneling microscopy (STM) alone. Here, we show that combining STM and IETS with atomic force microscopy enables carbon monoxide front atom identification, revealing that high IETS intensities for CO/Cu(111) are obtained for single atom tips, while the intensity drops sharply for multi-atom tips. Adsorbing the CO molecule on a Cu adatom [CO/Cu/Cu(111)] such that it is elevated over the substrate strongly diminishes the tip dependence of IETS intensity, showing that an elevated position channels most of the tunneling current through the CO molecule even for multi-atom tips, while a large fraction of the tunneling current bypasses the CO molecule in the case of CO/Cu(111).

cond-mat.mes-hall

cond-mat

Influence of atomic tip structure on the intensity of inelastic tunneling spectroscopy data analyzed by combined scanning tunneling spectroscopy, force microscopy and density functional theory Norio Okabayashi1,2,*, Alexander Gustafsson3, Angelo Peronio1, Magnus Paulsson3, Toyoko Arai2, and Franz J. Giessibl1 1Institute of Experimental and Applied Physics, University of Regensburg, D-93053 Regensburg, Germany 2Graduate School of Natural Science and Technology, Kanazawa University, 920-1192 Ishikawa, Japan 3Department of Physics and Electrical engineering, Linnaeus University, 391 82 Kalmar, Sweden * okabayashi@staff.kanazawa-u.ac.jp PACS numbers: 68.37.Ef, 68.37.Ps, 73.23.-b Abstract Achieving a high intensity in inelastic scanning tunneling spectroscopy (IETS) is important for precise measurements. The intensity of the IETS signal can vary up to a factor three for various tips without an apparent reason accessible by scanning tunneling microscopy (STM) alone. Here, we show that combining STM and IETS with atomic force microscopy enables carbon monoxide front atom identification, revealing that high IETS intensities for CO/Cu(111) are obtained for single atom tips, while the intensity drops sharply for multi-atom tips. Adsorbing the CO molecule on a Cu adatom [CO/Cu/Cu(111)] such that it is elevated over the substrate strongly diminishes the tip dependence of IETS intensity, showing that an elevated position channels most of the tunneling current through the CO molecule even for multi-atom tips, while a large fraction of the tunneling current bypasses the CO molecule in the case of CO/Cu(111). 1 Inelastic electron tunneling spectroscopy (IETS) with scanning tunneling microscopy (STM) is an effective method to analyze the vibrational modes of a single adsorbed molecule with sub-nanometer lateral resolution [1,2]. The vibrational energy of a molecule on a substrate strongly depends on the surrounding environment, such as the substrate structure and composition [3]. By studying these subtle changes of the vibrational energy using STM-IETS with a molecular functionalized tip, it has been demonstrated that STM-IETS can provide information on the inner structure of a molecule [4,5] similar to atomic force microscopy (AFM) [6]. These advantages of STM-IETS have accelerated research in related fields [7-16]. Owing to recent progress in the theoretical description of IETS [17-22], the qualitative understanding has been improved considerably, where the symmetry of the wave functions of a tip and a molecule on a substrate and a vibrational mode of the molecule is predicted to influence the efficiency of the inelastic process (inel) for the tunneling current involving the molecule. In order to discuss inel from the intensity of IETS we have to consider that IETS intensity is described by the multiplication of the two factors: (1) the ratio of the tunneling current passing through a molecule to the total tunneling current (Imolecule/Itotal) and (2) the efficiency of the inelastic process (inel). These factors should in principle be affected by the geometrical structure of the substrate and the tip. The geometrical structure of a metal tip apex can be determined by using carbon monoxide (CO) front atom identification (COFI) provided by AFM [23,24], where the tip apex of a force sensor is probed by a CO molecule that stands upright on a metal surface (inset of Fig. 1(e)). The metallic tip apex atom has a dipole moment induced by the Smoluchowski effect [25], whose direction is the same as that of the CO molecule [26]. Thus in the distance regime where the electrostatic interaction between the tip and the molecule dominates, the force between them is attractive. When the tip is scanned over the CO molecule, this attractive force appears as a dip (smaller value) in the frequency shift image for each atom at its apex, i.e., the number of the attractive force minima provides the number of atoms composing the tip apex [27]. In this paper, we have investigated the tip-structure dependent IETS for individual CO molecules on a Cu surface by combining STM and AFM. We have found that a tip with a single atom on its apex (single-atom tip) gives a stronger IET signal compared with a blunt tip consisting of four atoms on its apex (four-atom tip) for a CO molecule on a Cu(111) surface. However, the intensity between the two tips becomes comparable when a Cu adatom is inserted between the CO molecule and the Cu(111) substrate. From these findings, we will 2 discuss the opposite electrode geometry dependent inelastic efficiency (inel) and demonstrate the validity of the modern IETS theory. The experiments are carried out in an ultra-high vacuum low-temperature (4.4 K) combined STM and AFM (LT-STM/AFM, Omicron Nanotechnology, Taunusstein, Germany). A Cu(111) surface was cleaned by repeated sputtering and annealing before adsorbing CO molecules on it. The force acting between a CO molecule and the apex of the metallic tip is measured by a qPlus sensor [28]. The sensor whose stiffness is k=1800 N/m oscillates at f0=47375 Hz with a constant amplitude of 20 pm during all STM/AFM measurements. When an average force gradient <kts> acts between the tip and the CO molecule, the sensor frequency is shifted by f=f0<kts>/2k [29]. The current <It> is averaged for many cycles of the sensor oscillation, since the bandwidth of the current amplifier is small compared to f0. In order to measure the conductance (dI/dV) and IET signal (d2I/dV2), a modulation voltage (2338.7 Hz, 1 mVrms) is added to the sample bias and the first and second harmonics in the current are detected by a lock-in amplifier (HF2LI, Zurich Instruments, Zürich, Switzerland). Throughout the whole text, the IET signal is normalized with the differential conductance, i.e., IETS=(d2I/dV2)/(dI/dV) [10,14,18,20]. The tip is formed from an etched tungsten wire, cleaned by field evaporation and repeatedly poked into the Cu substrate to prepare various tip apexes [23,24,30]. The repeated poking processes probably cover the tip apex with Cu atoms. The poking processes also scatter Cu adatoms on the Cu(111) surface which are employed as another target by adsorbing a CO molecule and as an opposite electrode for the CO-functionalized tip [31]. Calculations of the IETS are performed with the density functional theory program Siesta [32]. From the relaxed geometries obtained by Siesta, elastic transport properties are calculated by attaching electrodes with TranSiesta [33]. Vibrational frequencies and IETS are obtained from the TranSiesta calculations using the post-processing package Inelastica [34]. The Siesta (TranSiesta) calculations utilize a supercell consisting of a 7 (17) layer thick 4×4 Cu slab together with the CO-molecule and a pyramidal tip modeled by 4 Cu atoms on the reverse side of the slab. The computations were performed using the following parameters: Perdew-Burke-Ernzerhof functional, 400 Ry real space cutoff, 4×4 k-points and a DZP (SZP) basis set for CO (Cu). The IETS calculations are then performed in the gamma-point. Fig. 1(a)[(c)] shows a constant-height current image of a CO molecule on the Cu(111) surface obtained by a single-atom tip [a four-atom tip], as confirmed by the simultaneously acquired f images (Fig. 1(b)[(d)]). For both tips we see the dip in the current image at the 3 position of the CO molecule [35,36], where the current on top of the CO molecule is larger for the four-atom tip than for the single-atom tip owing to the larger tip area from which electrons can tunnel. Fig. 1(e) shows the IETS for CO molecules by the single-atom tip [37] and the four-atom tip, where identical current set-points are used for both tips. As described in the Appendix A, a background IETS measured on the copper surface is subtracted from that on the CO molecule. In the case of the single-atom tip, the frustrated translational (FT; ~4 meV) and frustrated rotational (FR; ~35 meV) modes of the CO molecule [7-9] are clearly seen in its IETS. However, the IETS intensity acquired with the four-atom tip is considerably smaller than that acquired with the single-atom tip: 65% decrease for both the FT and FR modes. A reduced IETS intensity is also observed for tips with two atoms and three atoms on its apex. The strong intensity of the IETS provided by single-atom tips is confirmed by preparing different tips, which by COFI measurements (Fig. 2(a)) are single-atom tips. Cross-sections of constant-height current image are shown in Fig. 2(b) [38]. Note that tip #1 is the one used in Fig 1(a) and (b). In the case of tips #1 through #5, the minimum current acquired on the CO molecules is almost identical and the value is 24% of that on the Cu surface, thus these single-atom tips are judged to be sharp. The normalized IETS is also consistent for the different sharp single-atom tips (tips #1 to #5) (Fig. 2(c))(Appendix A). On the other hand, the IETS intensity with a single-atom tip having secondary-atoms outside of the apex which can contribute to the tunneling, is considerably weaker (Appendix A). This decrease originates from the increased fraction of tunneling electrons that do not interact with CO molecule and pass directly between the tip and the substrate. The decrease in IETS intensity for the four-atom tip can be similarly rationalized as a decreased ratio of tunneling current involving the CO molecule to the total current (ICO/Itotal). To investigate how the inelastic efficiency (inel) depends on the geometry, we now present IETS measurements for a system where the tunneling current is dominantly passing through a CO molecule. Figure 3(a) shows IETS for a CO molecule on a Cu adatom on the Cu(111) surface (Appendix A) with the single-atom tips [37] and the four-atom tip used in Fig. 1(e), where the current set-point is identical. In this case, the IETS intensity of the frustrated rotational (FR) mode with the four-atom tip is 20% smaller than the case of the single-atom tip, however, their overall intensity is comparable. Raising the vertical position of the CO molecule by one Cu adatom removes the direct tunneling channel between tip and substrate, causing almost all the current to pass through the CO molecule. The similarity of the intensity between the single-atom tips and four-atom tip indicates that inel does not depend on the tip electrode 4 geometry investigated here. The same conclusion can be derived from the IETS with a CO-functionalized tip over (1) a Cu adatom and (2) the bare Cu(111) surface [39]. In the two situations investigated in Fig. 3(b), tunneling electrons are emitted from or injected to the single-atom on which the CO molecule is adsorbed, thus the electron beam is focused and ICO/Itotal is expected to be large and similar. The similarity of the IETS intensity between two cases again indicates that the structure of the opposite electrode such as the Cu adatom and Cu plane does not strongly influence the inel for a CO molecule on the tip apex. The role of the electrode opposite to a CO molecule in inel has been further investigated theoretically by using the TranSiesta code [33]. The computations utilize a localized basis set causing the calculated tunneling current to preferentially pass through the CO molecule regardless of the opposite electrode geometry [35], i.e., the direct tunneling between the tip and the substrate is underestimated. Thus the calculated IETS dominantly reflects the contribution of inel rather than that of ICO/Itotal. Fig. 4(a) shows the calculated IETS for a CO on a Cu (111) with the single- and three- atom tip. Here we can see that the intensity of IETS for the rotational mode is almost identical between two tips, which support our conclusion that the inel does not depend on the opposite electrode geometry. The conclusion that inel does not depend on the opposite electrode geometry can be rationalized considering the symmetry of the tips states with respect to the molecular axis [18-20]. In the case of the CO molecule, the two-fold degenerate  symmetric molecular states dominantly contribute to the inelastic tunneling process (Fig. 4(b)), because these states are more localized on the O molecule than that of the  symmetric state [19] (Appendix B). Taking into account that the FT and FR modes have a  symmetric character, the tip state with  symmetry should effectively contribute to the inelastic tunneling process (Fig. 4(b)) (Appendix B). The relative contribution of the  state to the total transmission is 50% for the three-atom tip and 57% for the single-atom tip, i.e., the contribution of the  state drops about 12% from a single- to a three-atom tip, resulting in an almost identical inel in the calculation. In contrast to the metallic tips investigated here, for the case of a CO functionalizes tip, the symmetry of the tip state is drastically changed from  to , which results in the inversion of the STM image contrast for a CO molecule on the Cu(111) surface from dip to bump [31,35]. This change of the tip state is predicted to decrease the efficiency of the IETS considerably for a CO molecule on a metal surface [19] in contrary to the present case. In summary, by combining STM and AFM, we have investigated the dependence of the IETS intensity on the structure of the tip electrode for individual CO molecules for several Cu 5 substrates. We have found that for the system where the current dominantly pass through the CO molecule by positioning this molecule on top of a Cu adatom, the IETS intensity is almost identical regardless of the tip geometry. This result indicates that the inelastic tunneling efficiency (inel) is independent on the geometry of the tip electrode at least for a metallic tip. This conclusion demonstrates the validity of the modern IETS theory based on density functional theory and nonequilibrium Green's functions [18-20]. While we have found that single-atom tips provide a maximal IETS intensity and great reproducibility, single-atom tip are more reactive than multi-atom tips [23,24,40]. Therefore multi-atom tips may still be useful in cases where a high intensity is not key but a low perturbation to the vibrating molecule by the force field of a tip is desired. We are deeply indebted to Thomas Frederiksen, Aran Garcia-Lekue and Alfred. J. Weymouth for stimulating discussions, to Daniel Meuer and Florian Pielmeier for the sample preparation and sensor construction and for Jascha Repp for various advices including a method to improve the resolution of IETS. This study was partially supported by a funding (SFB 689) from Deutsche Forschungsgemeinschaft (F.J.G); by JSPS "Strategic Young Researcher Overseas Visits Program for Accelerating Brain Circulation" (T.A. and N.O.); by a Grant-in-Aid for Young Scientists (B) (25790055) from MEXT (N.O.); by a grant from the Swedish Research Council (621-2010-3762) (A.G. and M. P.). Appendix A: Complete set of IETS with related STM images Figure 5 displays IETS data from five different tips. All these tips are single-atom tips, although none of them shows a COFI image that is perfectly symmetric with respect to rotations around the z-axis. Instead, the COFI images show slight asymmetries that could be attributed to a slight tilt of the plane of the second atomic layer of the tip. Nevertheless, the IETS spectra are essentially identical after background subtraction. Decreased IETS for the single-atom, blunt tip is shown in Fig. 6 with the data by single-atom sharp tips, where the tip apex geometry is confirmed by COFI (Fig. 6(a)) and the sharpness of the tip apex is confirmed by the constant-height current measurement (Fig. 6(b)). When the tip is blunt, i.e., the tip constitutes a single-atom on its apex but has secondary-atoms outside of the apex (Fig. 6(c)), the IETS intensity is considerably decreased (Fig. 6(d)). Complete set of the IETS for the CO molecule on Cu adatom and Cu substrate are shown in Fig. 7: IETS for (a) [(b)] CO/Cu(111) and (d) [(e)] CO/adatom with the single-atom tip #1 6 [#2], and for (c) CO/Cu(111) and (f) CO/Cu adatom with the four-atom tip. The topographic image of a CO on a Cu adatom is shown in Fig. 8 with the image of a CO and a Cu adatom on the Cu(111) surface. Appendix B: Detail of the theoretical IETS Theoretical IETS between two different set-points are shown in Fig. 9, where we see that the intrinsic IETS depends weakly on the tip apex geometry. Three most transmitting eigenchannels are shown in Fig. 10 for (a) the single-atom and (b) the three-atom tip, whose contribution to inelastic tunneling process for the FT and FR modes is summarized in table 1. We see that the contribution of the sigma state at the tip to the transmission is similar between the single-atom (57%) and three atom tip (50%), which results in the similar intrinsic IETS intensity. The calculated vibrational energies are summarized in table 2. 7 References [1] W. Ho, J. Chem. Phys. 117, 11033 (2002). [2] B. C. Stipe, M. A. Rezaei, and W. Ho, Science 280, 1732 (1998). [3] H. J. Lee and W. Ho, Phys. Rev. B 61, R16347 (2000). [4] C. L. Chiang, C. Xu, Z. Han, and W. Ho, Science 344, 885 (2014). [5] P. Hapala, R. Temirov, F. S. Stefan Tautz, and P. Jelínek, Phys. Rev. Lett. 113, 226101 (2014). [6] L. Gross, F. Mohn, N. Moll, P. Liljeroth, and G. Meyer, Science 325, 1110 (2009). [7] L. J. Lauhon and W. Ho, Phys. Rev. B 60, R8525 (1999). [8] A. J. Heinrich, C. P. Lutz, J. A. Gupta, and D. M. Eigler, Science 298, 1381 (2002). [9] L. Vitali et al., Nano Lett. 10, 657 (2010). [10] N. Okabayashi, M. Paulsson, H. Ueba, Y. Konda, and T. Komeda, Phys. Rev. Lett. 104, 077801 (2010). [11] N. Okabayashi, M. Paulsson, H. Ueba, Y. Konda, and T. Komeda, Nano Lett. 10, 2950 (2010). [12] N. Okabayashi, M. Paulsson, and T. Komeda, Prog. Surf. Sci. 88, 1 (2013). [13] H. Gawronski and K. Morgenstern, Phys. Rev. B 89, 125420 (2014). [14] K. J. Franke, G. Schulze, and J. I. Pascual, J Phys Chem Lett 1, 500 (2010). [15] M. Grobis et al., Phys. Rev. Lett. 94, 136802 (2005). [16] K. Motobayashi, Y. Kim, H. Ueba, and M. Kawai, Phys. Rev. Lett. 105, 076101 (2010). [17] N. Lorente and M. Persson, Phys. Rev. Lett. 85, 2997 (2000). [18] M. Paulsson, T. Frederiksen, H. Ueba, N. Lorente, and M. Brandbyge, Phys. Rev. Lett. 100, 226604 (2008). [19] A. Garcia-Lekue, D. Sanchez-Portal, A. Arnau, and T. Frederiksen, Phys. Rev. B 83, 155417 (2011). [20] E. T. R. Rossen, C. F. J. Flipse, and J. I. Cerda, Phys. Rev. B 87, 235412 (2013). [21] G. Teobaldi, M. Penalba, A. Arnau, N. Lorente, and W. A. Hofer, Phys. Rev. B 76, 235407 (2007). [22] A. Troisi and M. A. Ratner, Phys. Rev. B 72, 033408 (2005). [23] J. Welker and F. J. Giessibl, Science 336, 444 (2012). [24] T. Hofmann, F. Pielmeier, and F. J. Giessibl, Phys. Rev. Lett. 112, 066101 (2014). [25] R. Smoluchowski, Phys. Rev. 60, 661 (1941). [26] M. Schneiderbauer, M. Emmrich, A. J. Weymouth, and F. J. Giessibl, Phys. Rev. Lett. 8 112, 166102 (2014). [27] M. Emmrich et al., Science 348, 308 (2015). [28] F. J. Giessibl, Appl. Phys. Lett. 76, 1470 (2000). [29] F. J. Giessibl, Phys. Rev. B 56, 16010 (1997). [30] M. Emmrich et al., Phys. Rev. Lett. 114. 146101 (2015). [31] L. Bartels, G. Meyer, and K. H. Rieder, Appl. Phys. Lett. 71, 213 (1997). [32] J. M. Soler et al., J. Phys.: Condens. Matter 14, 2745 (2002), [33] M. Brandbyge, J. L. Mozos, P. Ordejon, J. Taylor, and K. Stokbro, Phys. Rev. B 65, 165401 (2002). [34] T. Frederiksen, M. Paulsson, M. Brandbyge, and A. P. Jauho, Phys. Rev. B 75, 205413 (2007). [35] A. Gustafsson and M. Paulsson, arXiv:1512.00702v1. [36] R. K. Tiwari, D. M. Otalvaro, C. Joachim, and M. Saeys, Surf. Sci. 603, 3286 (2009). [37] The data is an average of IETS by the tip in Fig.1(b) (tip #1 in Fig. 2(a)) and the tip #2 in Fig. 2(a), since these two tips are adopted for IETS measurements of CO/Cu/Cu(111) in addition to CO/Cu(111). - [38] The current on the Cu(111) surface is slightly modulated depending on the substrate position owing to the standing waves. The averaged current <It> in Fig. 2(b) is normalized such that the current on the leftmost position is 1.5 nA. [39] The tip before the CO functionalization has a single-atom on its apex, and is sharp in the sense of Fig. 2. [40] T. Hofmann, F. Pielmeier, and F. J. Giessibl, Phys. Rev. Lett. 115, 109901 (2015). 9  Fig. 1. (color online) Constant-height, (a)[(c)] current and (b)[(d)] frequency shift images (1.5 nm×1.5 nm) for a CO molecule adsorbed on Cu(111) by a single-atom tip [four-atom tip]. The tip height is set on the Cu(111) substrate at a sample bias Vt=−1 mV and an average current <It>=1.5 nA. (e) Normalized IETS for a CO molecule at a set-point of Vt=−50 mV and <It>=5 nA, where the IETS on the Cu(111) surface is subtracted. Fig. 2. (color online) (a) COFI images (1.5nm×1.5nm) and (b) cross-sections of the constant-height current images on a CO molecule with different single-atom (tip #1 to #5) tips at the set-point of Vt=−1 mV and <It>=1.5 nA on the Cu(111) surface. As a reference, the cross-section of the current image with the four-atom tip in Fig. 1(c) is added in (b). (c) IETS for CO molecules with the same tips where the set-point on a CO molecule is Vt=−50 mV and <It>=5 nA. 10  Fig. 3. (color online) (a) IETS with the single-atom tips (red) and the four-atom tip (blue) for a CO on a Cu adatom. (b) IETS of a CO-functionalized tip for a Cu adatom (red) and the bare Cu(111) surface (blue). In both cases [(a) and (b)], the tip-height is set at Vt=−50 mV and <It>=5 nA on the measurement points. Fig. 4. (color online) (a) Calculated IETS with a single-atom tip (red) and three-atom tip (blue) for a CO on a Cu(111), where the tips is positioned such that the calculated elastic current is nearly identical (Vt=−50 mV and It≈5 nA). Broadening by the modulation voltage (1.0 mVrms) and temperature (4.4 K) is included. (b) Most important tip (upper panels) and molecular (lower panels) scattering states for the inelastic scattering by the rotational and translational modes. 11  Fig. 5 (color online) Reproducibility of IETS spectra for five different tips #1 to #5. The data is the same as that shown in Fig. 2(c), but we display the spectra individually including background subtraction here. The top row shows the constant height COFI (carbon monoxide front atom identification) frequency shift profiles (in Hz) for the five different tips. The center row displays the IETS signal in color and the background spectra in black. The bottom row displays the net IETS signal without background. 12  Fig. 6. (color online) (a) COFI images (1.5nm×1.5nm) and cross-sections of the constant-height current images for a CO molecule with various single-atom tips (tip #6 to #11). The set-point on the Cu(111) surface is Vt=−1 mV and <It>=1.5 nA for the COFI measurements and Vt=−1 mV and <It>=1 nA for current measurements. (c) Schematic images of the single-atom sharp and blunt tips. (d) IETS for CO molecules with the tip #6 to #11 where the set-point on a CO molecule is Vt=−50 mV and <It>=10 nA. Note that the modulation voltage used here is 3.5 mVrms, which is larger than the value adopted for the case of the main text (1 mVrms). 13  Fig. 7. (color online) Set of IETS on the CO molecule and the Cu(111) substrate (upper panel) and after the background subtraction (lower panel). Experimental conditions: the single-atom tip (tip #1[#2]) to (a)[(b)] CO/Cu(111) and (d)[(e)] CO/Cu/Cu(111); the four-atom tip to (c) CO/Cu(111) and (f) CO/Cu/Cu(111). For the background measurements, IETS on the Cu(111) surface are measured and averaged for 16 different points. A set-point of Vt=−50 mV and <It>=5 nA on the CO molecule has been used for all measurements. 14  Fig. 8. (color online) Constant-current images of a Cu adatom and a CO molecule on a Cu adatom by a single-atom tip (Vt=−10 mV, <It>=100 pA). The inset shows a CO molecule on the Cu(111) surface. Fig. 9. (color online) Calculated IETS data between a single-atom tip and a three-atom tip are displayed for two different set-points where the transmission acquired by TranSiesta is almost identical between the single-atom- and three-atom tip: Vt=−50 mV, It≈5 nA and Vt=−50 mV, It≈1 nA . Note that Fig. 4(a) in the manuscript shows the data for the former set-point . 15  Fig. 10. (color online) Isosurface plot of the three most transmitting eigenchannels between the CO molecule on the Cu(111) surface and (a) the single-atom tip [(b) the three-atom tip] at a set-point of Vt=−50 mV, It≈5 nA. The top (bottom) row of scattering states originates from the tip (substrate) side. 16 Table. 1. Average (over two degenerated vibration) partial contribution (%) of each scattering state (see Fig. 9) in the inelastic process of the FT and FR modes with the elastic transmission for the single-atom and three-atom tips at a set-point of Vt=−50 mV, It≈5 nA. 1 atom tip T1() T2() T3() FT (%) FR (%) S1() S2() S3() S1() S2() S3() Transmission ×10-4 Transmission (%) 0.19 46.65 44.64 0.27 48.00 46.32 9.00 57 1.57 0.01 0.01 2.00 0.02 0.02 3.37 22 1.63 0.10 0.07 1.95 0.12 0.07 3.29 21 3 atom tip T1() T2() T3() FT (%) FR (%) S1() S2() S3() S1() S2() S3() Transmission ×10-4 Transmission (%) 0.09 45.17 43.78 0.16 37.63 35.37 8.49 50 3.36 0.16 0.08 6.22 0.11 0.31 4.27 25 3.33 0.08 0.15 6.53 0.20 0.85 4.20 25 Table. 2. Vibrational energy (meV) of FT and FR modes at a set-point of Vt=−50 mV, It≈5 nA, which are calculated by allowing the CO molecule and the one (three) Cu atoms of the 1 (3)-atom tip to move (dynamical region). FT1 4.59 4.25 FT2 4.61 4.28 FR1 31.6 31.7 FR2 31.8 31.9 1atom 3atom tip tip 17 

1703.00184

2

1703

Unexpectedly High Cross-plane Thermoelectric Performance in Layered Carbon Nitrides

Organic thermoelectric (TE) materials create a brand new perspective to search for high-efficiency TE materials, due to their small thermal conductivity. The overlap of pz orbitals, commonly existing in organic {\pi}-stacking semiconductors, can potentially result in high electronic mobility comparable to inorganic electronics. Here we propose a strategy to utilize the overlap of pz orbitals to increase the TE efficiency of layered polymeric carbon nitride (PCN). Through first-principles calculations and classical molecular dynamics simulations, we find that A-A stacked PCN has unexpectedly high cross-plane ZT up to 0.52 at 300 K, which can contribute to n-type TE groups. The high ZT originates from its one-dimensional charge transport and small thermal conductivity. The thermal contribution of the overlap of pz orbitals is investigated, which noticeably enhances the thermal transport when compared with the thermal conductivity without considering the overlap effect. For a better understanding of its TE advantages, we find that the low-dimensional charge transport results from strong pz-overlap interactions and the in-plane electronic confinement, by comparing {\pi}-stacking carbon nitride derivatives and graphite. This study can provide a guidance to search for high cross-plane TE performance in layered materials.

cond-mat.mes-hall

cond-mat

*Manuscript Click here to view linked References Unexpectedly High Cross-plane Thermoelectric Performance of Layered Carbon Nitrides Zhidong Ding1,#, Meng An1,2,#, Shenqiu Mo1,3, Xiaoxiang Yu1,3, Zelin Jin1,3, Yuxuan Liao4, Jing-Tao Lü5,*, Keivan Esfarjani6, Junichiro Shiomi4,7,*, Nuo Yang1,3,* 1 State Key Laboratory of Coal Combustion, Huazhong University of Science and Technology, Wuhan 430074, China 2 College of Mechanical and Electrical Engineering, Shaanxi University of Science and Technology, Xi'an, 710021, China 3 Nano Interface Center for Energy (NICE), School of Energy and Power Engineering, Huazhong University of Science and Technology, Wuhan 430074, China 4 Department of Mechanical Engineering, The University of Tokyo, 7-3-1 Hongo, Bunkyo, Tokyo 113-8656, Japan 5 School of Physical and Wuhan National High Magnetic Field Center, Huazhong University of Science and Technology, Wuhan 430074, China 6 Department of Mechanical and Aerospace Engineering, University of Virginia, Charlottesville 22904, United States 7 Center for Materials Research by Information Integration, National Institute for Materials Science, 1-2-1 Sengen, Tsukuba, Ibaraki 305-0047, Japan # Z.D. and M.A. contributed equally to this work. * To whom correspondence should be addressed. E-mail: (J.-T.L.) jtlu@hust.edu.cn, (J.S.) shiomi@photon.t.u-tokyo.ac.jp, (N.Y.) nuo@hust.edu.cn Abstract: Organic thermoelectric (TE) materials create a brand new perspective to search for high-efficiency TE materials, due to their small thermal conductivity. The overlap of pz orbitals, commonly existing in organic π-stacking semiconductors, can potentially result in high electronic mobility comparable to inorganic electronics. Here we propose a strategy to utilize the overlap of pz orbitals to increase the TE efficiency of layered polymeric carbon nitride (PCN). Through first-principles calculations and classical molecular dynamics simulations, we find that A-A stacked PCN has unexpectedly high cross-plane ZT up to 0.52 at 300 K, which can contribute to n-type TE groups. The high ZT originates from its one-dimensional charge transport and small thermal conductivity. The thermal contribution of the overlap of pz orbitals is investigated, which noticeably enhances the thermal transport when compared with the thermal conductivity without considering the overlap effect. For a better understanding of its TE advantages, we find that the low-dimensional charge transport results from strong pz-overlap interactions and the in-plane electronic confinement, by comparing π-stacking carbon nitride derivatives and graphite. This study can provide a guidance to search for high cross-plane TE performance in layered materials. Introduction Thermoelectric (TE) technology can directly convert heat into electricity as well as being used in solid-state cooling. Thermoelectric devices are considered for a variety of applications [1 -- 5], for example, distributed power generation, waste heat recovery from vehicles, and power supply to electronic devices. The efficiency of TE conversion can be measured by a dimensionless figure of merit ZT that is defined as . is the electrical conductivity; is the Seebeck coefficient; is the absolute temperature; is the electronic thermal conductivity; is the lattice thermal conductivity. Reducing while maintaining the power factor (PF= ), is a general strategy to increase ZT. Such methods have been demonstrated to be effective in reducing , for example, reducing dimensions to bring in size effects on phonon scattering [6,7], creating nanostructures to intensify boundary scatterings [8 -- 10], and changing morphology in a micron scale [10 -- 12]. These strategies aim at designing electron-crystal-phonon-glass structures. In addition, lower-dimensional band structures can augment PF to further boost ZT [13,14]. Organic TE materials provide an alternative blueprint for TE application due to their small [15 -- 19]. Compared with inorganic TE materials, organic TE materials have advantages, for example, light weight, high flexibility, and low cost. Thermal conductivity of organic materials is generally below 1 Wm-1K-1. Thus conducting polymers, for example, PEDOT and P3HT, are considered as materials with high TE potential [20,21]. Their TE performance can be improved through doping or optimizing molecular morphology [22,23]. The ZT of PEDOT:PSS is up to 0.42 at 300K from experimental measurements [20]. Another type of promising TE candidates is small- molecular semiconductors that are easier to be purified and crystallized than polymers. In small-molecular semiconductors, strong intermolecular electronic couplings provide main paths for their salient charge transport [24]. The mobility of single-crystal rubrene [25] can be up to 43 cm2V-1s-1. Pentacene thin films doped with iodine [17] have of )κT/(κσSZTphe2σSTeκphκphκ2σSphκphκσ110 S/cm. This excellent electronic transport and poor thermal transport make small- molecular semiconductors competitive in TE applications. More interestingly, the charge transport in three-dimensional (3D) structures of small-molecular semiconductors is strongly anisotropic, where an inter-planar charge transport dominates. Thus TE performance along a non-bonds direction is worthy of exploration. Recent investigations have shown that Cn-BTBTs and bis(dithienothiophene) (BDT) have high ZT along the inter-planar direction due to their low-dimensional charge transport [26,27]. Overall, low and the special electronic properties make stacking organic semiconductors become promising candidates in TE applications. Polymeric carbon nitrides (PCN) have recently attracted a considerable interest as crystalline all-organic wide-bandgap semiconductors in solar water-splitting [28] and photovoltaics [29]. Tyborski et al experimentally found that their samples of PCN mainly consist of flat and buckled A-A-stacked structures with a stacking distance of 3.28 Å [30]. The flat structure is shown in Figure 1(a) and Figure 1(b). Merschjann et al demonstrated that PCN exhibit a one-dimensional (1D) inter-planar charge transport, which means that the charge transport in PCN is essentially confined to move along one dimension, the c direction of the crystal structure [31]. This low-dimensional charge transport property is similar to BDT, where the 1D band structure of BDT results from a pz-orbital overlap and is considered to be responsible for its high ZT [27]. Similarly, a previous model of π-stacked molecular junctions has been reported to possess high TE coefficients [32,33]. For this reason, we are intended to investigate cross-plane TE performance of PCN under the influence of the pz-orbital overlap. Another reason is the low-dimensional advantage of their charge transport. Common small-molecular semiconductors possess herringbone structures that result in two-dimensional (2D) charge transport [34]. The A-A stacked PCN has an 1D charge transport that is preferable in TE applications [35]. In this work, we calculate TE coefficients of A-A stacked PCN along the c direction phκusing density function theory (DFT) combined with Boltzmann transport equation (BTE). Its lattice thermal conductivity is calculated by molecular dynamics simulations. To accurately predict the lattice thermal conductivity of A-A stacked PCN along the c direction, we consider van der Waals (vdW) interaction and electrostatic forces to describe the overlap of pz orbitals. Methods The lattice and electronic structure of PCN are calculated using the Vienna Ab Initio Simulation Package (VASP) [36,37]. The Perdew−Burke−Ernzerhof (PBE) version of the generalized gradient approximation (GGA) is used for the exchange−correlation functional [38]. The DFT-D2 method of Grimme is used to take into account the van der Waals (vdW) interactions among different layers [39]. The cutoff energy of the plane wave expansion is set to 600 eV. The reciprocal space is sampled by a 9×9×15 Monkhorst-Pack k-mesh. A finer mesh of 11×11×31 is used for density of state (DOS). Other computational details are provided in Supporting Information (SI). The electronic transport coefficients are calculated with BoltzTraP [40] that is based on the semi-classical Boltzmann transport theory. Different from an assumption of constant relaxation time, we calculate electronic relaxation time via a deformation potential approximation (DPA) that is proposed by Bardeen and Shockley [22,26,41,42]. The longitudinal acoustic phonon mode is considered in electron-phonon (e-ph) coupling along the direction of electronic flow. As PCN has 1D electronic transport properties, we use a 1D formula for relaxation time that is expressed as follows. In the formula, is the Boltzmann constant; T is the absolute temperature; is the deformation potential constant; is the reduced Planck constant; is the elastic constant along the c direction; is the group velocity. Details of formula derivations k2BvCTEk)kτ(1BkECkvand transport coefficient calculations are provided in SI. The thermal conductivity of PCN is calculated from equilibrium molecular dynamics (EMD) using the Green-Kubo formula, where κ, V, and T are the thermal conductivity, the volume of simulation cell, and temperature, respectively. is the heat current autocorrelation function (HCACF). The angular bracket denotes ensemble average. In this study, all the MD simulations are performed using LAMMPS packages [43]. For such layered structured PCN, the overlap of pz orbitals introduces a new electronic transport channel. For thermal transport, AMBER force field [44] is used because it has been shown to successfully reproduce thermal transport of layered structures. Previous investigations demonstrated that the geometry of the overlap of pz orbitals are controlled by electrostatic forces and vdW interactions [45,46]. To accurately predict the lattice thermal conductivity between interlayers, both are considered. The charge distribution of PCN is computed by an atomic charge calculator, which is based on the Electronegativity Equalization Method (EEM) [47 -- 51]. The periodic boundary condition is used in three directions to simulate the infinite system. However, the calculated thermal conductivity still exhibits the finite size effects related to the variation in the longest phonon wavelength that the system can reproduce. In our simulations, 6×6×10 unit cells (100.35×76.11×7.28 nm3) are used, where the calculated thermal conductivity saturates to a constant. More details about EMD simulations can be found in the SI. All the results reported in our studies are ensemble averaged over 12 independent runs with different initial conditions. 02Bdτ(0)J) (τJVT3k1κ(0)J) (τJResults and discussion From Figure 2(a), for the conduction band (CB), we observe the band along Г−Z−Г is much more dispersive than other directions, as well as the bandwidth. Then we use the 1D tight-binding model to fit the CB (seen in SI), and obtain the hopping energy along the c direction up to 0.224 eV. This value is close to hopping energy of hole in BDT, and thus indicates an equivalent strong electronic coupling along the c direction. . Comparatively, the band Г−X and Г−Y directions is flat, which indicates a smaller electronic hopping energy and thus smaller in-plane electronic couplings. On the other hand, from the projected band structure and the partial density of states, we can learn that the strong coupling originates from the overlap of vertical pz orbitals. Meanwhile, the charge density of CB in Figure 2(b) shows that nitrogen atoms at the top corners of triangles do not contribute pz orbitals to CB. That is, electrons are confined within the a-b plane but preferable to move along the c direction. This feature is consistent with the conclusion from the reference that PCN has a predominantly inter-planar electronic transport [31]. The predominantly inter-planar, or 1D-like electronic transport property of PCN is advantageous in TE applications [52]. The density of states (DOS) of PCN also show a large asymmetry around the Fermi level. It indicates large Seebeck coefficients of PCN as Seebeck coefficients involve an integral over eigenvalues around the Fermi level. With the knowledge of the inter-planar electronic transport of PCN, we have explored another two similar types of stacked carbon nitrides, g-CN1 and g-CN2 [53], lattice structures of which are shown in Figure 1(c)-1(f). We calculate their band structures that are shown in Figure 2(e) and Figure 2(f). The CB and valence band (VB) of both band structures are more dispersive along Г−M and Г−K than those along Г−Α, which indicates stronger in-plane electronic couplings thus larger electrical conductivities along in-plane directions, compared with those along the out-of-plane direction. That is to say, the electronic transport along in-plane directions dominates in g-CN1 and g-CN2. Therefore, we focus on PCN in following TE performance calculations.  We calculate TE coefficients of PCN along the c direction, using the band model and the Boltzmann transport equation. To validate our band model for charge transport in organic materials, we calculate hopping rates of the CB of PCN, and the reorganization energy of melem monomers that consist of PCN. Details are provided in SI. The hopping rates of the CB are comparable to the reorganization energy, so the band model is acceptable [24,54]. Figure 3(a)-3(d) show the TE coefficients as a function of doping concentration. Seebeck coefficients under 300 K decrease from 50 to 500 μV/K when the carrier concentration increases from 1019 to 1021 cm-3. These Seebeck values are comparable to inorganic materials, for example, Silicon and single- layer MoS2 that have ZT maximum in the same concentration range and have Seebeck up to 300 μV/K under 300 K [8,55]. The electrical conductivity of PCN increases when the carrier concentration increases. In the range between 1019 and 1021 cm-3, the cross-plane electrical conductivity in PCN has an order of magnitude of 105 S/m. We calculated the intrinsic mobility of PCN that is about 8.55 cm2V-1s-1 for electrons under 300K, which give a reasonable explanation for its high electrical conductivity. This mobility value is high for organic materials but not restricted to PCN. As Troisi suggests [56], if the reorganization energy is similar to or smaller than rubrene (the inner part is 159 meV [57]), and the vdW interaction between atoms is involved in the frontier orbitals of neighboring molecules, essentially any highly purified crystalline molecular material may have a sufficiently large mobility in the 1 -- 50 cm2V-1s-1 range. It indicates that π- stacking organic materials with low reorganization energy values have a favorable cross-plane charge transport. In organic materials, polaron theories and band theories can be both used to describe interactions between charges and the structure deformation [56,58]. The band theories, such as polaronic bands and electronic bands can successfully predict the negative mobility dependence on the temperature [59]. In addition, the band theories are able to predict the high mobility that is consistent with experiments. Shuai et al adopted the electronic band model to calculate electronic mobility of pentacene and rubrene that agrees well with experimental measurements [60 -- 62]. In terms of the band theories that are widely used in inorganic semiconductors, electron-phonon couplings are an important contribution to the electronic relaxation time. In our calculations, we have considered the contribution of acoustic phonon modes. Specially, if the polarity of materials is strong, the contribution of polar optical phonon (POP) scattering can be also important, such as GaAs and GaN [63,64].In these materials, the electronegativity is a key parameter to evaluate the polarity. Back to the carbon nitrides, since the electronegative difference between carbon and nitrogen is larger than that between gallium and arsenic, and we can find that in Figure 2(b), the electron density distribution of PCN is not symmetric, we believe that the polarity of PCN and its POP scattering effects make our calculations overestimate the relaxation time and the electrical conductivity of PCN. Figure 3(b) shows that the thermal conductivity of PCN, electronic part κe and lattice part κph. Within the range from 1020 to 1021 cm-3, κe is comparable to or even larger than κph. As for phonon transport, the lattice thermal conductivity is 0.77 Wm-1K-1 at 300 K. To clarify the contribution of the overlap of pz orbitals to lattice thermal conductivity, other simulations are performed just considering vdW interactions between interlayers. It is found that the lattice thermal conductivity is 0.55 W-m-1K-1 at 300 K within the range (0.1 -- 1.0 m-1K-1) [65 -- 69] of the common amorphous polymers due to the weak intermolecular interactions of vdW nature (more details in SI). Therefore, compared with the weak vdW interactions between interlayers, the overlap of pz orbitals in PCN does noticeably enhance the thermal transport. With the electronic and phononic properties, we obtain ZT profiles of PCN that are shown in Figure 3(c). The optimal ZT values at the room temperature are 0.52 with an n-type carrier concentration of 1.23 × 1020 cm-3 and 0.28 with a p-type carrier concentration of 5.45 × 1020 cm-3. At the optimal carrier concentration of electrons, the PF is 17.92 μWcm-1K-2 and the total thermal conductivity is 1.03 Wm-1K-1. The p-type thermoelectric performance is overestimated, as the valence band does not exhibit a large bandwidth along the Г−Z direction. The hole hopping energy along the c direction are much smaller than the electronic one, indicating that the band model may lose its validity and overestimate the conductivity of holes. PCN can be doped as n-type materials and contribute to the n-type TE group. Figure 3(d) shows maximum ZT values from 100 to 400 K. The ZT of PCN has the same order of magnitude as typical herringbone stacked molecules. The calculated optimal ZT of pentacene and rubrene are 1.8 and 0.6 at room temperature, respectively [60]. Herringbone stacked BDT is predicted to have peculiar 1D band structure and an optimal ZT of 1.48 at room temperature [27]. Due to a capability for TE conversion along the cross-plane direction, PCN can be distinguished from layered materials, specifically inorganics materials such as, graphene, MoS2 and black phosphorus, and thus regarded as a complementary material to 2D TE materials. This feature does not only exist in PCN, as layered inorganic heterostructures also exhibit high ZT along the perpendicular directions. Esfarjani et al studied TE performance of vdW graphene/phosphorene/graphene heterostructures with the predicted ZT of 0.13 under 600K [70]. Similarly, graphene/MoS2 vdW heterostructures are calculated to have ZT as large as 2.8 at room temperature [71]. The phonon transport along the cross-plane direction is highly suppressed in both structures. Compared with these nanoscale heterostructures, PCN is a bulk material and should be easier to prepare in experiments and to scale up. Before closing, we combine PCN, g-CN1, g-CN2 and A-A stacked graphite, to search for the possible lattice structure that materials can have to obtain high cross- plane ZT. The cross-plane direction corresponds to the small lattice thermal conductivity because of no covalent bonds and suppressed lattice vibrations, so strong couplings of electrons or holes along the cross-plane direction are preferable for high ZT values. In organic materials, the layer distance and the stacking motif are considered to affect the couplings [72]. For g-CN1 and g-CN2, they have nearly the same layer distances as PCN, but their band structures exhibit weak electronic couplings between neighboring layers. Our calculated charge density of g-CN1 and g-CN2 shows a consistent conclusion in Figure 4. We found that pz orbitals of the CB in Figure 4(a) and 4(c) indicate the existence of the overlap of pz orbitals. But due to their A-B stacked structures, the coupling is much weaker than PCN as the coupling distance is nearly twice as that in PCN. The pz orbitals of the VB in Figure 4(b) and 4(d) lie horizontally, showing that holes cannot transfer effectively along the cross-plane direction. So A-A stacked motif or a small coupling distance will be essential for strong inter-planar electronic couplings and the excellent cross-plane TE performance. Besides, the in-plane electronic confinement is necessary. For A-A stacked graphite, the stacking motif and the electronic coupling distance are nearly the same as PCN. However, electrons and holes are more likely to move within the plane of graphite, because pz orbitals in graphite form delocalized π bonds in plane. In contrast, the in- plane electronic confinement is caused in PCN because the nitrogen atoms connecting heptazine-like units contribute no pz orbitals of the CB. Thus, the pz orbitals of PCN are discontinuously distributed and π bonds are localized along in-plane directions. In such case, charges are more likely to transport from one layer to another and form the 1D- like charge transport. Overall, we conclude that small electronic coupling distances along cross-plane direction and the in-plane confinement effect are responsible for promising cross-plane TE performance. Conclusions We investigate TE performance of A-A stacked PCN using first-principles calculations and classical molecular dynamics simulations. We find that PCN has a high ZT up to 0.52 at 300 K along the c direction, which can contribute to n-type TE groups. The overlap of pz orbitals and the in-plane electronic confinement induce a 1D charge transport and large cross-plane power factors that are comparable to silicon and single- layer MoS2. The lattice thermal conductivity contributed by the overlap of pz orbitals is considered and found to enhance the thermal transport. By comparing three carbon nitrides and A-A stacked graphite, we find that small electronic coupling distances along cross-plane direction and the in-plane confinement effect induce the 1D charge transport and TE advantages. This study can benefit the pursuing of a high cross-plane ZT in other layered materials. Acknowledgments The work were supported by the National Natural Science Foundation of China, Grant No. 51576076 (NY), 51711540031 (NY), 61371015 (JL), Hubei Provincial Natural Science Foundation of China (2017CFA046), and NSFC-JSPS cooperative projects, Grant No. 51711540031 (NY and JS). The authors are grateful to Dr. Jinyang Xi, Prof. Zhigang Shuai and Prof. Gang Zhang for useful discussions. The authors thank the National Supercomputing Center in Tianjin (NSCC-TJ) and China Scientific Computing Grid (ScGrid) for providing assistance in computations. References [1] L.E. Bell, Cooling, Heating, Generating Power, and Recovering Waste Heat with Thermoelectric Systems, Science (2008) 1457 -- 1461. doi:10.1126/science.1158899. ). 321 (80-. [2] A.J. Minnich, M.S. Dresselhaus, Z.F. Ren, G. Chen, Bulk nanostructured thermoelectric materials: current research and future prospects, Energy Environ. Sci. 2 (2009) 466 -- 479. doi:10.1039/b822664b. [3] Y. Lan, A.J. Minnich, G. Chen, Z. Ren, Enhancement of thermoelectric figure- of-merit by a bulk nanostructuring approach, Adv. Funct. Mater. 20 (2010) 357 -- 376. doi:10.1002/adfm.200901512. [4] A. Shakouri, Recent Developments in Semiconductor Thermoelectric Physics and Materials, Annu. Rev. Mater. Res. 41 (2011) 399 -- 431. doi:10.1146/annurev- matsci-062910-100445. [5] M. Zebarjadi, K. Esfarjani, M.S. Dresselhaus, Z.F. Ren, G. Chen, Perspectives on thermoelectrics: from fundamentals to device applications, Energy Environ. Sci. 5 (2012) 5147 -- 5162. doi:10.1039/c1ee02497c. [6] A.I. Boukai, Y. Bunimovich, J. Tahir-Kheli, J.-K. Yu, W.A.G. Iii, J.R. Heath, Silicon nanowires as efficient thermoelectric materials, in: Mater. Sustain. Energy A Collect. Peer-Reviewed Res. Rev. Artic. from Nat. Publ. Gr., 2011: pp. 116 -- 119. [7] A.I. Hochbaum, R. Chen, R. Diaz Delgado, W. Liang, E.C. Garnett, M. Najarian, A. Majumdar, P. Yang, Enhanced thermoelectric performance of rough silicon nanowires, Nature. 451 (2008) 163 -- 167. doi:10.1038/nature06381. J.-H. Lee, G.A. Galli, J.C. Grossman, Nanoporous Si as an Efficient Thermoelectric Material, 3750 -- 3754. doi:10.1021/nl802045f. (2008) Nano [8] Lett. 8 [9] Y. Tian, M.R. Sakr, J.M. Kinder, D. Liang, M.J. Macdonald, R.L.J. Qiu, H.-J. Gao, X.P.A. Gao, One-Dimensional Quantum Confinement Effect Modulated Thermoelectric Properties in InAs Nanowires, Nano Lett. 12 (2012) 6492 -- 6497. doi:10.1021/nl304194c. [10] L. Yang, N. Yang, B. Li, Extreme Low Thermal Conductivity in Nanoscale 3D Si Phononic Crystal with Spherical Pores, Nano Lett. 14 (2014) 1734 -- 1738. doi:10.1021/nl403750s. J.-W. Jiang, N. Yang, B.-S. Wang, T. Rabczuk, Modulation of Thermal Conductivity in Kinked Silicon Nanowires: Phonon Interchanging and Pinching Effects, Nano Lett. 13 (2013) 1670 -- 1674. doi:10.1021/nl400127q. [11] [12] Q. Song, M. An, X. Chen, Z. Peng, J. Zang, Y. Nuo, Adjustable thermal resistor by reversibly folding a graphene sheet, Nanoscale. 8 (2016) 14943 -- 14949. doi:10.1039/c6nr01992g. [13] H. Ohta, S. Kim, Y. Mune, T. Mizoguchi, K. Nomura, S. Ohta, T. Nomura, Y. Nakanishi, Y. Ikuhara, M. Hirano, H. Hosono, K. Koumoto, Giant thermoelectric Seebeck coefficient of a two-dimensional electron gas in SrTiO3, Nat. Mater. 6 (2007) 129 -- 134. doi:10.1038/nmat1821. [14] M. Zebarjadi, K. Esfarjani, Z. Bian, A. Shakouri, Low-temperature thermoelectric power factor enhancement by controlling nanoparticle size distribution, Nano Lett. 11 (2011) 225 -- 230. doi:10.1021/nl103581z. [15] O. Bubnova, X. Crispin, Towards polymer-based organic thermoelectric generators, Energy Environ. Sci. 5 (2012) 9345 -- 9362. doi:10.1039/c2ee22777k. [16] J. Yang, H.-L. Yip, A.K.-Y. Jen, Rational Design of Advanced Thermoelectric Materials, Adv. Energy Mater. 3 (2013) 549 -- 565. doi:10.1002/aenm.201200514. [17] Q. Zhang, Y. Sun, W. Xu, D. Zhu, Organic Thermoelectric Materials: Emerging Green Energy Materials Converting Heat to Electricity Directly and Efficiently, Adv. Mater. 26 (2014) 6829 -- 6851. doi:10.1002/adma.201305371. [18] Y. Chen, Y. Zhao, Z. Liang, Solution processed organic thermoelectrics: towards fl exible thermoelectric modules As featured in: Solution processed organic thermoelectrics: towards flexible thermoelectric modules, Energy Environ. Sci. 8 (2015) 401 -- 422. doi:10.1039/c4ee03297g. [19] C. Li, H. Ma, Z. Tian, Thermoelectric properties of crystalline and amorphous polypyrrole: A computational study, Appl. Therm. Eng. 111 (2017) 1441 -- 1447. doi:10.1016/j.applthermaleng.2016.08.154. [20] G.-H. Kim, L. Shao, K. Zhang, K.P. Pipe, Engineered doping of organic semiconductors for enhanced thermoelectric efficiency, Nat. Mater. 12 (2013) 719 -- 723. doi:10.1038/NMAT3635. [21] Q. Zhang, Y. Sun, W. Xu, D. Zhu, Thermoelectric energy from flexible P3HT films doped with a ferric salt of triflimide anions, Energy Environ. Sci. 5 (2012) 9639 -- 9644. doi:10.1039/c2ee23006b. [22] W. Shi, T. Zhao, J. Xi, D. Wang, Z. Shuai, Unravelling Doping Effects on PEDOT at the Molecular Level: From Geometry to Thermoelectric Transport Properties, 12929 -- 12938. doi:10.1021/jacs.5b06584. J. Am. Chem. Soc. 137 (2015) [23] S. Qu, Q. Yao, L. Wang, Z. Chen, K. Xu, H. Zeng, W. Shi, T. Zhang, C. Uher, L. Chen, Highly anisotropic P3HT films with enhanced thermoelectric performance via organic small molecule epitaxy, NPG Asia Mater. 8 (2016) e292. doi:10.1038/am.2016.97. [24] V. Coropceanu, J. Cornil, D.A. Da, S. Filho, Y. Olivier, R. Silbey, J.-L. Brédas, Charge Transport in Organic Semiconductors, Chem. Rev. 107 (2006) 926 -- 952. doi:10.1021/cr050140x. [25] M. Yamagishi, J. Takeya, Y. Tominari, Y. Nakazawa, T. Kuroda, S. Ikehata, M. Uno, T. Nishikawa, T. Kawase, High-mobility double-gate organic single-crystal transistors with organic crystal gate insulators, Appl. Phys. Lett. 90 (2007) 182117. doi:10.1063/1.2736208. [26] W. Shi, J. Chen, J. Xi, D. Wang, Z. Shuai, Search for Organic Thermoelectric Materials with High Mobility: The Case of 2,7-Dialkyl[1]benzothieno[3,2- b][1]benzothiophene Derivatives, Chem. Mater. 26 (2014) 2669 -- 2677. doi:10.1021/cm500429w. [27] X.-Y. Mi, X. Yu, K.-L. Yao, X. Huang, N. Yang, J.-T. Lu , Enhancing the Thermoelectric Figure of Merit by Low-Dimensional Electrical Transport in Phonon-Glass doi:10.1021/acs.nanolett.5b01491. Crystals, Nano Lett. 15 (2015) 5229 -- 5234. [28] X. Wang, K. Maeda, A. Thomas, K. Takanabe, G. Xin, J.M. Carlsson, K. Domen, M. Antonietti, A metal-free polymeric photocatalyst for hydrogen production from water under visible (2009) 76 -- 80. doi:10.1038/NMAT2317. light, Nat. Mater. 8 [29] J. Xu, T.J.K. Brenner, L. Chabanne, D. Neher, M. Antonietti, M. Shalom, Liquid- Based Growth of Polymeric Carbon Nitride Layers and Their Use in a Mesostructured Polymer Solar Cell with Voc Exceeding 1 V, J. Am. Chem. Soc. 136 (2014) 13486 -- 13489. doi:10.1021/ja508329c. [30] T. Tyborski, C. Merschjann, S. Orthmann, F. Yang, M.-C. Lux-Steiner, T. Schedel-Niedrig, Crystal structure of polymeric carbon nitride and the determination of its process-temperature-induced modifications Related content, J. Phys. Condens. Matter. 25 (2013) 395402. doi:10.1088/0953- 8984/25/39/395402. [31] C. Merschjann, S. Tschierlei, T. Tyborski, K. Kailasam, S. Orthmann, D. Hollmann, T. Schedel-Niedrig, A. Thomas, S. Lochbrunner, Complementing Graphenes: 1D Interplanar Charge Transport in Polymeric Graphitic Carbon Nitrides, Adv. Mater. 27 (2015) 7993 -- 7999. doi:10.1002/adma.201503448. [32] G. Kiršanskas, Q. Li, K. Flensberg, G.C. Solomon, M. Leijnse, Designing π- stacked molecular structures to control heat transport through molecular junctions, Therm. Conduct. through Mol. Wires J. Chem. Phys. 105 (2014) 233102. doi:10.1063/1.1603211. [33] Q. Li, M. Strange, I. Duchemin, D. Donadio, G.C. Solomon, A Strategy to Suppress Phonon Transport in Molecular Junctions Using π‑Stacked Systems, J. Phys. Chem. C. 121 (2017) 7175 -- 7182. doi:10.1021/acs.jpcc.7b02005. [34] H. Dong, X. Fu, J. Liu, Z. Wang, W. Hu, 25th Anniversary Article: Key Points for High-Mobility Organic Field-Effect Transistors, Adv. Mater. 25 (2013) 6158 -- 6183. doi:10.1002/adma.201302514. I.D. Hicks, M.S. Dresselhaus, Thermoelectric figure of merit of a one- dimensional 16631. doi:10.1103/PhysRevB.47.16631. conductor, (1993) Phys. Rev. [35] B. 47 [36] G. Kresse, J. Furthmüller, Efficient iterative schemes for ab initio total-energy calculations using a plane-wave basis set, Phys. Rev. B. 54 (1996) 11169. doi:10.1103/PhysRevB.54.11169. [37] G. Kresse, D. Joubert, From ultrasoft pseudopotentials to the projector 1758. Phys. Rev. B. (1999) 59 augmented-wave method, doi:10.1103/PhysRevB.59.1758. [38] J.P. Perdew, K. Burke, M. Ernzerhof, Generalized Gradient Approximation Made Simple, Phys. Rev. Lett. 77 (1996) 3865. doi:10.1103/PhysRevLett.77.3865. [39] S. Grimme, Accurate description of van der Waals complexes by density functional theory including empirical corrections, J. Comput. Chem. 25 (2004) 1463 -- 1473. doi:10.1002/jcc.20078. [40] G.K.H. Madsen, D.J. Singh, BoltzTraP. A code for calculating band-structure dependent quantities, Comput. Phys. Commun. 175 doi:10.1016/j.cpc.2006.03.007. (2006) 67 -- 71. [41] J. Bardeen, W. Shockley, Deformation Potentials and Mobilities in Non-Polar Crystals, Phys. Rev. 80 (1950) 72 -- 80. doi:10.1103/PhysRev.80.72. [42] J. Xi, M. Long, L. Tang, D. Wang, Z. Shuai, First-principles prediction of charge mobility in carbon and organic nanomaterials, Nanoscale. 4 (2014) 4348 -- 4369. doi:10.1039/c2nr30585b. [43] S. Plimpton, Fast parallel algorithms for short-range molecular dynamics, J. Comput. Phys. 117 (1995) 1 -- 19. doi:10.1006/jcph.1995.1039. [44] W.D. Cornell, P. Cieplak, C.I. Bayly, I.R. Gould, K.M. Merz, D.M. Ferguson, D.C. Spellmeyer, T. Fox, J.W. Caldwell, P.A. Kollman, A Second Generation Force Field for the Simulation of Proteins, Nucleic Acids, and Organic Molecules, J. Am. Chem. Soc. 117 (1995) 5179 -- 5197. doi:10.1021/ja00124a002. [45] J.K.M. Hunter, C.A., Sanders, The nature of. pi.-. pi. interactions, J. Am. Chem. Soc. 112 (1990) 5525 -- 5534. [46] B.R. Brooks, C.L.B. III, J. A. D. Mackerell, L. Nilsson, R.J. Petrella, B. Roux, Y. Won, G. Archontis, C. Bartels, S. Boresch, A. Caflisch, L. Caves, Q. Cui, A.R. Dinner, M. Feig, S. Fischer, J. Gao, M.W.I. Hodoscek, M. Karplus, Assessment of standard force field models against high‐quality ab initio potential curves for prototypes of π -- π, CH/π, and SH/π interactions, J. Comput. Chem. 30 (2009) 2187 -- 2193. doi:10.1002/jcc. [47] C.M. Ionescu, D. Sehnal, F.L. Falginella, P. Pant, L. Pravda, T. Bouchal, R. Svobodová Vařeková, S. Geidl, J. Koča, AtomicChargeCalculator: Interactive web-based calculation of atomic charges in large biomolecular complexes and drug-like molecules, J. Cheminform. 7 (2015) 50. doi:10.1186/s13321-015- 0099-x. [48] B. Nebgen, N. Lubbers, J.S. Smith, A.E. Sifain, A. Lokhov, O. Isayev, A.E. Roitberg, K. Barros, S. Tretiak, Transferable Dynamic Molecular Charge Assignment Using Deep Neural Networks, J. Chem. Theory Comput. (2018). doi:10.1021/acs.jctc.8b00524. [49] C.H. Chuang, M. Porel, R. Choudhury, C. Burda, V. Ramamurthy, Ultrafast Electron Transfer across a Nanocapsular Wall: Coumarins as Donors, Viologen as Acceptor, and Octa Acid Capsule as the Mediator, J. Phys. Chem. B. 122 (2018) 328 -- 337. doi:10.1021/acs.jpcb.7b11306. [50] A.M. Raj, M. Porel, P. Mukherjee, X. Ma, R. Choudhury, E. Galoppini, P. Sen, V. Ramamurthy, Ultrafast Electron Transfer from Upper Excited State of Encapsulated Azulenes to Acceptors across an Organic Molecular Wall, J. Phys. Chem. C. 121 (2017) 20205 -- 20216. doi:10.1021/acs.jpcc.7b07260. [51] D. Luo, F. Wang, J. Chen, F. Zhang, L. Yu, D. Wang, R.C. Willson, Z. Yang, Z. Ren, Poly(sodium 4-styrenesulfonate) Stabilized Janus Nanosheets in Brine with Retained 3694 -- 3700. doi:10.1021/acs.langmuir.8b00397. Amphiphilicity, Langmuir. 34 (2018) [52] M.S. Dresselhaus, G. Chen, M.Y. Tang, R. Yang, H. Lee, D. Wang, Z. Ren, J.P. Fleurial, P. Gogna, New directions for low-dimensional thermoelectric materials, Adv. Mater. 19 (2007) 1043 -- 1053. doi:10.1002/adma.200600527. [53] W. Wei, T. Jacob, Strong excitonic effects in the optical properties of graphitic carbon nitride g-C3 N4 from first principles, Phys. Rev. B. 87 (2013) 85202. doi:10.1103/PhysRevB.87.085202. [54] H. Kobayashi, N. Kobayashi, S. Hosoi, N. Koshitani, D. Murakami, R. Shirasawa, Y. Kudo, D. Hobara, Y. Tokita, M. Itabashi, Hopping and band mobilities of pentacene, and 2,7-dioctyl[1]benzothieno[3,2- b][1]benzothiophene (C8-BTBT) from first principle calculations, J. Chem. Phys. 139 (2013) 14707. doi:10.1063/1.4812389. rubrene, [55] Z. Jin, Q. Liao, H. Fang, Z. Liu, W. Liu, Z. Ding, T. Luo, N. Yang, A Revisit to High Thermoelectric Performance of Single-layer MoS2, Sci. Rep. 5 (2015) 18342. doi:10.1038/srep18342. [56] A. Troisi, Charge transport in high mobility molecular semiconductors: classical theories, Chem. Soc. Rev. 40 (2011) 2347 -- 2358. models and new doi:10.1039/c0cs00198h. [57] D.A. Da Silva Filho, E.G. Kim, J.L. Brédas, Transport properties in the rubrene crystal: Electronic coupling and vibrational reorganization energy, Adv. Mater. 17 (2005) 1072 -- 1076. doi:10.1002/adma.200401866. [58] T. Holstein, Studies of Polaron Motion: Part I. The Molecular-Crystal Model, Ann. Phys. (N. Y). 8 (1959) 325 -- 342. doi:10.1016/0003-4916(59)90002-8. [59] T. Sakanoue, H. Sirringhaus, Band-like temperature dependence of mobility in a solution-processed organic semiconductor, Nat. Mater. 9 (2010) 736. doi:10.1038/NMAT2825. [60] D. Wang, L. Tang, M. Long, Z. Shuai, First-principles investigation of organic semiconductors for thermoelectric applications, J. Chem. Phys. 131 (2009) 224704. doi:10.1063/1.3270161. [61] M. Yamagishi, J. Takeya, Y. Tominari, Y. Nakazawa, T. Kuroda, S. Ikehata, M. Uno, T. Nishikawa, T. Kawase, High-mobility double-gate organic single-crystal transistors with organic crystal gate insulators, Appl. Phys. Lett. 90 (2007) 182117. doi:10.1063/1.2736208. [62] O.D. Jurchescu, M. Popinciuc, B.J. Van Wees, T.T.M. Palstra, Interface- controlled, high-mobility organic transistors, Adv. Mater. 19 (2007) 688 -- 692. doi:10.1002/adma.200600929. [63] C.M. Wolfe, G.E. Stillman, W.T. Lindley, Electron mobility in high-purity GaAs, J. Appl. Phys. 41 (1970) 3088 -- 3091. doi:10.1063/1.1659368. [64] C. Bulutay, B.K. Ridley, N.A. Zakhleniuk, Full-band polar optical phonon scattering analysis and negative differential conductivity in wurtzite GaN, Phys. Rev. B. 62 (2000) 15754 -- 15763. doi:10.1103/PhysRevB.62.15754. [65] A. Henry, Thermal Transport in Polymers, Annu. Rev. Heat Transf. 17 (2013) 485 -- 520. doi:10.1615/AnnualRevHeatTransfer.2013006949. [66] S. Li, X. Yu, H. Bao, N. Yang, High Thermal Conductivity of Bulk Epoxy Resin by Bottom-Up Parallel-Linking and Strain: A Molecular Dynamics Study, J. Phys. Chem. C. 122 (2018) 13140 -- 13147. doi:10.1021/acs.jpcc.8b02001. [67] X. Xu, J. Chen, J. Zhou, B. Li, Thermal Conductivity of Polymers and Their Nanocomposites, Adv. Mater. 30 (2018) 1 -- 10. doi:10.1002/adma.201705544. [68] X. Xie, K. Yang, D. Li, T.-H. Tsai, J. Shin, P. V. Braun, D.G. Cahill, High and low thermal conductivity of amorphous macromolecules, Phys. Rev. B. 95 (2017) 035406. doi:10.1103/PhysRevB.95.035406. [69] C. Huang, X. Qian, R. Yang, Thermal conductivity of polymers and polymer 1 -- 22. (2018) Eng. Sci. 132 nanocomposites, Mater. doi:10.1016/j.mser.2018.06.002. R. [70] X. Wang, M. Zebarjadi, K. Esfarjani, First principles calculations of solid-state thermionic transport in layered van der Waals heterostructures, Nanoscale. 8 (2016) 14695 -- 14704. doi:10.1039/c6nr02436j. [71] H. Sadeghi, S. Sangtarash, C.J. Lambert, Cross-plane enhanced thermoelectricity and phonon suppression in graphene/MoS2 van der Waals heterostructures, doi:10.1088/2053- 1583/4/1/015012. 2D Mater. 015012. (2017) 4 [72] V. Coropceanu, J. Cornil, D.A. da Silva Filho, Y. Olivier, R. Silbey, J.L. Brédas, Charge transport in organic semiconductors, Chem. Rev. 107 (2007) 926 -- 952. doi:10.1021/cr050140x. Figure 1. Lattice structures of PCN ((a) and (b)), g-CN1 (g-C3N4) ((c) and (d)) and g- CN2 (tri-g-C3N4) ((e) and (f)).  Figure 2. (a) Projected band structure and pDOS of PCN. (b) Charge density of conduction band of PCN. The purple space shows the charge density. The isosurfaces value is set as 2×10-7 e/Bohr3 for positive and negative cases. (c) The Brillouin zone of the orthorhombic lattice and high-symmetric paths for the PCN band structure. (d) The Brillouin zone of the hexagonal lattice and high-symmetric paths for band structures of g-CN1 and g-CN2. The a* b* and c* are reciprocal lattice vectors. (e) The band structure of g-CN1. (f) The band structure of g-CN2. Figure 3. (a) (b) (c) TE coefficients of PCN versus n-type (n) and p-type (p) carrier concentrations under 300 K: (a) The electrical conductivity (σ) and Seebeck coefficients; (b) electronic (κe), lattice (κph) and total (κ) thermal conductivity; (c) ZT and the power factor; (d) The ZT maximum dependence of the temperature from 100 to 400 K. Figure 4. The charge density of CB and VB of g-CN1 and g-CN2: (a) CB of g-CN1; (b) VB of g-CN1; (c) CB of g-CN2; (d) VB of g-CN2. For (a) and (b), the isosurfaces value is set to be 0.01 e/Bohr3 for positive and negative cases. For (c) and (d), the isosurfaces value is set to be 0.005 e/Bohr3. 

1810.12516

3

1810

Strong Hot Carrier Effects Observed in a Single Nanowire Heterostructure

We use Transient Rayleigh Scattering to study the thermalization of hot photoexcited carriers in single GaAsSb/InP nanowire heterostructures. By comparing the energy loss rate in single bare GaAsSb nanowires which do not show substantial hot carrier effects with the core-shell nanowires, we show that the presence of an InP shell substantially suppresses the LO phonon emission rate at low temperatures leading to strong hot carrier effects.

cond-mat.mes-hall

cond-mat

Strong Hot Carrier Effects in Single Nanowire Heterostructures Iraj Abbasian Shojaei, Samuel Linser, Giriraj Jnawali, N. Wickramasuriya, Howard E. Jackson, Leigh M. Smith Department of Physics, University of Cincinnati, Cincinnati, OH 45221 Fariborz Kargar, Alexander A. Balandin Department of Electrical and Computer Engineering, University of California, Riverside CA 92521 USA Xiaoming Yuan School of Physics and Electronics, Hunan Key Laboratory for Supermicrostructure and Ultrafast Process, Central South University, 932 South Lushan Road, Changsha, Hunan 410083, P. R. China Philip Caroff, Hark Hoe Tan, and Chennupati Jagadish Department of Electronic Materials Engineering, Research School of Physics and Engineering, The Australian National University, Canberra, ACT 2601, Australia ABSTRACT: We use transient Rayleigh scattering to study the thermalization of hot photoexcited carriers in single GaAs0.7Sb0.3 / InP nanowire heterostructures. By comparing the energy loss rate in single core-only GaAs0.7Sb0.3 nanowires which do not show substantial hot carrier effects with the core- shell nanowires, we show that the presence of an InP shell substantially suppresses the longitudinal optical phonon emission rate at low temperatures which then leads to strong hot carrier effects. Keywords: Hot Carrier Effects, Hot Phonons, Carrier Thermalization 1 Introduction: The understanding of thermalization processes in semiconductors has significant implications for a wide variety of applications. For instance, one of the barriers to higher efficiency solar cells is the substantial loss of solar energy to heat as carriers thermalize before they are collected electrically.1 -- 5 When electrons and holes are created with substantial excess energy, they first rapidly relax (< 1 ps) through the emission of longitudinal optical (LO) phonons through polar Frohlich interactions.6 This creates a large non-thermal distribution of LO phonons which must equilibrate to the lattice temperature. In many materials, such as GaAs, GaSb and the ternary alloy GaAsSb, these non-thermal LO phonons can efficiently thermalize through the Klemens channel where a single zone center LO phonon decays into two counter propagating longitudinal acoustic ( LA) phonons. However, in materials such as InP where ħω LO > 2 ħω LA, this thermalization process is inhibited and so the LO phonon populations can remain in a non-thermal state for extended periods which inhibits cooling of the electrons and holes resulting in a pronounced hot carrier effect.7-9 The subject of hot carrier effects in semiconductors has been of strong interest for the past twenty years since it was first realized that LO phonon emission could be strongly suppressed in certain materials or structures. 10 -- 12 Substantial progress in understanding hot carrier thermalization in bulk and two dimensional materials has been made, but much less is known about how nanoscale heterostructures might impact thermalization dynamics.13,14 Numerous measurements have shown that hot carrier effects are more prominent in InP materials than GaAs-based nanowires (NWs),9,15 while others have shown that hot carrier effects are more dominant in thinner NWs,16 and terahertz measurements have shown that defects such as stacking faults can enhance hot carrier effects even further.17 -- 19 In this paper, we use transient Rayleigh scattering (TRS) to measure directly the average energy per carrier as a function of time in single NWs at 10 and 300 K and extract the energy loss rate for core-only GaAs0.7Sb0.3 NWs and GaAs0.7Sb0.3 / InP NW core-shell heterostructures. We show that while the core-only GaAs0.7Sb0.3 shows the expected optic-phonon dominated rapid thermalization, the growth of an InP shell surrounding GaAs0.7Sb0.3 core exhibits extremely strong hot-carrier effects. Sample Morphology and Experimental Setup: Core-only GaAs0.7Sb0.3 and core-shell GaAs0.7Sb0.3 / InP NWs were grown via the vapor- liquid-solid (VLS) method using gold catalysts in a metal-organic vapor phase epitaxy (MOVPE) system.20,21 Figure 1a shows an SEM image of the morphology of the core-shell NWs. Figure 1b shows a TEM image of a cross-section of a core-shell NW which displays the hexagonal inner core with non-polar {110} facets surrounded by outer truncated triangular-shaped InP shell with {112} facets. The diameter of the GaAs0.7Sb0.3 core is 70 nm and the thickness of the InP shell is a maximum of 30 nm. Details on the growth and morphology of the core-only and core-shell NWs can be found in Refs. 20 and 21. Under lattice-matched conditions GaAs0.5Sb0.5 / InP heterostructures have been shown to display a type II band alignment (see Figure 1c) with holes confined to the GaAs0.5Sb0.5 valence band (VB) with a 0.4 eV InP barrier, while the electrons are confined to the InP conduction band (CB) 2 with a relatively modest GaAs0.5Sb0.5 barrier.22 In GaAs0.7Sb0.3 / InP NWs, the GaAs0.7Sb0.3 core is under tensile strain while the InP shell is under compressive strain. It is expected that the type I to type II transition should occur at approximately 30% to 40% Sb composition. Thus, our expectation is that the holes are strongly confined to the GaAs0.7Sb0.3 core, while the electrons see nearly flat- band conditions in the CB between the core and shell. In the TRS experiments, a Coherent Chameleon Ti:Sapphire laser with 150 fs 1.5 eV pulses is used to excite single NWs with light polarized parallel to the long axis. The power of the pump beam is kept low enough so that there is no potential for heating the lattice. The change in the polarized scattering efficiency is monitored by a delayed mid-infrared output pulse (150 fs) from a Coherent Chameleon OPO laser with energies ranging from 0.79 to 1.13 eV (1100 to 1570 nm). The polarization of the probe beam oscillates at 100 kHz between parallel and perpendicular to the NW axis using a photoelastic modulator, and the scattered light from the NW is detected using a LN2- cooled InSb detector and a lock-in amplifier. Using a mechanical delay line, the probe pulse can be delayed relative to the pump pulse (-100 to 2000 ps) to investigate the effect of carrier decay and thermalization on scattering efficiency. For this measurement, the pump pulse train is chopped at 800 Hz and the output of the first lock-in is detected by a second lock-in amplifier tuned to this frequency. The second lock-in thus measures the change in the polarized scattering efficiency due to the presence of the photoexcited carriers: R´ = R´on - R´off where R´ = R - RThe normalized TRS scattering efficiency, R´/ R´ = (R´on - R´off) / R´off has a derivative-like behavior as a function of energy and depends on the geometry of the NW and the change of both the real and imaginary parts of complex index of refraction which is function of energy, carrier density and temperature. 9,23,24 3  Figure 1: (a) SEM images GaAs0.7Sb0.3 / InP nanowires. Scale bar is 1 m. (b) Cross sectional TEM image of GaAs0.7Sb0.3 / InP NW. Scale bar is 50 nm. (c) Schematic band diagram of type II GaAs1-xSbx / InP nanowire. (d, e) Solid black lines show exponential decay fitting of selected transient Rayleigh scattering carrier dynamics data for core-only (solid circle) and core- shell (open circle) nanowires at (d) 10 K (blue) and (e) 300 K (red) at fixed 0.83 eV probe energy. Experimental Results: Figure 1d shows the time decay of the TRS scattering efficiency for the core-only (solid circles) and core-shell (open circles) NWs at 10 K at a fixed 0.83 eV probe energy just below the band edge. The core-only NW decays rapidly to background with a 10 ps exponential decay, while the core-shell NW shows a remarkably long 1800 ps exponential decay. This indicates that photoexcited carrier recombination in the core-only GaAs0.7Sb0.3 NW is dominated by nonradiative surface recombination. The InP shell clearly passivates the surface states of the GaAsSb core, resulting in a two orders of magnitude longer lifetime. In contrast, at room temperature (300 K), the core-only NW shows a similar 10 ps fast decay, while the core-shell exhibits only a 150 ps time decay (Figure 1e). This is consistent with the CB of the InP shell providing only very weak 4 confinement for electrons of ~30 meV relative to the GaAs0.7Sb0.3 CB edge. If the band alignment is Type-I with electrons and holes confined to the core, the 30 meV confinement potential in the conduction band would not be sufficient to confine the electrons to the core if they have a thermal energy of ~30 meV at 300 K and so they would see the unpassivated surface states of the InP. If the band alignment is Type-II with the electrons confined to the InP, at low temperatures the Coulomb attraction of the electrons to the strongly confined holes in the core is sufficient to keep them close to the heterointerface. However, at 300 K it is expected that the thermal energy of the electrons would enable them to scatter more frequently with the surface states in the InP. Whether the weak confinement in the CB is type I or II is not possible to determine from our experimental results. By measuring the energy dependence of the polarized scattering efficiency one can obtain a more detailed understanding of the carrier dynamics.25,26 As noted previously, the resulting line- shapes exhibit a derivative-like spectra where the zero-crossing and linewidths depend directly on the density and temperature of the electron hole pairs, and the diameter of the nanowires. Using the analysis described in refs (23, 24) and detailed in the Supplemental Information, we are able to extract directly the density and temperature of the electron-hole pairs as a function of time after photoexcitation by the pump pulse. Figure 2 a,b shows such TRS spectra measured at ~10 K at three different delay times of the probe pulse for (a) GaAs0.7Sb0.3 core-only and (b) GaAs0.7Sb0.3 / InP core-shell NWs. As described by Montazeri et al., 23,24 the zero crossing point of NWs TRS spectra occurs approximately at the band gap of the structure. The line-shape fit is sensitively dependent on the diameter of the nanowire, and the density and temperature of the electron and hole distributions. The core-only NWs display a zero crossing at ~ 0.9 eV which shifts slightly to lower energy at later times. The line-shape for the core-only NW is very broad at early times and narrows within 20 ps which is indicative of filling of the conduction and valence bands with a dense and hot electron hole plasma which decays and thermalizes rapidly. The behavior of the core-shell NW is dramatically different with a smaller line- shape narrowing which extends over a two-orders of magnitude longer time of 2000 ps. 5  Figure 2: Theoretical fitting (blue lines) of transient Rayleigh scattering spectra at three different time delays at 10 K for (a) core-only (70 nm diameter) and (b) core-shell (130 nm diameter) nanowires. (c) Normalized carrier density of core-only and core-shell nanowires at 10 K have been marked on transient Rayleigh scattering time scan data with red hexagons and red squares respectively. Initial carrier density for core-only nanowires is around 4×1018 cm-3 and for core-shell nanowires is about 2×1018 cm-3. (d) Carrier temperature from modeling of transient Rayleigh scattering spectroscopy data at different measured delay time for core-only and core-shell nanowires at 10 K. By minimizing the difference between the theoretical line-shape and the data points (see section S1 in Supplemental Information), we are able to determine the electron-hole density and temperature for each delay time. The diameter of the nanowire is determined so that the chi-square is minimized for all spectra. In the fits displayed in Figure 2 the NW diameters determined in this way are 70 and 130 nm for core-only and core-shell NWs respectively, which are consistent with cross-sectional TEM measurements.21 Using these NW diameters and assuming the number of light and heavy holes equals the number of electrons, it is straightforward to determine the carrier density and temperature from each spectrum (see solid lines in Fig. 2 a and b). The time dependent density and temperature extracted from the fits of spectra at 10 K are shown in Figure 2c,d. The red squares 6 (red hexagons) in Figure 2c show the normalized carrier densities in the core-shell (core-only) NWs at times after photoexcitation which is in good agreement with the time decays shown previously. The initial carrier density for core-only NWs is ~ 4×1018 cm-3 and ~ 2 × 1018 cm-3 for the core-shell NWs. The fits confirm that the carrier density in the core-shell NWs takes 600 times longer to decay. This suggests that the band alignment of the core and shell is type-II (with holes confined to the core), but is not conclusive. The temperature (Figure 2d) of the photoexcited electrons and holes in the core-only NWs drops from 400 to 200 K in 20 ps, while in the core-shell NWs it takes nearly 2000 ps to drop to 140 K from the same initial temperature. These spectra also confirm that the lattice temperature does not change from the nominal 10 K. Figure 3a,b shows TRS spectra for core-only and core-shell NWs at ~300 K. The behavior of the line-shapes for the core-only NWs is very similar to that at low temperature, while the line- shape for the core-shell NWs exhibits a somewhat faster decay and thermalization with time, consistent with the shorter time decays observed at room temperature. The fits to the 300 K spectra (see Figure 3) again result in the same 70 and 130 nm diameters for the core-only and core-shell NWs respectively. The density and temperature extracted from these fittings at room temperature are shown in Figure 3c,d respectively. The normalized carrier densities obtained from the fits are shown in Figure 3c with blue hexagons for core-only NWs and blue squares for core-shell NWs. These points display good agreement with the related TRS time scan. The initial densities are 5×1018 cm-3 for the core-only NWs and 6.3×1018 cm-3 for the core-shell NWs. These fits show extremely rapid carrier thermalization in the core-only NWs from 500 K to 320 K within ~20 ps after the pump pulse. The core-shell NWs, on the other hand, show a slower thermalization from 550 to 320 K within 150 ps (Figure 3d). 7  Figure 3: Theoretical fitting (red lines) of transient Rayleigh scattering spectra at three different time delays at 300 K for (a) core-only and (b) core-shell nanowires. (c) Normalized carrier density of core- only and core-shell nanowires at 300 K have been marked on transient Rayleigh scattering time scan data with blue hexagons and blue squares respectively. Initial carrier density for core-only nanowires is around 5×1018 cm-3 and for core-shell nanowires is about 6.3×1018 cm-3. (d) Carrier temperature from modeling of transient Rayleigh scattering spectroscopy data at different measured delay time for core- only and core-shell nanowires at 300 K. From the fits to the time-resolved scattering spectra of these NWs suggest several conclusions. The first is that the InP shell clearly passivates non-radiative surface states in the GaAs0.7Sb0.3 NWs at both 10 and 300 K resulting in substantially longer recombination lifetimes in the core-shell NWs. The much larger lifetime enhancement observed at low temperatures may indicate that the band alignment of the core-shell NW is marginally type-II with electrons confined to the InP with a 30 meV confinement energy. However, the carrier temperature dynamics also indicate that the presence of the InP shell clearly causes a substantial slowing of the thermalization times although GaAsSb (like GaAs) is not known to be a material which shows substantial hot carrier effects. In the following sections we quantify the change in the energy loss rate in the core-shell NWs. 8  Carrier Thermalization: E(cid:3404) (cid:2871)(cid:2870) k(cid:2886)T (cid:2890)(cid:3119)(cid:3118) (cid:4666)(cid:2967)(cid:4667) (cid:2890)(cid:3117)(cid:3118) (cid:4666)(cid:2967)(cid:4667) (1) Through the fitting process described above we can determine the electron and hole densities (their quasi-Fermi energies) and temperature as a function of time. We can therefore calculate the dynamic change in the average energy per carriers using the expression:27 -- 29 where  is the quasi-fermi energy and Fi() is the ith Fermi integral defined in the usual manner. Using this result and the measured dynamics of the temperature and Fermi energies for electrons and holes we can calculate the average energy per electron-hole pair as a function of time after photoexcitation for both the core-only and core-shell NWs. The thermalization of electrons and holes is determined by the scattering (emission) rate of the carriers with LO and LA phonons which determines their energy loss rate.11,30,31 From the average energy per pair, we can calculate the energy loss rate (ELR) versus time simply by calculating the numerical derivative. The ELR calculated in this way is shown in Figure 4a,c for the core-only NWs at 10 K and 300 K respectively and Figure 4b,d for the core-shell NWs at 10 K and 300 K respectively. The difference between the core-only and core-shell NWs at 10 K is immediately obvious as the ELR of the core-only is three orders of magnitude larger than that of the core-shell NWs. 9  Figure 4: dE/dt and carrier energy loss rate due to optical and acoustic phonon emission for core-only and core-shell nanowires respectively at 10 K (a, b) and 300 K (c, d). Dashed lines in these graphs show energy loss rate due to optical phonon emission calculated by the Ridley expression. The reduction in the energy per particle reflects the thermalization of the carriers as a function of time. The derivative dE/dt also shows the dynamics of the ELR as a function of time. Because the thermalization process is dominated by emission of both LO and LA phonons, it is clear that:28 -- 30 〈(cid:2914)(cid:2889)(cid:2914)(cid:2930)〉(cid:3404)c 〈(cid:2914)(cid:2889)(cid:4666)(cid:2898)(cid:4666)(cid:2930)(cid:4667),(cid:2904)(cid:4666)(cid:2930)(cid:4667)(cid:4667) (cid:2914)(cid:2930) 〉(cid:2896)(cid:2899)(cid:3397) 〈(cid:2914)(cid:2889)(cid:4666)(cid:2889)(cid:3159)(cid:3161),(cid:2898)(cid:4666)(cid:2930)(cid:4667),(cid:2904)(cid:4666)(cid:2930)(cid:4667)(cid:4667) (cid:2914)(cid:2930) 〉(cid:2896)(cid:2885) (2) The left hand side of the equation is determined directly from the TRS measurements. Hot carrier effects in semiconductors occur because of a large suppression of the LO phonon emission rate because of hot phonons which cannot down convert efficiently to LA phonons.27,28 On the other 10 hand, if one knows the acoustic deformation potential, the ELR for LA phonons is quite well understood.32 This means that it is possible to extract the LO phonon ELR simply by subtracting the ELR for LA phonons directly from dE/dt calculated from the TRS data. For example, dE/dt from the TRS data for the core-only NWs at 10 K shows a total ELR which is ~1010 eV/s. Given the deformation potential of 1.6 eV, the LA phonon ELR is just below 108 eV/s.9 This means that the thermalization of hot carriers in the core-only NWs is completely dominated by LO phonon emission which explains the rapid decrease in temperature of the carriers (see Figure 4a). In contrast, the 10 K measurements for the core-shell NWs show a radically different behavior. While the total ELR starts at 109 eV/s, it falls rapidly to mid 107 eV/s and decreases slowly after that. We adjust the deformation potential to 1.6 eV in order to fit the late time response of the energy loss rate for the 10 K core-shell nanowires. By subtracting the LA phonon ELR from dE/dt, we therefore obtain the dynamics of the change in the LO phonon ELR in the core-shell NWs. This shows that the LO phonon ELR dominates at times less than 200 ps after the pump pulse but falls rapidly below the LA phonon ELR at later times (see Figure 4b). The black dashed lines in Figure 4 display the ELR due to LO phonon emission based on Ridley expression.27,33 -- 35 In this calculation, the LO phonon ELR depends on carrier temperature and density, lattice temperature and reabsorption of LO phonons which is represented by a coefficient in that formalism. For a given 0.0025 reabsorption coefficient, the ELR due to LO phonon emission is close to t h e carrier ELR in t h e core-only NW with an excellent correspondence to the LO phonon emission extracted from the dynamic measurements. But in the core-shell NWs by using same value of the reabsorption coefficient (0.0025), the LO ELR is close to our experimental result at early times, but is orders of magnitude too high at times greater than 200 ps. The LO phonon ELR can be related to the emission rate of the LO phonons through ħ/*, where * is the time between LO phonon emissions.36 This shows that the LO phonons emission rate for the core-only NWs is ~2 ps, while that for the core-shell NWs at 10 K is 10 ps at the earliest times but rapidly increases by three orders of magnitude to 4000 ps by 1 ns after the pump pulse. This result is shown in more detail in Fig. S1 in the Supplemental Information. Similar analysis of TRS data taken at room temperature (~300 K) shows that the LO phonon emission time for core-only NWs drops to 100 fs, while that for the core-shell NWs remains stable at 5 ps. This implies that even at room temperature hot carrier effects are not negligible in GaAs0.7Sb0.3 / InP nanostructures (see Figure 4c,d). The black dashed lines in this graph show the ELR based on the Ridley expression with a 0.025 reabsorption coefficient. This coefficient is 10 times larger than the 10 K value but matches well with the time-dependent ELR extracted from core- only NWs. Comparing the results in Fig. 4 c and d we see that the calculated LO phonon ELR from the core-only NW is almost one order of magnitude larger than the ELR extracted from TRS measurements in the core-shell NW. Discussion: The central conclusion from the above analysis is that a GaAs0.7Sb0.3 semiconductor NW made of material which should not show hot carrier effects, now shows very strong hot carrier effects at 11 low temperature if a thin 30 nm InP shell is added to the NW. In the discussion above, we considered mainly ELR of the charged carriers. We now consider how the phonon populations are affected after excitation by the pump pulse. Electrons and holes are created with nearly 600 meV of excess energy. This means that nearly 20 hot phonons per pair are created by the rapid relaxation of hot electrons and holes to the band edge through the Frohlich interaction. This is potentially different in the core- shell NW because Froehlich coupling in the InP shell is three times larger than the GaAs0.7Sb0.3 core (0.15 vs 0.05).37 This may suggest that during the initial relaxation of the hot carriers, substantially more InP-like LO phonons are created than GaAs0.7Sb0.3 phonons. Secondly, to thermalize with the lattice, these hot phonons need to decay anharmonically to the lower frequency LA phonon branches. Hot carrier effects happen because such anharmonic decays are inhibited and so hot phonons persist for much longer times which in turn inhibits thermalization of the hot carriers. 12 Over the past decade there has been intense interest in phonon engineering whereby one can use nanoscale heterostructures to tune the phonons in a material and also their interactions.38,39 Spatial confinement of phonons in nanostructures have been shown to strongly impact their dispersion, group velocity and density of states.38,40,41 While there have been a few papers which claim theoretically that the electron-phonon coupling can be impacted by the presence of a heterostructure, most conclusions are that such an effect is small.41 Several papers, however, indicate that a shell can strongly impact the LA phonons in the NW, particularly if there is a large impedance mismatch between the core and the shell. 42-46 The impedance, defined as  =  vs, where  is the density and vs is the sound velocity in the material, is substantially larger (40%) for the GaAs0.7Sb0.3 core than for the InP shell. Thus, the outer shell is "softer" than the core, and this has been shown to deplete the density of states of the phonons in the core and confining the phonons to the shell, particularly for large wave-vectors (high frequencies).42,43 Several papers have shown theoretically that the thermal conduction in both two- and one-dimensional structures can be strongly suppressed with the addition of a softer cladding layer.44,45 Others have shown that the mobility in the core can be enhanced by suppression of the LA phonon modes in the core.41 Similarly, Stroscio and Dutta have shown that in a nanostructure where the LO phonons are confined (keeping LA phonons not confined) that the anharmonic decay of the LO phonons is suppressed, resulting in a factor of two longer lifetimes. 47 The question is whether in the present case of a 70 nm diameter GaAs0.7Sb0.3 core and a 30 nm thick InP shell such confinement effects could be relevant. Two estimates of size scales where phonon confinement effects can be seen are when the diameter of the NW is either comparable to T = hvs / kBT or the diameter of the NW is less than the phonon mean free path, dMFP = 3K / (CV vs), where K is the thermal conductivity, Cv is the specific heat and vs is the phonon velocity. We use tabulated values for these parameters. At 10 K the thermal wavelength is approximately 20 nm while the mean free path is typically >200 nm. Recently Balandin and co-workers have shown using Brillouin scattering spectroscopy that confinement of LA phonons can be observed at room temperature in GaAs NWs with diameters as large as 130 nm.48 It has also been demonstrated experimentally that LA phonon confinement may affect thermal transport in nanostructures with the feature sizes of 25 nm.49 While the exact mechanism of the phonon dispersion modification on the charge carrier relaxation requires a separate theoretical investigation, one can conclude that confinement of phonons is certainly applicable for the 30 nm InP shell, and the core can produce some effects at low temperatures. The fact that the hot carrier effects in the core-shell NWs are stronger at low temperature may reflect the temperature dependence of the thermal phonon wavelength and the mean free path.  Figure 5: Calculated phonon dispersion in nanowires with and without a shell layer. The red and blue curves show the phonon dispersion along the nanowire axis in a GaAs0.7Sb0.3 nanowire with D=70 nm and GaAs0.7Sb0.3 / InP nanowire with the same inner diameter with an InP shell layer of 30 nm thickness. In Figure 5, we present the results of numerical simulations of the LA phonon dispersion in these NWs with a core diameter of 70 nm and a shell thickness of 30 nm. The simulations reveal noticeable differences between core-only and core-shell NWs, which potentially might be responsible for the experimentally observed phenomena. We see that the shell causes a significant change in the dispersion of all phonon polarization branches. In particular, one can see a decrease in the group velocity of the LA phonon polarization branch. The shell layer also induces bending of the LA phonon branch at smaller phonon wave-vectors compared to the NW without shell layers. The latter translates into substantially lower LA phonon energy at the Brillouin zone edge. While the differences among the phonon branches in terms of the absolute values of energy may not be large, the changed phonon group velocity and density of states can produce measurable effects at low temperature. This can be understood from the following considerations. The electron -- phonon scattering via deformation potential depends on the divergence of the phonon displacement (see for example Refs. [44-45]). The phonon displacement in a nanowire is different from that in bulk crystals and would depend on the diameter of the nanowire and mismatch between the nanowire core and the shell. This suggest that the electron relaxation will have a functional dependence on the specifics of the phonon dispersion, particularly at low temperature. Conclusions: We have shown using TRS to probe core-only GaAs0.7Sb0.3 and GaAs0.7Sb0.3 / InP core-shell NWs that the presence of the InP shell strongly influences hot carrier effects in these structures. For the core-only GaAs0.7Sb0.3 no hot carrier effects are seen, and the thermalization of photoexcited carriers is completely dominated by optical phonon emission at both 10 and 300 K. On the other hand, in the GaAs0.7Sb0.3 / InP core-shell NW at 10 K the LO phonon emission is completely suppressed at times longer than 200 ps and so thermalization is determined almost completely by LA phonon deformation potential scattering. At 300 K, thermalization of hot carriers in the core shell NW is determined by the LO phonon emission, but strong hot carrier effects are still observed with the emission rate reduced by an order of magnitude from the core-only NW. This provides the first 13 evidence that it might be possible to use concepts from phononic engineering to control hot carrier effects in semiconductors. Supporting Information. Description of fitting procedure of transient Rayleigh scattering lineshapes used to extract the radius of the nanowire and density and temperature of the photoexcited carriers. Graphic which shows optic phonon lifetimes versus time after pump pulse. ACKNOWLEDGEMENTS: We acknowledge the financial support of the NSF through grants DMR 1507844, DMR 1531373 and ECCS 1509706, and also the financial support of the Australian Research Council. The Australian National Fabrication Facility is acknowledged for access to the growth facility used in this work. A.A.B. acknowledges the support of DARPA project W911NF18-1-0041. X.M. Yuan thanks the financial support of the National Natural Science Foundation of China (No. 51702368). REFERENCES: (1) (2) (3) (4) (5) (6) (7) (8) Knig, D.; Casalenuovo, K.; Takeda, Y.; Conibeer, G.; Guillemoles, J. F.; Patterson, R.; Huang, L. M.; Green, M. A. Hot Carrier Solar Cells: Principles, Materials and Design. Phys. E Low‐Dimensional Syst. Nanostructures 2010, 42 (10), 2862 -- 2866. Le Bris, A.; Rodiere, J.; Colin, C.; Collin, S.; Pelouard, J. L.; Esteban, R.; Laroche, M.; Greffet, J. J.; Guillemoles, J. F. Hot Carrier Solar Cells: Controlling Thermalization in Ultrathin Devices. IEEE J. Photovoltaics 2012, 2 (4), 506 -- 511. Conibeer, G. J.; König, D.; Green, M. A.; Guillemoles, J. F. Slowing of Carrier Cooling in Hot Carrier Solar Cells. Thin Solid Films 2008, 516 (20), 6948 -- 6953. Le Bris, A.; Lombez, L.; Laribi, S.; Boissier, G.; Christol, P.; Guillemoles, J. F. Thermalisation Rate Study of GaSb‐Based Heterostructures by Continuous Wave Photoluminescence and Their Potential as Hot Carrier Solar Cell Absorbers. Energy Environ. Sci. 2012, 5 (3), 6225 -- 6232. Le Bris, A.; Guillemoles, J. F. Hot Carrier Solar Cells: Achievable Efficiency Accounting for Heat Losses in the Absorber and through Contacts. Appl. Phys. Lett. 2010, 97 (11), 113506. Lassnig, R. Polar Optical Interface Phonons and Fröhlich Interaction in Double Heterostructures. Phys. Rev. B 1984, 30 (12), 7132 -- 7137. Fritsch, J.; Pavone, P.; Schröder, U. Ab Initio Calculation of the Phonon Dispersion in Bulk InP and in the InP(110) Surface. Phys. Rev. B 1995, 52 (15), 11326 -- 11334. Vallée, F. Time‐Resolved Investigation of Coherent LO‐Phonon Relaxation in III‐V Semiconductors. Phys. Rev. B 1994, 49 (4), 2460 -- 2468. (9) Wang, Y.; Jackson, H. E.; Smith, L. M.; Burgess, T.; Paiman, S.; Gao, Q.; Tan, H. H.; Jagadish, C. Carrier Thermalization Dynamics in Single Zincblende and Wurtzite InP Nanowires. Nano Lett. 2014, 14 (12), 7153 -- 7160. (10) Gornik, E.; Stradling, R. A.; Findlay, P. C.; Kotitschke, R. T. Suppression of Lo Phonon Scattering in 14 Landau Quantized Quantum Dots. Phys. Rev. B ‐ Condens. Matter Mater. Phys. 1999, 59 (12), R7817 -- R7820. (11) Othonos, A. Probing Ultrafast Carrier and Phonon Dynamics in Semiconductors. J. Appl. Phys. 1998, 83 (4), 1789 -- 1830. (12) Sundaram, S. K.; Mazur, E. Inducing and Probing Non‐Thermal Transitions in Semiconductors Using Femtosecond Laser Pulses. Nat. Mater. 2002, 1 (4), 217 -- 224. (13) Michael Klopf, J.; Norris, P. Subpicosecond Observation of Photoexcited Carrier Thermalization and Relaxation in InP‐Based Films. Int. J. Thermophys. 2005, 26 (1), 127 -- 140. (14) (15) (16) (17) Elsaesser, T.; Woerner, M. Femtosecond Infrared Spectroscopy of Semiconductors and Semiconductor Nanostructures. Phys. Rep. 1999, 321 (6), 253 -- 305. Clady, R.; Tayebjee, M. J. Y.; Aliberti, P.; König, D.; Ekins‐Daukes, N. J.; Conibeer, G. J.; Schmidt, T. W.; Green, M. A. Interplay between the Hot Phonon Effect and Intervalley Scattering on the Cooling Rate of Hot Carriers in GaAs and InP. Prog. Photovolt: Res. Appl. 2012, 20 (3 -- 4), 82 -- 92. Tedeschi, D.; De Luca, M.; Fonseka, H. A.; Gao, Q.; Mura, F.; Tan, H. H.; Rubini, S.; Martelli, F.; Jagadish, C.; Capizzi, M.; et al. Long‐Lived Hot Carriers in III‐V Nanowires. Nano Lett. 2016, 16 (5), 3085 -- 3093. Johnston, M. B.; Whittaker, D. M.; Corchia, A.; Davies, A. G.; Linfield, E. H. Simulation of Terahertz Generation at Semiconductor Surfaces. Phys. Rev. B ‐ Condens. Matter Mater. Phys. 2002, 65 (16), 165301. (18) Beard, M. C.; Turner, G. M.; Schmuttenmaer, C. A. Transient Photoconductivity in GaAs as Measured by Time‐Resolved Terahertz Spectroscopy. Phys. Rev. B ‐ Condens. Matter Mater. Phys. 2000, 62 (23), 15764 -- 15777. (19) (20) (21) Joyce, H. J.; Docherty, C. J.; Gao, Q.; Tan, H. H.; Jagadish, C.; Lloyd‐Hughes, J.; Herz, L. M.; Johnston, M. B. Electronic Properties of GaAs, InAs and InP Nanowires Studied by Terahertz Spectroscopy. Nanotechnology 2013, 24 (21), 214006. Yuan, X.; Caroff, P.; Wong‐Leung, J.; Tan, H. H.; Jagadish, C. Controlling the Morphology, Composition and Crystal Structure in Gold‐Seeded GaAs 1−x Sb x Nanowires. Nanoscale 2015, 7 (11), 4995 -- 5003. Yuan, X.; Caroff, P.; Wang, F.; Guo, Y.; Wang, Y.; Jackson, H. E.; Smith, L. M.; Tan, H. H.; Jagadish, C. Antimony Induced {112}A Faceted Triangular GaAs<inf>1‐X</Inf>Sb<inf>x</Inf>/InP Core/Shell Nanowires and Their Enhanced Optical Quality. Adv. Funct. Mater. 2015, 25 (33), 5300 -- 5308. (22) Hu, J.; Xu, X. G.; Stotz, J. A. H.; Watkins, S. P.; Curzon, A. E.; Thewalt, M. L. W.; Matine, N.; Bolognesi, C. R. Type II Photoluminescence and Conduction Band Offsets of GaAsSb/InGaAs and GaAsSb/InP Heterostructures Grown by Metalorganic Vapor Phase Epitaxy. Appl. Phys. Lett. 1998, 73 (19), 2799 -- 2801. (23) Montazeri, M.; Wade, A.; Fickenscher, M.; Jackson, H. E.; Smith, L. M.; Yarrison‐Rice, J. M.; Gao, Q.; Tan, H. H.; Jagadish, C. Photomodulated Rayleigh Scattering of Single Semiconductor Nanowires: Probing Electronic Band Structure. Nano Lett. 2011, 11 (10), 4329 -- 4336. (24) Montazeri, M.; Jackson, H. E.; Smith, L. M.; Yarrison‐Rice, J. M.; Kang, J.‐H. H.; Gao, Q.; Tan, H. H.; Jagadish, C. Transient Rayleigh Scattering: A New Probe of Picosecond Carrier Dynamics in a Single Semiconductor Nanowire. Nano Lett. 2012, 12 (10), 5389 -- 5395. 15 (25) Sabbah, A. J.; Riffe, D. M. Femtosecond Pump‐Probe Reflectivity Study of Silicon Carrier Dynamics. Phys. Rev. B ‐ Condens. Matter Mater. Phys. 2002, 66 (16), 1 -- 11. (26) Mittendorff, M.; Wendler, F.; Malic, E.; Knorr, A.; Orlita, M.; Potemski, M.; Berger, C.; De Heer, W. A.; Schneider, H.; Helm, M.; et al. Carrier Dynamics in Landau‐Quantized Graphene Featuring Strong Auger Scattering. Nat. Phys. 2015, 11 (1), 75 -- 81. Leo, K.; Rühle, W. W.; Ploog, K. Hot‐Carrier Energy‐Loss Rates in GaAs/AlxGa1‐XAs Quantum Wells. Phys. Rev. B 1988, 38 (3), 1947 -- 1957. Žukauskas, A. Second Nonequilibrium‐Phonon Bottleneck for Carrier Cooling in Highly Excited Polar Semiconductors. Phys. Rev. B ‐ Condens. Matter Mater. Phys. 1998, 57 (24), 15337 -- 15344. Pugnet, M.; Collet, J.; Cornet, A. Cooling of Hot Electron‐Hole Plasmas in the Presence of Screened Electron‐Phonon Interactions. Solid State Commun. 1981, 38 (6), 531 -- 536. Lyon, S. A. Spectroscopy of Hot Carriers in Semiconductors. J. Lumin. 1986, 35 (3), 121 -- 154. Schoenlein, R. W.; Fujimoto, J. G.; Ippen, E. P. Femtosecond Absorption Saturation Studies of Hot Carriers in GaAs and AlGaAs. IEEE J. Quantum Electron. 1988, 24 (2), 267 -- 275. Cardona, M.; Christensen, N. E. Acoustic Deformation Potentials and Heterostructure Band Offsets in Semiconductors. Phys. Rev. B 1987, 35 (12), 6182 -- 6194. (27) (28) (29) (30) (31) (32) (33) Ridley, B. K. The Electron‐Phonon Interaction in Quasi‐Two‐ Dimensional Semiconductor Quantum‐Well Structures. J. Phys C : Solid State Phys. 1982, 15 (28), 5899‐5917. (34) (35) Zanato, D.; Balkan, N.; Ridley, B. K.; Hill, G.; Schaff, W. J. Hot Electron Cooling Rates via the Emission of LO‐Phonons in InN. Semicond. Sci. Technol. 2004, 19 (8), 1024 -- 1028. Lester, L. F.; Ridley, B. K. Hot Carriers and the Frequency Response of Quantum Well Lasers. J. Appl. Phys. 1992, 72 (7), 2579 -- 2588. (36) Ridley, B. K. Hot Electrons in Low‐Dimensional Structures. Rep. Prog. Phys. 1991, 54 (2), 169‐256. (37) Grundmann, M. Handbook on Semiconductors: The Physics of Semiconductors; Springer Berlin Heidelberg: New York, 2006. (38) Balandin, A. A.; Pokatilov, E. P.; Nika, D. L. Phonon Engineering in Hetero‐ and Nanostructures. J. Nanoelectron. Optoelectron. 2007, 2 (2), 140 -- 170. (39) Toberer, E. S.; Zevalkink, A.; Snyder, G. J. Phonon Engineering through Crystal Chemistry. J. Mater. Chem. 2011, 21 (40), 15843 -- 15852. Stroscio, M. A.; Dutta, M. Phonons in Nanostructures, first edition. Cambridge: UK, 2001. (40) (41) Ridley, B. K. Hybrid Phonons in Nanostructures, first edition. Oxford: UK, 2017. (42) Pokatilov, E. P.; Nika, D. L.; Balandin, A. A. Acoustic‐Phonon Propagation in Rectangular Semiconductor Nanowires with Elastically Dissimilar Barriers. Phys. Rev. B ‐ Condens. Matter Mater. Phys. 2005, 72 (11), 4 -- 7. Pokatilov, E. P.; Nika, D. L.; Balandin, A. A. Acoustic Phonon Engineering in Coated Cylindrical Nanowires. Superlattices Microstruct. 2005, 38 (3), 168 -- 183. Fonoberov, V. A.; Balandin, A. A. Phonon confinement effects in hybrid virus‐inorganic nanotubes for nanoelectronic applications. Nano Lett. 2005, 5 (10), 1920 -- 1923. (43) (44) 16 (45) (46) Fonoberov, V. A.; Balandin, A. A. Giant enhancement of the carrier mobility in silicon nanowires with diamond coating. Nano Lett. 2006, 6 (11), 2442 -- 2446. Ramayya, E. B.; Vasileska, D.; Goodnick, S. M.; Knezevic, I. Electron transport in silicon nanowires: The role of acoustic phonon confinement and surface roughness scattering. J. Appl. Phys. 2008, 104, 063711. (47) Datta, D.; Krishnababu, K.; Stroscio, M. A.; Dutta, M. Effect of Quantum Confinement on Lifetime of Anharmonic Decay of Optical Phonons in Semiconductor Nanostructures. J. Phys.: Condens. Matter. 2018, 30 (35), 355302. (48) (49) Kargar, F.; Debnath, B.; Kakko, J. P.; Saÿnätjoki, A.; Lipsanen, H.; Nika, D. L.; Lake, R. K.; Balandin, A. A. Direct Observation of Confined Acoustic Phonon Polarization Branches in Free‐Standing Semiconductor Nanowires. Nat. Commun. 2016, 7, 1 -- 7. Kargar, F.; Ramirez, S.; Debnath, B.; Malekpour, H.; Lake, R. K.; Balandin, A. A. Balandin. Acoustic phonon spectrum and thermal transport in nanoporous alumina arrays. Appl. Phys. Lett. 2015, 107 (17), 171904. 17 SUPPLEMENTAL INFORMATION: Strong Hot Carrier Effects in Single Nanowire Heterostructures Iraj Abbasian Shojaei, Samuel Linser, Giriraj Jnawali, N. Wickramasuriya, Howard E. Jackson, Leigh M. Smith Department of Physics, University of Cincinnati, Cincinnati, OH 45221 Fariborz Kargar, Alexander A. Balandin Department of Electrical and Computer Engineering, University of California, Riverside CA 92521 USA Xiaoming Yuan School of Physics and Electronics, Hunan Key Laboratory for Supermicrostructure and Ultrafast Process, Central South University, 932 South Lushan Road, Changsha, Hunan 410083, P. R. China Philip Caroff, Hark Hoe Tan, and Chennupati Jagadish Department of Electronic Materials Engineering, Research School of Physics and Engineering, The Australian National University, Canberra, ACT 2601, Australia S1. Theoretical Fitting: As discussed in M. Montazeri et al and Y. Wang et al,1-3 the line-shape of TRS spectra is sensitive to the three parameters: the density and temperature of the electron-hole plasma and the diameter of the NW. Since the NW diameter is much smaller than wavelength of the probe beam, back scattered light from NW is in the Rayleigh scattering regime. Also, because of large ratio of length to diameter of the NW and dielectric contrast of the NW with surroundings (air), polarization dependent classical Rayleigh scattering from a long uniform cylinder of radius r is a reliable approximation for analysis of scattered light from the NW. Using the Maxwell equations, It has been shown that scattered light by an infinite cylinder with radius r for light with incident polarization parallel (R║) and perpendicular (R) to the axis of cylinder depends on cylinder radius r and complex index of refraction (n)4. In general R║≠ R and thus the reflectance of the NW is polarization dependent. Thus, measuring R ´ = R║- R enable us to distinguish the scattered light of the NW from the large background reflection from the substrate. Since the complex index of refraction varies by with the carrier density and temperature, the carriers photoexcited by the pump beam changes R ´ , a n d ΔR´ = R´on - R´off shows the effect of the excited carries on the polarized scattering efficiency of the NW. Through band filling, the absorption (the complex part of the index of refraction) of the NW is modified by the presence of the electrons and holes. In turn, the real part of the index of the refraction is also modified. Thus, the normalized scattering efficiency R´/ R´ at different times after the pump excitation allows us to extract the carrier density and temperature and the diameter of the NW. The solid blue lines (red lines) in Figure 2a,b (Figure 3a,b) in the paper illustrate the line-shape of the theoretical modeling of R´/ R´ based on the following assumptions: (1) The electrons and holes are in thermal equilibrium at all times because of the extremely rapid carrier-carrier scattering rate. (2) The sum of the heavy and light hole densities in the VB equals the density of electrons (3) When the pump is off, the background carrier density and temperature is assumed to in the CB. be 1015 cm3 at 10 and 300 K. (4) All spectral fits at all times use the same value for the nanowire radius. (This radius is chosen to minimize the error for all fits). We calculate the energy dependence of the absorption by using direct band-to-band transition theory when the CB and VB are occupied by hot electrons and holes respectively. As shown previously,1-3 the TRS efficiency (  R ´ / R ´ ) depends only on the NW diameter and the change in the real and imaginary part of complex index of refraction: (cid:2940)(cid:2902)′(cid:2902)′(cid:3404) (cid:2940)(cid:4666)(cid:2902)‖(cid:2879) (cid:2902)(cid:3132)(cid:4667) (cid:2902)‖(cid:2879) (cid:2902)(cid:3132) ~ Re (cid:4670)e(cid:2919)(cid:2968)(cid:2722)(cid:1814)(cid:4671) (1) where is the modulation phase factor that depends on the NW diameter2, and Δn = Δn + i Δk is the change of the complex index of refraction due to the occupied CB and VB, where n is the index of refraction and k is proportional to absorption, k = (/4) Thus the derivative-like line-shape of  R ´ / R ´ can be expressed as: (cid:2902)´(cid:4666)(cid:2889),(cid:2930)(cid:4667)(cid:3404)Re (cid:4674)A e(cid:2919)(cid:2968)(cid:4672) Δn(cid:3397) i (cid:2971)(cid:2872)(cid:2976)Δα(cid:4673)(cid:4675)(cid:3404)A(cid:4668)cos(cid:4666)θ(cid:4666)r(cid:4667)(cid:4667)(cid:4670)n(cid:4666)E,N(cid:2915)(cid:2918)(cid:4666)t(cid:4667),T(cid:2915)(cid:2918)(cid:4666)t(cid:4667)(cid:4667)(cid:3398) (cid:2940)(cid:2902)´(cid:4666)(cid:2889),(cid:2930)(cid:4667) n(cid:4666)E,N(cid:2868),T(cid:2868)(cid:4667)(cid:4671)(cid:3398) (cid:2971)(cid:2872)(cid:2976)sin(cid:4666)θ(cid:4666)r(cid:4667)(cid:4667)(cid:4670)α(cid:4666)E,N(cid:2915)(cid:2918)(cid:4666)t(cid:4667),T(cid:2915)(cid:2918)(cid:4666)t(cid:4667)(cid:4667)(cid:3398)α(cid:4666)E,N(cid:2868),T(cid:2868)(cid:4667)(cid:4671)(cid:4669) (2) where A is an overall arbitrary amplitude factor. The individual time-resolved TRS spectra are modeled using this formula to extract the time-dependent carrier density and temperature in addition to the NW diameter. The absorption coefficient (E,N,T) is calculated using α(cid:4666)E,N,T(cid:4667)(cid:3404) (cid:2976)(cid:3118)(cid:2913)(cid:3118)(cid:2918)(cid:3119) (cid:2924)(cid:3118)(cid:2889)(cid:3118)(cid:4666)(cid:2870)(cid:2976)(cid:4667)(cid:3119)B(cid:1516) (cid:2889)(cid:2879)(cid:2889)(cid:3165) (cid:2868) ρ(cid:2913)(cid:4666)E´(cid:4667) ρ(cid:2932)(cid:4666)E´(cid:3398)E(cid:4667)(cid:4670)f(cid:2922)(cid:4666)E(cid:3398)E(cid:2917)(cid:3398)E´(cid:4667)(cid:3398)f(cid:2931)(cid:4666)E´(cid:4667)(cid:4671) dE′ (3) where B is the radiative bimolecular coefficient, n is the average index of refraction where we have used the average values from literature, i(E) is the 3D density of states in the CB and VB, f(E) = (1 + exp[(E -- EF) / kBT])-1 is the Fermi-Dirac distribution probability that upper and lower states involved in the transition are occupied by electrons, with the quasi-Fermi energy EF(N,T) related to both the carrier density N and temperature T, and E´ is the upper state energy above the CB minimum. Using the Kramers-Kronig relation we transform the calculated absorption coefficient to acquire the index of refraction n(E,N,T) as a function of energy, carrier density and temperature. To fit the TRS spectra as a function of time, the absorption coefficient and index of refraction ( and n) are calculated as a function of energy, carrier density and temperature (E, N, T) until the spectra are best fit to the resulting line-shapes using the expression for  R ´ / R ´ . By minimizing the difference between the theoretical line-shape and the data points, we are able to determine the electron-hole density and temperature for each delay time. S2. Effective relaxation time of LO phonons: The effective relaxation time of the LO phonons (*) for core-only and core-shell NWs at both 10 and 300 K extracted from our calculation for ELR (see Fig. 4 in the manuscript) is shown in Figure S1. For core-only NW at 10 K, it exhibit a constant value of 2 ps for the whole relaxation (Figure S1 a). On the other hand, by adding InP shell the * display a huge change. It starts around 10 ps and after 1 ns we see the value of around 4000 ps (Figure S1 b). At room temperature, also, the InP shell cause that * increases around one order of magnitude for all relaxation time (Figure S1 c, d). The black dashed lines in Figure S1 shows * based on Ridley expression which have nice corresponding for core-only NWs, but it is off from our calculations for core-shell NWs. (a) (b) (c) (d) Figure S1: The effective relaxation time of optical phonons for the core-only and core-shell nanowires at 10 and 300 K. Black dashed lines in the graph show effective relaxation time of optical phonons for core-shell nanowires calculated by Ridley expression. REFERENCES: ( 1 ) Wang, Y.; Jackson, H. E.; Smith, L. M.; Burgess, T.; Paiman, S.; Gao, Q.; Tan, H. H.; Jagadish, C. Carrier Thermalization Dynamics in Single Zincblende and Wurtzite InP Nanowires. Nano Lett. 2014, 14 (12), 7153 -- 7160. ( 2 ) Montazeri, M.; Wade, A.; Fickenscher, M.; Jackson, H. E.; Smith, L. M.; Yarrison-Rice, J. M.; Gao, Q.; Tan, H. H.; Jagadish, C. Photomodulated Rayleigh Scattering of Single Semiconductor Nanowires: Probing Electronic Band Structure. Nano Lett. 2011, 11 (10), 4329 -- 4336. ( 3 ) Montazeri, M.; Jackson, H. E.; Smith, L. M.; Yarrison-Rice, J. M.; Kang, J.-H. H.; Gao, Q.; Tan, H. H.; Jagadish, C. Transient Rayleigh Scattering: A New Probe of Picosecond Carrier Dynamics in a Single Semiconductor Nanowire. Nano Lett. 2012, 12 (10), 5389 -- 5395. van de Hulst, H. C. Light scattering by small particles; Dover Publication Inc.: New York, 1981. ( 4 ) 

1706.04493

1

1706

Bulk-edge correspondence in topological transport and pumping

The bulk-edge correspondence (BEC) refers to a one-to-one relation between the bulk and edge properties ubiquitous in topologically nontrivial systems. Depending on the setup, BEC manifests in different forms and govern the spectral and transport properties of topological insulators and semimetals. Although the topological pump is theoretically old, BEC in the pump has been established just recently [1] motivated by the state-of-the-art experiments using cold atoms [2,3]. The center of mass (CM) of a system with boundaries shows a sequence of quantized jumps in the adiabatic limit associated with the edge states. Although the bulk is adiabatic, the edge is inevitably non-adiabatic in the experimental setup or in any numerical simulations. Still the pumped charge is quantized and carried by the bulk. Its quantization is guaranteed by a compensation between the bulk and edges. We show that in the presence of disorder the pumped charge continues to be quantized despite the appearance of non-quantized jumps.

cond-mat.mes-hall

cond-mat

a Bulk-edge correspondence in topological transport and pumping Ken-Ichiro Imura1, Yukinori Yoshimura1,2, Takahiro Fukui3, Yasuhiro Hatsugai4 1Department of Quantum Matter, AdSM, Hiroshima University, 739-8530, Japan 2MathAM-OIL, AIST & Tohoku University, Sendai 980-8577, Japan 3Department of Physics, Ibaraki University, Mito 310-8512, Japan 4Institute of Physics, University of Tsukuba, Tsukuba 305-8571, Japan E-mail: imura@hiroshima-u.ac.jp Abstract. The bulk-edge correspondence (BEC) refers to a one-to-one relation between the bulk and edge properties ubiquitous in topologically nontrivial systems. Depending on the setup, BEC manifests in different forms and govern the spectral and transport properties of topological insulators and semimetals. Although the topological pump is theoretically old, BEC in the pump has been established just recently [1] motivated by the state-of-the-art experiments using cold atoms [2, 3]. The center of mass (CM) of a system with boundaries shows a sequence of quantized jumps in the adiabatic limit associated with the edge states. Although the bulk is adiabatic, the edge is inevitably non-adiabatic in the experimental setup or in any numerical simulations. Still the pumped charge is quantized and carried by the bulk. Its quantization is guaranteed by a compensation between the bulk and edges. We show that in the presence of disorder the pumped charge continues to be quantized despite the appearance of non-quantized jumps. 1. Introduction Recently, the role of topology is often highlighted in condensed-matter physics. This new trend in condensed-matter dates back to a few papers published three decades ago. One of them is the so-called TKNN paper [4], in which topological interpretation was given to quantum Hall effect (QHE), which had been discovered experimentally a few years earlier. The concept of topological pump is a variant of the idea of TKNN, applied to a temporal evolution of the system, instead of to the Brillouin zone. In Ref. [4], quantization of the Hall plateaus was given an interpretation as manifestation of an underlying topological order, which encodes a nontrivial phase property of the bulk wave function. Apparently, Thouless, one of the authors of Ref. [4] had the idea of applying the same scenario almost at the same time to the system of adiabatic pump [5]. Yet no clear experimental demonstration of this idea had been reported until in 2015 two experimental groups embodied the idea of topological pump a la Ref. [5] in the system of cold atoms; [2, 3] not to mention that the original proposal was based on an electronic system. One of the issues yet to be investigated in the experimental studies is on the role of disorder in pumping, which we focus on in this paper. 2. Contribution of the bulk vs. the edge: the bulk-edge correspondence A natural way to quantify pumping is to keep truck of a temporal evolution of the many body wave function. In case of topological pump this temporal evolution is adiabatic so that following the snapshot ¯x(t) of the center of mass (CM) of the ground state ¯x(t) = Xi xini (1) is enough to describe the pumping where ni is a many body particle number at xi [1]. Here taking a rescaling as xi = , i0 = , (i = 1, · · · , L) (2) i − i0 L L 2 is essential (L is the system size). To demonstrate this we consider a 1 + 1-dimensional (i.e., 1 spatial + 1 temporal dimensions) model; a one-dimensional (1D) model with an (effective) time-dependent potential [see Eq. (7)]. For the practical numerical simulation we employ the pump version of the so-called Harper, or Aubry-Andre model. [1] Then, we impose the boundary condition such that the system is periodic in time, while it is open i.e., with boundaries 1 in the space direction. This means that the snapshots of the CM is well-defined and periodic in time: ¯x(t0 + T ) − ¯x(t0) = 0, (3) where t0: initial time, T : pumping cycle. This might seem to imply that it is impossible to quantify pumping in this way. Let us recall, however, in the typical situation we consider in topological pump, the (one-body) spectrum of the system is characterized by the existence of edge modes traversing the bulk energy gap [see FIG. 1 (a)]. Therefore, the Fermi energy set typically in the gap intersects with such edge modes in the course of the time evolution. Then, if we consider an evolution of CM of the (many-body) ground state ¯x(t), it has two distinct parts which can be readily separable; patches of continuous curves and discrete jumps [see FIG. 1 (b)]. The jumps are necesarilly associated with edge modes in the clean limit, and their magnitudes ∆¯xjump are always quantized to be half integral: ∆¯xjump = ±1/2. [1] In topological insulators and related systems a one-to-one relation can be established between the appearance of edge/surface modes and the topological non-triviality in the bulk. The bulk-edge correspondence (BEC) refers to this one-to-one relation. in our choice of the boundary condition (open in one and periodic in the other), which was also the case in the so-called Laughlin's argument, [7] the effects of the bulk and edges are superposed and interconnected. The evolution of the CM in continuous patches is due to the bulk, while the jumps are due to the edge modes. To quantify pumping and reveal the compensating roles of the bulk and the edge, we attempt to separate the effect of the bulk and that of the edges. [6] Here, To concretize this BEC we reconnect the discrete patches of the CM curve by eliminating the discontinuities, and form a continuous CM curve over the cycle. Note that the resulting continuous curve is no longer periodic in time but it acquires a net gain (or loss) ∆¯xnet per cycle. One can interpret this ∆¯xnet as the net pumped charge, transported through the bulk. This is a polarization of the bulk. Since the net effect of bulk and edge contributions cancel after a cycle [see Eq. (3)], ∆¯xnet = − X{jn} ∆¯xjump(tjn), (4) where the summation is over the jumps, i.e., discontinuities of ¯x(t) due to the "appearance" or "disappearance" of an edge mode in the ground state subspace. Here, we are in a "grand- canonical point of view," [1] in which all the states below ǫF is occupied (in the ground state). 1 may sound paradoxical, but this seems to be the standard terminology in the field. {jn} = {j1, j2, · · ·} represents a set of time slices where the jumps occur (see Sec. 3 for its precise definition). The number Ne of the occupied states below ǫF , i.e., the number of electrons changes by 1 at such jumps. Since after a complete cycle of time T this number must get back to the original value, the number of times an edge state "appears" must be equal to the number of times an edge state "disappears" so that the total number Njump of jumps is even. This immediately results in that the pumped charge ∆¯xnet per cycle is quantized to be an integer[1]. This integral quantization of ∆¯xnet has a profound mathematical meaning. In parallel with the case of QHE one can directly relate ∆¯xnet to a topological (Chern) number [5]. In QHE, quantization of the Hall conductance σxy was attributed to the existence of an underlying topological number. [4, 8] Here, quantization of ∆¯xnet has the same mathematical origin. In case of FIG. 1 more than two: nF ≥ 2 bands are fully occupied below the Fermi energy ǫF , which is set to be between the nF -th and nF + 1-th bands. Then, ∆¯xnet becomes the sum of all the Chern numbers associated with a filled band: ∆¯xnet = C(nF ), C(nF ) = nF Xn=1 Cn, (5) where Cn is a Chern number associated with nth band. Or if one rather defines I(nF ) = −P{jn} ∆¯xjump(tjn), then I(nF ) represents the number of (the pair of) edge modes with suitable sign that appear at ǫF . The edge quantity I(nF ) is connected to the bulk quantity C(nF ) through Eq.(4). This is the BEC relation[1, 6, 9] in the topological pump: I(nF ) = C(nF ), I(nF ) − I(nF − 1) = CnF . (6) 3. Quantify pumping by the snapshots To implement the features of topological pump, it is convenient to work on a simple theoretical model. Even experimentally, such an approach is proven to be useful, e.g., in the system of cold atoms, [2, 3] already referred to. In Refs. [2, 3] the so-called Rice-Mele model was presumed, and an effective situation in which a description by the Rice-Mele model has been realized experimentally in a optical lattice. On the other hand, Ref. [1] considers the case of Harper model, in which situations represented by a high (≥ 2) Chern number can be readily realized. In the practical implementation we also consider this model, but for the time being, we can still work on a general 1D model with a time-dependent potential: H(ky) = L Xi=1 (cid:18)xi+1itxhxi + xiit∗ xhxi+1 + xii(V (xi, ky) + W (xi))hxi(cid:19), (7) where the replacement: ky → 2πt/T is presumed. [1, 5] In case of the Harper model the potential term becomes V (xi, ky) = 2ty cos(ky − 2πφi), where in this original 2D representation, φ represents the strength of a magnetic flux piercing a plaquette, while φ enters the 2D hopping Hamiltonian through Peierls substitution. W (x) is a site random potential distributed uniformly in the range W (x) ∈ [−W0/2, W0/2] 2. Note that W (x) is randomly distributed in space, while once this distribution is chosen, it stays static. W0 specify the strength of disorder. Eq. (7) is a Fourier transform of that 2D-hopping Hamiltonian. In Eq. (7), xi represents a Bloch state x, kyi = Py eiky yx, yi, while we make the replacement: ky → 2πt/T , in order to make Eq. (7) a 1D pump Hamiltonian. 2 This one dimensional randomness was discussed as a fictitious one in the original 2D problem[9]. It is now realized as it is in the 1D topological pump. 2.0 1.5 E 1.0 0.0 0.2 0.4 0.6 0.8 1.0 (a) t/T ) t ( x 0.6 0.4 0.2 0.0 -0.2 -0.4 -0.6 0.0 0.2 0.4 0.6 0.8 1.0 (b) t/T Figure 1. Time evolution of (a) the snapshot spectrum ǫα(t), and (b) the center of mass ¯x(t) of the ground state in case of the dis- ordered Harper model. In panel (b) the magnitude of the 6 jumps are in the order of their appear- ance 0.497119, 0.494151, 0.494077, 0.496461, 0.182738, −0.188663. Model is given in Eq. (7). Parame- ters are specified toward the end of Sec. 3. Let us consider snapshots of such a Hamiltonian as given in Eq. (7) with ky = 2πt/T at t = j∆t (j = 1, 2, 3, · · ·). At each time slice tj = j∆t we diagonalize the Hamiltonian Eq. (7) and find eigenstates, which include both the bulk and the edge states. Then, we consider the ground state of the system in which all the states below ǫF is occupied; both at the edge and in the bulk. We typically consider the case in which the Fermi energy ǫF is in the gap, since the pumped charge is topologically quantized in this case. Since the present case is non interacting, the center of mass ¯x(tj) of the many body ground state is given as L ¯x(tj) = Xi=1 xini(tj), ni(tj) = Xα ′ ψα(xi, tj)2, (8) at different time slices tj, where the summation Pα ′ is taken over all the occupied states α in the ground state. ψα(xi, tj) represents the eigenwavefunction corresponding to the eigenenergy ǫα(tj). FIG. 1 shows the evolution of the spectrum ǫα(t) [panel (a)]; here, the erratic behavior of the spectrum is due to the disorder potential, and that of the center of mass ¯x(t) in the ground state [panel (b)]. In practice, we plot simply the spectrum ǫα(tj) and ¯x(tj) at different time slices to visualize their evolution. In panel (a) one can observe that four branches of edges modes appear and traverse the energy gap. In panel (b) the CM curve ¯x(t) shows predominantly a continuous evolution except at a few [actually, six in the specific case of panel (b)] discontinuities (jumps), which occur typically when the equi-energy line at ǫF intersects with either of these edge branches. In panel (b) this is the case at the first four jumps, while the remaining jumps are due to impurities. Suppose that at t = tj1 a "right" edge mode localized in the vicinity of the right edge at x = +1/2. becomes available in the subspace of the ground state: ǫ ≤ ǫF . Then, the number Ne(t) = Pi ni(t) of occupied states increases by one at this time slice: Ne(tj1)−Ne(tj1−1) = 1. Correspondingly, ¯x(t) shows a quantized jump of +1/2; i.e., ∆¯xj1=+1/2. In panel (b) this seems to happen at the second and at the fourth jump. Generally, such a change of the occupied states Ne(t) occurs at a set of a finite number of time slices: t = tj1, tj2, · · ·, and there, ¯x(t) possibly shows discontinuities. {jn} in Eq. (4) specify the set of time slices {j1, j2, · · ·} at which the jumps occur. In the clean limit and if ǫF is in the gap, all of such intersections are associated with an edge mode, and at each tjn (n = 1, 2, · · ·) ¯x(t) shows a quantized jump of magnitude 1/2, since the localization length tends to zero after rescaling Eq.(2) at the extremity of the system. The sign of the jump depends on the location and the slope of the edge mode: [1] 1 2 ∆¯x= sgn(xedge)[−sgn(slope)], (9) where xedge = ±1/2 represents the location of the edge state, while when its slope is positive (negative) the state becomes empty (occupied) at the time slice in question. In other words, the factor [−sgn(slope)] is a measure of the appearance/disappearance of the state in question in/from the ground state. P′ {jn} ∆¯xjump(tjn)= + 2, where we used the notation P′ In the case of panel (b) in FIG. 1 the contribution of the first four jumps (associated with an edge state) to the summation on the right hand side of Eq. (4) is close to +2, i.e., {jn} to make explicit that only the contribution from the edge states is considered (by just counting the discontinuities). Recall that these half-integral quantized jumps always appear in pairs. Accordingly, the pumped charge in the bulk becomes ∆¯xnet= − 2, which is indeed identical to the bulk topological (Chern) number. In the situation of panel (b) the flux φ and ǫF are chosen such that φ = 1/7, L = 351 and ǫF = 1.4 with tx = ty = 1. As a result, 5 of 7 bands are fully occupied; ǫF is in the gap between the 5th and 6th bands. The Chern number Cn characterizing the occupied bands are +1, +1, +1, −6, +1, +1, +1 from the bottom to the valence band so that they sum up to −2. 4. Role of disorder: non-quantized jumps vs. quantized pumped charge In FIG. 1(b) one can observe that in addition to the quantized jumps we have focused on so far, there are additional jumps which are not quantized and appear "trivially" in pairs. The fifth and sixth jumps in FIG. 1(b) fall on this category. These additional jumps are due to impurity states. The direction and magnitude of such jumps are such that ∆¯x= ximp[−sgn(slope)], (10) 3 represents the location (CM) of the impurity state, while the factor [−sgn(slope)] where ximp indicates whether that appear or disappear in/from the ground state at the particular time slice. Another implication of this factor is that a given impurity state gives a pair of contributions to the summation on the right hand side of Eq. (4) with the same magnitude but with opposite signs, so that their contributions simply cancel each other. This is contrasting to the case of half-integral quantized jumps due to edge modes which also appear "in pairs" but in a different sense. In the case of quantized jumps the factor sgn(xedge) allows them to evade this cancellation and can still give a non-vanishing (though integral quantized) contribution to ∆¯xnet. Thanks to this cancellation, the calculated value of ∆¯xnet in the case of FIG. 1(b) is ∆¯xnet = −1.97588, which is close to the ideal value −2 in the clean limit [in the case of two panels in FIG. 1 the strength of disorder W0 is set as W0 = 1]. This example shows that despite the appearance of non-quantized jumps the pumped charge can still be quantized in the presence of disorder. 3 Due to the scaling Eq.(2), the position of the localized state is unambiguously specified as far as the localization length is finite. 5. Concluding remarks: comments on the experimental situations Let us summarize what we have argued so far. For the actual time evolution to be strictly identical to the collection of snapshots, the system must be adiabatic. This is the case when the pumping cycle T is long enough, satisfying the inequality T ≫ ¯h/∆ǫ, where ∆ǫ is the characteristic energy scale. In the "bulk regime" in which ¯x(t) shows a continuous evolution, ∆ǫ = ǫg (scale of the bulk energy gap), i.e., the adiabaticity is controlled by ǫg. In the vicinity of the jumps, on contrary, or in the "edge regime" ∆ǫedge → 0, since the edge is gapless, so that the typical time tedge = ¯h/∆ǫedge tends to be infinity. This means that for the adiabatic condition to be strictly satisfied at the edge the pumping cycle T must be infinite. Fortunately, this condition will not be satisfied experimentally; the pumping cycle T = Texp is well between the above two time scales: tbulk ≪ Texp ≪ tedge, i.e., the bulk is adiabatic, while the edges in experiments are non adiabatic, that is, described by the "sudden" approximation. Therefore, the jumps are not seen in the the experiments, even not in the numerical simulations of time evolution, while the "reconnection" is justified; i.e., one can safely skip the jumps. This is because in the regime of Texp ≪ tedge, the sudden approximation is fully justified at the edge. 4 The system behaves before and after the jump as if the jump does not exist. Finally, let us recall that the CM is only well-defined for an open system although the pumped charge is described by the polarization of the bulk, which is compensated by the discontinuities due to the edge states in the extreme adiabatic limit. Even though we may not see the jumps in the experiment, this part must underlie for the whole phenomenon to occur. Acknowledgments The authors are supported by KAKENHI: 15K05131 (KI), 15H0370001 (KI), 17H06138 (KI, TF, YH) and 16K13845 (YH). References [1] Hatsugai Phys. http://link.aps.org/doi/10.1103/PhysRevB.94.041102 Fukui 2016 and Y T Rev. B 94(4) 041102 URL [2] Nakajima S, Tomita T, Taie S, Ichinose T, Ozawa H, Wang L, Troyer M and Takahashi Y 2016 Nat Phys 12 296 -- 300 URL http://dx.doi.org/10.1038/nphys3622 [3] Lohse M, Schweizer C, Zilberberg O, Aidelsburger M and Bloch I 2016 Nat Phys 12 350 -- 354 URL http://dx.doi.org/10.1038/nphys3584 [4] Thouless D J, Kohmoto M, Nightingale M P and den Nijs M 1982 Phys. Rev. Lett. 49(6) 405 -- 408 URL http://link.aps.org/doi/10.1103/PhysRevLett.49.405 [5] Thouless D J 1983 Phys. Rev. B 27(10) 6083 -- 6087 URL http://link.aps.org/doi/10.1103/PhysRevB.27.6083 [6] Hatsugai Y 1993 Phys. Rev. Lett. 71(22) 3697 -- 3700 URL https://link.aps.org/doi/10.1103/PhysRevLett.71.3697 [7] Laughlin R B 1981 Phys. Rev. B 23(10) 5632 -- 5633 URL https://link.aps.org/doi/10.1103/PhysRevB.23.5632 [8] Kohmoto M 1985 354 http://www.sciencedirect.com/science/article/pii/0003491685901484 0003-4916 Physics Annals ISSN URL 160 of 343 -- [9] Hatsugai Y 1993 Phys. Rev. B 48(16) 11851 -- 11862 URL https://link.aps.org/doi/10.1103/PhysRevB.48.11851 4 A. Messiah, "Quantum mechanics", chap. XVII, §7. 

1609.05769

1

1609

Valley filtering by a line-defect in graphene: quantum interference and inversion of the filter effect

Valley filters are crucial to any device exploiting the valley degree of freedom. By using an atomistic model, we analyze the mechanism leading to the valley filtering produced by a line-defect in graphene and show how it can be inverted by external means. Thanks to a mode decomposition applied to a tight-binding model we can resolve the different transport channels in k-space while keeping a simple but accurate description of the band structure, both close and further away from the Dirac point. This allows the understanding of a destructive interference effect (Fano resonance or antiresonance) on the p-side of the Dirac point leading to a reduced conductance. We show that in the neighborhood of this feature the valley filtering can be reversed by changing the occupations with a gate voltage, the mechanism is explained in terms of a valley-dependent Fano resonance splitting. Our results open the door for an enhanced control of valley transport in graphene-based devices.

cond-mat.mes-hall

cond-mat

a Valley filtering by a line-defect in graphene: quantum interference and inversion of the filter effect L. H. Ingaramo1 and L. E. F. Foa Torres2 1)Instituto de F´ısica Enrique Gaviola (CONICET) and FaMAF, Universidad Nacional de C´ordoba, Argentina 2)Departamento de F´ısica, Facultad de Ciencias F´ısicas y Matem´aticas, Universidad de Chile, Santiago, Chile (Dated: 30 September 2018) Valley filters are crucial to any device exploiting the valley degree of freedom. By using an atomistic model, we analyze the mechanism leading to the valley filtering produced by a line-defect in graphene and show how it can be inverted by external means. Thanks to a mode decomposition applied to a tight-binding model we can resolve the different transport channels in k-space while keeping a simple but accurate description of the band structure, both close and further away from the Dirac point. This allows the understanding of a destructive interference effect (Fano resonance or antiresonance) on the p-side of the Dirac point leading to a reduced conductance. We show that in the neighborhood of this feature the valley filtering can be reversed by changing the occupations with a gate voltage, the mechanism is explained in terms of a valley-dependent Fano resonance splitting. Our results open the door for an enhanced control of valley transport in graphene-based devices. I. INTRODUCTION Graphene1 has two inequivalent Dirac cones related by time-reversal symmetry.2 -- 4 This endows the electronic states in graphene with a binary, spin-like, flavor. Harnessing this valley degree of freedom is an exciting avenue that may lead to new "valleytronics" applications. The interest has surged in the last few years not only in the context of graphene5 -- 7 but also for other two-dimensional materials like MoS2.8 A very first step for the operation of a valleytronics device is generating a valley polarization, i.e. selectively populating a single valley. Proposals for achieving such a valley filter in- clude the use of a constriction with zig-zag edges,5 and scat- tering against a (8-5-5) line-defect.9 The experimental real- ization of line-defects further contributes to the growing inter- est in their use as valley filter.10 -- 14 Indeed, besides their natu- ral occurrence in polycrystalline samples,15,16 different groups have demonstrated controlled growth of 8-5-5 line defects ei- ther via chemical vapor deposition on a Nickel step17 or by Joule heating.18 Filtering via line-defects has been elegantly demonstrated based on symmetry arguments9,19 which were also confirmed through atomistic calculations.9,20,21 But pre- vious works were mostly focused on the immediate proximity of the Dirac points where a low energy model can be used. Here we re-examine valley filtering in graphene with a 8-5- 5 line-defect. Our analysis is based on a tight-binding model using a mode-decomposition which helps us to resolve the conductive modes in k-space. This allows us to explore the mechanisms leading to valley filtering both in the neighbor- hood and also further away from the Dirac point. Close to the Dirac point we recover the results previously reported in the literature. Gating the system to the p-doped side one finds a conductance dip which we interpret as resulting from destruc- tive interference (known in the literature as Fano resonance or antiresonance). Interestingly, we find that around this con- ductance dip the valley filtering effect can be reversed by ap- plying a gate voltage. This is, if around the Dirac point elec- trons incident on the defect at a given angle are predominantly transmitted on the K valley, close to the conductance dip one can tune the occupation so that they are transmitted mainly on the K ′ valley. The mechanism is explained in terms of a valley-dependent Fano resonance splitting. This reversal of the filtering effect could be useful, for example, to produce the analog of a spin-valve effect. In the following we introduce our model Hamiltonian and the scheme used to resolve the scattering in k-space produced by the defect. Later on we examine the operation of the valley filter both close and away from the Dirac point and present our main results. II. HAMILTONIAN MODEL AND MODE DECOMPOSITION Let us consider a simple π-orbitals Hamiltonian2,4 for graphene: He = X i Eic† i ci − X hi,ji γi,j[c† i cj + h.c.] (1) where c† i and ci are the electronic creation and anihilation op- erators at site i, Ei is the site energy and hi, ji denote that the summation is restricted to nearest neighbors. The transfer in- tegrals between nearest neighbors is chosen as γ0 = 2.7eV 4. Since we are interested in the two-dimensional limit, we im- pose periodic boundary conditions along the armchair edge. This makes the quasi-momentum along the vertical direction (see Fig. 1(a)) a good quantum number which we will ex- ploit later on. The defect is modelled by taking into account the changes in the topology of the lattice according to a 8-5-5 linear structure (see Fig. 1(c)) as in Ref. 9. A tight-binding model is, a priori, not well suited for ob- taining transmission probabilities as a function of the incident angle of the electrons, i.e. kx and ky. Since the Hamiltonian is written in a real space basis, the momentum resolved infor- mation is hidden when one follows the standard procedure to compute the transmission probabilities from left to right.22,23 (a) ... 1 2 3 1 2 3 1 1 2 2 3 3 n-1 n-1 n-1 n-1 n n n n ... 2 γ0 cos(kqa) −γ0 cos(kqa) iγ0 sin(kqa) 0 if q = q′ ∈ [1, . . . , n/2]. if q = q′ ∈ [n/2 + 1, . . . , n]. if q′ = q + n/2. otherwise   hkqHdkq′i = (3) Therefore, as a result of placing the defect in the sample, the dimers are now connected in pairs as represented in Fig. 1(d). At low energies these two modes are not simultaneusly metal- lic, which means that an electron cannot be scattered from one valley to the other. FIG. 1. (a) Scheme showing a graphene ribbon lattice decomposed into succesive interconnected layers. A mode decomposition of the lattice shown in (a) leads to the independent modes represented in (b). (c) Detail of the line defect considered in the text. The defect couples two of the modes represented in (b) as shown in panel (d). To gain resolution in reciprocal space we exploit the peri- odic boundary conditions and switch to a basis where ky is well defined, a similar strategy was followed in Refs.20,21. In this basis the Hamiltonian of a pristine armchair ribbon with periodic boundary conditions can be written in block-diagonal form,4 where each block can be represented as shown in Fig. 1(b). Since the line-defect doubles the periodicity of the lat- tice along y it couples only those modes with ky differing in ky − ky ′ = 2π/2a, where a is the lattice parameter, which leads to the ladder model represented in Fig. 1(d). The eigenvectors defining the new basis can be worked out analytically. By choosing slices of the ribbon as shown in 1(a), so that the carbon atoms are along the same vertical line, the ribbon has a periodicity of four of such slices. Then, the prob- lem can be decoupled by changing to the following basis:4 kqi = 1 √n n X j=1 exp(ikqja)ji , (2) where kq = 2πq/na, q = 1, 2, ..., n (n being an even inte- ger number) and the sum on the right hand side is over the lattice sites on each slice of Fig. 1(a). Therefore, one can see that in this basis one gets independent modes as shown in Fig.1(b). These modes are indexed by kq, the quasimomen- tum in the vertical direction, and are represented by dimers with hoppings γ0 and γq = 2γ0 exp(−iπq/n) cos(qπ/n). These modes will be mixed by the line defect. The cor- responding matrix elements can be obtained by writing the matrix of the defect Hd in the basis of Eq. (2). This gives: Each leg of the ladder corresponds to a well defined ky. Thus, we still need to resolve the information on kx. This can be done by noticing that the asymptotic states also have a well defined kx which is fixed by the dispersion relation of bulk graphene and a boundary condition (direction of inci- dence). Therefore, an interesting aspect of this representation is that both the longitudinal and transversal quasi-momentum of the asymptotic states can be resolved within a tight-binding model. III. VALLEY FILTERING AND INVERSION OF THE FILTER EFFECT To motivate our discussion let us examine the transmission probability through a very wide ribbon containing a line de- fect perpendicular to the transport direction (x) as introduced in the previous section. The result of a tight-binding calcula- tion is shown in Fig. 2 with a full line, as a reference the result for a pristine system is shown with a the dashed line. Overall, we can see a reduction of the transmission probability consis- tent with a defect-induced enhanced backscattering. The most prominent difference is the dip observed in the p-doped region (i.e. for energies below the Dirac point). We will come back to the physical origin of this feature later on. Figure 2a provides information on the scattering for differ- ent energies of the incident electrons but it does not discrim- inate between valleys. One might be tempted to infer that the transmission probabilities for electrons entering the sam- ple at a given incident angle is independent on whether the quasimomentum lies close to K or K ′. Fig. 2b shows that this is not the case. There, we observe the dependence of the transmission probability with the angle of incidence and val- ley calculated from our tight-binding model at a Fermi energy close to the Dirac point. At normal incidence (0◦) the value of the transmission is equivalent for electrons from both valleys. In contrast, at high incident angle we see a larger difference of transmission probability between electrons from different valleys, which means a larger valley polarization for those an- gles. This behavior is stable at energies close to the Fermi level and is consistent with the results reported in Ref.9. Now, exploiting the capabilities of our atomistic descrip- tion, we turn to the study of the valley polarization further away from the Dirac point. We are interested, in particular, in the behavior close to the dip observed for negative doping in Fig. 2a. Interestingly, we find that the valley polarization is 3 θ K K' ◦ FIG. 2. (color online) (a) Transmission probability as a function of the incident electronic energy for a ribbon with a line-defect (solid line). The simulations are for a system 93,7 µm wide. The transmission for a pristine system of the same dimensions is shown for reference with dotted line. Notice the dip around Ef = −540meV . (b-d) Representative polar plots of the transmission probability as a function of the angle of incidence on the line defect (0◦ corresponds to normal incidence, see scheme). (b-d) differ in the energy of the incoming electrons which are marked on panel (a) with grey vertical lines (dash-dot): (b) is for ε = −23.3 meV, (c) is for ε = −594 meV and (d) is for ε = −729 meV. The solid line is for scattering around the K point while the dashed line is for the K ′ point. (b) and (c) show that the valley polarization gets inverted close to the transmission dip in (a). reversed around the condutance dip. Figs. 2c and 2d illustrate the dramatic change in angular dependence of the transmis- sion probability when the Fermi energy is slightly shifted. Let us examine the changes in the valley filter effect as we move away from the Dirac point. For energies above the Dirac point we find that the lobes shift their dominant angle from grazing angles to angles closer to normal incidence as the en- ergy increases. On the other hand, for negative energies we see a richer behavior: As the energy moves further away from the Dirac point the transmission degrades, especially at nor- mal incidence, the lobes become thinner and their maxima do not reach unity, see Fig. 2-c. Furthermore, for each valley one notices an incipient new lobe in the opposite quadrant, which becomes dominant as the energy is lowered even further, see Fig. 2-d. Therefore, the operation of the valley filter can be inverted by introducing a small change in the Fermi energy (e.g. through an applied gate voltage). The mechanism behind the valley filter inversion turns out to be closely related to that of the conductance suppression. The conductance dip reported earlier is the manifestation of a destructive interference effect, known as Fano resonance24 -- 26 or antiresonance27,28. This can be visualized by analyzing the individual modes represented in Fig. 1d, a set of repre- sentative transmissions is shown in Fig. 3a-d. The Green's function determining the transmission contains two contri- butions that can compete with each other: One correspond- ing to direct transmission from left to right on the same leg of the ladder, and another one which where the mode on the opposite leg (which has a larger energy gap) is explored. This can be captured by a simple two-pole approximation for the retarded Green's function determining the transmission: GR,L ∼ A/(ε − Ed) + B/(ε − Ed). The pole in the first term is located at ℜ(Ed) > 0.5 γ0 and dominates over the second one which is on the p-doped side. This is because the second pole comes from the non-conducting leg of the lad- der and therefore can only provide for a virtual process. The competition between these two poles leads to a destructive interference located closer to the non-dominant pole, on the p-doped side of the spectrum. As shown in Fig. 3, the pre- cise position of this antiresonance or Fano-resonance feature changes slightly from one mode to the other giving the overall behavior shown in Fig. 2a. Now, the question is how is this destructive interference linked to the inversion of the valley filter effect. To ratio- nalize it let us examine more closely the panels in Fig. 3. One can see that while moving away from the Dirac point in a chosen direction (as in Figs. 3b-d) the poles in the two-pole approximation get closer together in one valley while they get further apart in the other. Therefore, as one moves away from the Dirac point, the destructive interferences on each valley, which are degenerate in Fig.3a, follow the movement of the pole on the p-doped side (3b-d), thereby suffering a valley- dependent splitting (the position of the Fano resonances is marked with arrows in Fig.3). This splitting, in turn, pro- duces the observed change in the valley polarization direction around the destructive interference. This behavior can also be verified analytically from the ladder model. K K' ky kx K K' K K' k i i n o s s m s n a r T 4 would allow for better control of valley filtering in graphene and one could envisage, for example, the realization of the analog of a spin-valve29 by using two valley filters in series with their easy valley axis inverted. Indeed, previous works showed that line defects in se- ries could provide for an enhanced control of the valley polarization.21 Nonetheless, since the authors focused in the vicinity of the Dirac point, where the mechanism that we pro- pose is not active, the use of creative configurations together with a gate voltage remains as an interesting problem for fur- ther study. We acknowledge partial funding by Program 'Inserci´on' 2016, University of Chile and support from CONICET (Ar- gentina). LEFFT acknowledges the support of the Abdus Salam ICTP associateship program. LEFFT is on leave from Universidad Nacional de C´ordoba (Argentina) and CONICET. F. of and "The "The electronic properties S. Roche, Foa Torres, J. C. Charlier, rise of graphene," and M. S. Dresselhaus, 2-based carbon nanostructures," 1A. K. Geim and K. S. Novoselov, Nat Mater 6, 183 (2007). 2A. H. Castro Neto, F. Guinea, N. M. R. Peres, K. S. Novoselov, and A. K. Geim, graphene," Rev. Mod. Phys. 81, 109 (2009). 3V. Meunier, A. G. Souza Filho, E. B. Barros, "Physical properties of low-dimensional sp Rev. Mod. Phys. 88, 025005 (2016). 4L. E. Introduction to Graphene-Based Nanomaterials: From Electronic Structure to Quantum Transport (Cambridge University Press, 2014). 5A. Rycerz, J. Tworzydlo, and C. W. J. Beenakker, "Valley filter and valley valve in graphene," Nat Phys 3, 172 (2007). 6W. Yao, D. Xiao, and Q. Niu, "Valley-dependent optoelectronics from in- version symmetry breaking," Phys. Rev. B 77, 235406 (2008). 7R. V. Gorbachev, J. C. W. Song, G. L. Yu, A. V. Kretinin, F. Withers, Y. Cao, A. Mishchenko, I. V. Grigorieva, K. S. Novoselov, L. S. Levitov, and A. K. Geim, "Detecting topological currents in graphene superlattices," Science 346, 448 (2014). 8S. Wu, J. S. Ross, G.-B. Liu, G. Aivazian, A. Jones, Z. Fei, W. Zhu, D. Xiao, W. Yao, D. Cobden, and X. Xu, "Electrical tuning of valley magnetic mo- ment through symmetry control in bilayer mos2," Nat Phys 9, 149 (2013). 9D. Gunlycke and C. T. White, "Graphene valley filter using a line defect," Phys. Rev. Lett. 106, 136806 (2011). 10O. V. Yazyev and S. G. Louie, "Electronic transport in polycrystalline graphene," Nat Mater 9, 806 (2010). 11J. Song, H. Liu, H. Jiang, Q.-f. Sun, and X. C. Xie, "One-dimensional quan- tum channel in a graphene line defect," Phys. Rev. B 86, 085437 (2012). 12J. Rodrigues, N. Peres, ing by linear Phys. Rev. B 86, 214206 (2012). defects and J. Lopes dos Santos, "Scatter- A continuum approach," in graphene: 13H.-B. Yao, Z. Liu, M.-F. Zhu, filter line-defect-induced in EPL (Europhysics Letters) 109, 37010 (2015). valley and Y.-S. Zheng, strained "The graphene," 14X. Xu, W. Yao, D. Xiao, and T. F. Heinz, "Spin and pseudospins in layered transition metal dichalcogenides," Nat Phys 10, 343 (2014). 15A. W. Cummings, D. L. Duong, V. L. Nguyen, D. Van Tuan, J. Ko- and S. Roche, "Charge in polycrystalline graphene: Challenges and opportunities," takoski, J. E. Barrios Vargas, Y. H. Lee, transport Adv. Mater. 26, 5079 (2014). 16O. V. Yazyev and Y. P. Chen, "Polycrystalline graphene and other two- FIG. 3. (color online) Panels (a-d) show the transmission probabili- ties for individual channels corresponding to a well defined ky (rep- resented in the insets, the empty dot being the nearest Dirac point). ky (as measured from the nearest Dirac points) are set to 0, 2π∆/a, 4π∆/a, and 6π∆/a respectively (∆ = 1/130). e) Total transmis- sion probability for electrons from the K and K ′ valleys in solid and dashed lines respectively, for the [0◦ − 90◦] quadrant. Inversion of dominant polarization current occurs at EF ∼ −620meV (marked with a vertical grey line). To better visualize the valley-dependent splitting of the Fano resonance let us consider the sum of all the transmis- sion probabilities from electrons with incident angles in one quadrant (0◦ < θ < 90◦) on each valley separately. This is shown in Fig. 3e where the total transmission for the K and K ′ valleys are plotted in solid and dashed lines respectively. Near the charge neutrality point, transmission from K valley is dominant. This is consistent with the results for low en- ergy observed in Fig. 3a-d. At EF ∼ −620meV (gray verti- cal line) and one has balanced transmission from both valleys, which means zero polarization for the whole quadrant. In con- trast, for EF < −620meV the valley polarization is inverted. IV. CONCLUSIONS dimensional materials," Nat Nano 9, 755 (2014). 17J. Lahiri, Y. Lin, P. Bozkurt, I. I. Oleynik, and M. Batzill, "An extended We have shown a new mechanism leading to an inversion of the valley filtering effect in graphene: the valley-dependent splitting of the Fano resonances. Our results for graphene with a line-defect show that this could be achieved on the p- doped side where a destructive interference is present. This defect in graphene as a metallic wire," Nat Nano 5, 326 (2010). 18J.-H. Chen, G. Autes, N. Alem, F. Gargiulo, A. Gautam, M. Linck, C. Kisielowski, O. V. Yazyev, S. G. Louie, and A. Zettl, "Controlled growth of a line defect in graphene and implications for gate-tunable valley filter- ing," Phys. Rev. B 89, 121407 (2014). 19D. Gunlycke and C. T. White, "Specular graphene transport barrier," Phys. Rev. B 90, 035452 (2014). 20J. Zhou, S. Cheng, W.-L. You, and H. Jiang, "Effects of intervalley scat- tering on the transport properties in onedimensional valleytronic devices," Scientific Reports 6, 23211 (2016). 21Y. Liu, J. Song, Y. Li, Y. Liu, valley polarization using graphene multiple topological Phys. Rev. B 87, 195445 (2013). and Q.-f. Sun, "Controllable line defects," 22S. Datta, Electronic Transport in Mesoscopic Systems (Cambridge Univer- sity Press, 1995). 23C. H. Lewenkopf greens cursive Journal of Computational Electronics 12, 203 (2013). function and E. R. Mucciolo, for method "The re- graphene," 24U. Fano, "Sullo spettro di assorbimento dei gas nobili presso il limite dello spettro darco," Il Nuovo Cimento, 12, 154 (1935). 25M. L. L. d. Guevara, F. Claro, and P. A. Orellana, "Ghost fano resonance in a double quantum dot molecule attached to leads," Phys. Rev. B 67, 195335 (2003). 26A. E. Miroshnichenko, S. Flach, and Y. S. Kivshar, "Fano resonances in 5 nanoscale structures," Rev. Mod. Phys. 82, 2257 (2010). 27J. L. D'Amato, H. M. Pastawski, and J. F. Weisz, "Half-integer and integer quantum-flux periods in the magnetoresistance of one-dimensional rings," Phys. Rev. B 39, 3554 (1989). 28P. R. Levstein, H. M. Pastawski, interaction ing Journal of Physics: Condensed Matter 2, 1781 (1990). through-bond the and J. L. D'Amato, "Tun- in problem," two-centre a 29K. Wakabayashi and T. Aoki, "Electrical conductance of zigzag nanographite ribbons with locally applied gate voltage," International Jour- nal of Modern Physics B, Int. J. Mod. Phys. B 16, 4897 (2002). 

1112.0645

2

1112

Probing the quantum behaviors of a nanomechanical resonator coupled to a double quantum dot

We propose a current correlation spectrum approach to probe the quantum behaviors of a nanome-chanical resonator (NAMR). The NAMR is coupled to a double quantum dot (DQD), which acts as a quantum transducer and is further coupled to a quantum-point contact (QPC). By measuring the current correlation spectrum of the QPC, shifts in the DQD energy levels, which depend on the phonon occupation in the NAMR, are determined. Quantum behaviors of the NAMR could, thus, be observed. In particular, the cooling of the NAMR into the quantum regime could be examined. In addition, the effects of the coupling strength between the DQD and the NAMR on these energy shifts are studied. We also investigate the impacts on the current correlation spectrum of the QPC due to the backaction from the charge detector on the DQD.

cond-mat.mes-hall

cond-mat

Probing the quantum behaviors of a nanomechanical resonator coupled to a double quantum dot Zeng-Zhao Li,1, 2 Shi-Hua Ouyang,1, 2 Chi-Hang Lam,2 and J. Q. You1, 3, ∗ 1Department of Physics, State Key Laboratory of Surface Physics, Fudan University, Shanghai 200433, China 2Department of Applied Physics, Hong Kong Polytechnic University, Hung Hom, Hong Kong, China 3Beijing Computational Science Research Center, Beijing 100084, China (Dated: August 30, 2018) We propose a current correlation spectrum approach to probe the quantum behaviors of a nanome- chanical resonator (NAMR). The NAMR is coupled to a double quantum dot (DQD), which acts as a quantum transducer and is further coupled to a quantum-point contact (QPC). By measuring the current correlation spectrum of the QPC, shifts in the DQD energy levels, which depend on the phonon occupation in the NAMR, are determined. Quantum behaviors of the NAMR could, thus, be observed. In particular, the cooling of the NAMR into the quantum regime could be examined. In addition, the effects of the coupling strength between the DQD and the NAMR on these energy shifts are studied. We also investigate the impacts on the current correlation spectrum of the QPC due to the backaction from the charge detector on the DQD. PACS numbers: 85.85.+j, 03.67.Mn, 42.50.Lc I. INTRODUCTION The observation of quantum-mechanical behaviors in nanoelectromechanical systems, in particular, nanome- chanical resonators (NAMRs) for testing the basic prin- ciples of quantum mechanics1 -- 3 has become a topic of considerable interest and activity. Besides their wide range of potential applications,1,2 e.g., serving as ultra- sensitive sensors in high-precision displacement measure- ments, and detection of gravitational waves, quantized NAMRs are potentially useful for quantum-information processing. For example, NAMRs may serve as a unique intermediary for transferring quantum information be- tween microwave and optical domains because they can be coupled to electromagnetic waves of any frequency.4 At very low temperatures (in the milli-Kelvin range), NAMRs of high-vibration frequencies (gigahertz range) have recently been experimentally verified to approach the quantum limit.5 -- 11 However, low-frequency (. 100 MHz) mechanical oscillators have the distinct advan- tages of high-quality factors, long phonon lifetimes, and large motional state displacements, which are important for future testing of quantum theory12 and other appli- cations. A formidable challenge (see, e.g., Refs. 5 -- 10) in this field is to detect the quantum quivering (zero- point motion) of an NAMR so as to quantitatively verify whether it has been cooled into the quantum-mechanical regime or not. To directly detect the extremely small dis- placements of an NAMR vibrating at gigahertz frequen- cies by using available displacement-detection techniques is very difficult.8 -- 11 The usual position-measurement method is also severely limited by the "zero-point dis- placement" fluctuations in the quantum regime,13 al- though near-Heisenberg-limited measurements have been performed in recent experiments.14 In this paper, we propose a current spectroscopic ap- proach to study the behaviors of an NAMR. It is based on the detection of the current correlation spectrum in a charge detector, e.g., a quantum-point contact (QPC), which is indirectly coupled to an NAMR via a dou- ble quantum dot (DQD) acting as a quantum-electro- mechanical transducer.15 Based on this proposal, we show that one can observe the quantum behaviors of the NAMR and can further verify whether it has been cooled into the quantum regime. In contrast to a previous ap- proach based on the superconducting qubit coupled to a cavity, which involves Rabi splitting,16 our proposed setup is expected to provide better tunability via the gate voltages. Moreover, we also study the effects of the back- action from the charge detector on the DQD. We consider a coupled NAMR-DQD system in the strong dispersive regime where the coupling strength g is much smaller than the frequency detuning δ between the DQD and the NAMR. In this regime, energy quanta in the NAMR are only virtually exchanged between the DQD and the NAMR. Thus, the coupling of the DQD to the NAMR does not change the occupation state of the electron in the DQD, but only results in phonon- number-dependent frequency shifts in the DQD energy levels. These shifts are analogous to Stark shifts and can be further detected by measuring the current correlation spectrum of the QPC. This paper is organized as follows. In Sec. II, the cou- pled system is explained. The effective dispersive Hamil- tonian is derived in Sec. III. The quantum dynamics of the coupled NAMR-DQD system in the presence of the QPC are derived in Sec. IV. In Sec. V, results related to the observation of quantum behaviors, the verifica- tion of the ground-state cooling of an NAMR, as well as the backaction from the QPC on the DQD are analyzed. Conclusions are given in Sec. VI. II. THE COUPLED NAMR-DQD-QPC SYSTEM The total Hamiltonian of the whole system is H = Hsys + Hint + Hdet, with an unperturbed part, Hsys = HNAMR + HDQD + HQPC, (2) where (after setting  = 1), HNAMR = ωmb†b, HDQD = ∆ 2 (a† 2a2 − a† 1a1) + Ω(a† 2a1 + a† 1a2), HQPC =Xk ωSkc† SkcSk +Xq ωDqc† DqcDq. 2 (1) (3) (4) (5) (6) The device layout of an NAMR capacitively cou- pled to a lateral DQD, which is further measured by a QPC, is presented in Fig. 1(a). Here, we consider a Coulomb-blockade regime with strong intradot and in- terdot Coulomb interactions so that only one electron is allowed in the DQD. The states of the DQD are denoted by occupation states 1i and 2i, representing one elec- tron in the left and the right dots, respectively. The stere- ographical diagram of this device is shown in Fig. 1(b). The lateral DQD is formed by properly tuning the volt- ages applied to the gates. Also, an electron can be in- jected from the left reservoir to the DQD by changing the gate voltages. A metallic NAMR is fabricated above the DQD, and the displacement of the NAMR from its equilibrium position modulates the mutual capacitance between the NAMR and the DQD.17 The current IQPC through the QPC depends on the electron occupation of the DQD. ( )a DQD QPC ( )b W 1 2 NAMR QD1 QD2 QPCI NAMR ( )c e 2Ω g nδ eε gε nω + nχ nω mω + χ nω - nχ n-1ω mω - χ n mω n-1 DQD effective levels NAMR effective levels FIG. 1: (Color online) (a) Schematic of an NAMR capac- itively coupled to a DQD, which is under measurement by a nearby QPC. The energy detuning between the two dot states in the DQD is zero, and the interdot coupling strength between them is Ω. (b) A stereographical diagram of the de- vice where an electron is injected from the left reservoir to the DQD by changing the gate voltages, and the displacement of the metallic NAMR from its equilibrium position modulates the capacitance between the NAMR and the DQD. (c) Effec- tive energy levels for the DQD (horizontal solid lines in the left panel) in the dispersive DQD-NAMR coupling regime: εe = Ω + (n + 1/2) χ and εg = −Ω − (n + 1/2) χ with energy detuning δn = 2Ω + (2n + 1) χ; effective energy levels for the NAMR (horizontal solid lines in the right panel) in the dis- persive DQD-NAMR coupling regime: ωn + nχ and ωn − nχ with ωn = nωm. The effective phonon level differences are ωm + χ and ωm − χ. The interaction parts are Hint = −(g2a† Hdet =Xkq (cid:16)T0 − ζ2a† 2a2 − ζ1a† 2a2 + g1a† 1a1)(cid:0)b† + b(cid:1) , 1a1(cid:17)(cid:16)c† SkcDq + c† DqcSk(cid:17) . (7) Here HNAMR, HDQD, and HQPC, respectively, are the free Hamiltonians of the NAMR, the DQD, and the QPC without the tunneling terms. The phonon operators b† and b, respectively, create and annihilate an excitation of frequency ωm in the NAMR. In Eq. (4), ∆ is the energy detuning between the two dots, and Ω is the interdot cou- pling. Below, we consider, for simplicity, the degenerate- state case with ∆ = 0 [see Fig. 1(a)]. cSk (cDq) is the annihilation operator for electrons in the source (drain) reservoir of the QPC with momentum k (q). Here, we define pseudospin operators σz ≡ a† 1a1 and σx ≡ a† 1a2 with a1 (a2) being the annihilation op- erator for an electron staying at the left (right) dot. Hint is the electromechanical coupling between the NAMR and dots 1 and 2 with coupling strengthes g1 and g2. The relative coupling strengths g ≡ (g2 − g1)/2 is about 0.1ωm ∼ 0.5ωm for typical electromechanical couplings (see, e.g., Ref. 18). Hdet describes tunnelings in the QPC, which depends on the electron occupation of the DQD, owing to the electrostatic coupling between the DQD and the QPC. We define T ≡ T0 − (ζ2 + ζ1)/2 and ζ ≡ (ζ2 − ζ1)/2 so that the transition amplitude of the QPC when an extra electron stays at the left and right dots equals T + ζ or T − ζ, respectively. 2a2 − a† 2a1 + a† III. EFFECTIVE DISPERSIVE HAMILTONIAN In the eigenstate basis, the DQD Hamiltonian can be written as HDQD = Ωz, (8) eae − a† where z = a† gag with the two eigenstates of the DQD given by gi = (1i − 2i) /√2 and ei = (1i + 2i) /√2 and the energy splitting between these two eigenstates is 2Ω. Then, the total Hamiltonian be- comes H = ωmb†b + Ωz + gx(cid:0)b† + b(cid:1) [T + ζx](cid:16)c† +HQPC +Xkq SkcDq + c† DqcSk(cid:17) ,(9) where x = a† eag + a† gae. In the dispersive DQD-NAMR coupling regime with η < 1, where η = g/δ and δ = 2Ω − ωm, applying a canonical transformation U HU † on the Hamiltonian (9), where U = es with s = η(cid:0)+b − −b†(cid:1) , one obtains an effective dispersive Hamiltonian. Under the rotating- wave approximation, this dispersive Hamiltonian can be written, up to second order in η, as H = H0 + HI, with H0 = ωmb†b + Ωz + χ(cid:18) 1 (T + ζx)(cid:16)c† HI =Xkq + b†b(cid:19) z + HQPC, DqcSk(cid:17) . SkcDq + c† 2 (10) (11) Here, χ = g2/δ. The third term in Eq. (10) is a dispersive interaction that can be viewed as either a DQD-state- dependent frequency shift in the NAMR or a phonon- number-dependent shift in the DQD transition frequency. This interaction implies that, when the DQD state is excited (deexcited), an energy 2χ is effectively added to (removed from) each NAMR phonon. A similar fre- quency shift also appears in analogous systems in quan- tum optics.19 The dispersive NAMR-DQD energy levels, described by the first three terms in Eq. (10), are the quantum version of the ac Stark effect. When there is no interaction (g = 0) between the NAMR and the DQD, energy differences between adjacent levels of the NAMR or the DQD are simply ωm or 2Ω, respectively. However, for g > 0, these eigenstates are dressed by the dispersive interaction. The corresponding phonon level differences become ωm − χ for the DQD state gi and ωm + χ for state ei, whereas the DQD energy split is δn ≡ 2Ω + (2n + 1)χ (12) for phonon number n in the NAMR. Figure 1(c) shows these effective energy-level differences. The phonon- number-dependent frequency shift in the DQD as well as the DQD-state-dependent shift in the NAMR can be detected as will be explained below. IV. QUANTUM DYNAMICS OF THE COUPLED NAMR-DQD SYSTEM 3 picture with the dispersive Hamiltonian H0 in Eq. (10), the interaction Hamiltonian HI [Eq. (11)] can be written as HI (t) = S (t) Y (t) , (13) with S (t) = (14) (15) Pj eibωj t, 3Xj=1 Y (t) =Xkq (cid:2)F † 2 + b†b(cid:1) , bω2 = −bω1, bω3 = 0, F † kq (t) + Fkq (t)(cid:3), where P1 = χ+, P2 = χ−, P3 = T, bω1 = 2Ω + 2χ(cid:0) 1 kq (t) = c† SkcDq+ei(ωSk−ωDk), and Fkq (t) = c† DqcSke−i(ωSk−ωDk). Applying the Born-Markov approximation and tracing over the degrees of freedom of the QPC, quantum dy- namics of the NAMR-DQD system are governed by ρI (t) = TrS,D(cid:8) − i [HI (t) , ρtot (0)] −Z ∞ 0 [HI (t) , [HI (t′) , ρtot (t)]](cid:9). (16) Here, ρtot (t) is the density operator of the whole system including the QPC as well. Substituting HI from Eq. (13) into Eq. (16) and converting the resulting equation into the Schrodinger picture, we obtain the master equation, ρ (t) = Lρ (t) = −i [HDQD, ρ (t)]+Ldρ (t)+γdD [−] ρ (t) , (17) with Ldρ (t) = (cid:8) 3Xi=1 D [Pi] ρ (t) + 3Xj=1(j6=i) 3Xi=1 ×2πgSgDζ2(cid:2)Θ (eVQPC − ωi) +Θ (−eVQPC − ωi)(cid:3), D [Pi, Pj] ρ (t)(cid:9) (18) where Θ (x) = (x + x) /2 and gS,D denotes the den- sity of states in the source and drain reservoirs of the QPC, which has a bias voltage VQPC. ωi is the eigen- value of the operator bωi with the NAMR in the ni state. Here, for simplicity, the temperatures of the reservoirs in the DQD-QPC system (instead of the temperature Tm of the NAMR) are chosen to be T = 0K because related quantum-dot experiments are performed at ex- tremely low temperatures (see, e.g., Ref. 20). The super- operator D, acting on any single or double operators, is defined as D [A] ρ ≡ AρA† − 1 2 A†Aρ − 1 2 ρA†A, (19) We now derive a master equation to describe the quan- tum dynamics of the coupled system. In the interaction D [A, B] ρ ≡ 1 2(cid:0)AρB† + BρA† − B†Aρ − ρA†B(cid:1). (20) where γ0 = γ1 + γd 2 , νn =qδ2 n − γ2 1 , 4 (29) (30) and pn is the probability that the NAMR is in state ni. In the calculation, we have assumed 0 < γ1 < δn (see typical parameters listed in Sec. V A). V. CURRENT CORRELATION SPECTRUM OF THE QPC The dc current through the QPC is given by22 I (t) = Ilρ11 (t) + Irρ22 (t) , (31) where Il = eD and Ir = eD′ are the currents through the QPC when dots 1 and 2, respectively, are occupied.22 Here, D = 2πgSgD (T − ζ)2 eVQPC and D′ = 2πgSgD (T + ζ)2 eVQPC are the corresponding rates of electron tunneling through the QPC, which follows from Eq. (7). Using ρ11 + ρ22 = 1, one can define the current operator as bI (t) = I0 − I1σz, (32) 2 (D + D′) and I1 = e with I0 = e 2 (D − D′) and x = −σz in the degenerate-state case with ∆ = 0. According to the Wiener-Khintchine theorem, when the phonon in the NAMR is in state ni, the QPC current correlation power spectrum Sn (ω) is given in terms of the two-time correlation function as19 0 Sn (ω) = 2ℜ +∞Z dτ eiωτ(cid:2)(cid:10)bI (t)bI (t + τ )(cid:11)n −(cid:10)bI (t + τ )(cid:11)n(cid:10)bI (t)(cid:11)n(cid:3). Substituting Eqs. (25)−(28) and (32) into Eq. (33) and using S (ω) = S0 +Pn pnSn (ω) , we get (33) ω γ1 p2 2γ1γ2 γ1 + γ2 Xn γ0 (cid:18)1 + nn νn(cid:19) + ×h1 + 0 + (νn + ω)2h1 + γ2 νnio. γ+ − γ− − γd 2γ0 δn γ0 + + γ0 0 + (νn − ω)2 γ2 γ+ − γ− − γd νn(cid:19) γ0 (cid:18)1 − 2γ0 γ1 ω δn νni (34) S (ω) = 1 + pn, S0 To account for the coupling of the DQD to other degrees of freedom, such as hyperfine interactions and electron- phonon couplings, we have phenomenologically included an additional relaxation term [the third term on the right-hand side of Eq. (17)] describing transitions from excited state ei to ground state gi .21 In the strong dis- persive regime, as we have mentioned before, the phonon in the NAMR neither is absorbed nor induces any transi- tions in the DQD and, hence, does not change the occu- pation probability of the DQD. Instead, the occupation state of the DQD is only affected by the backaction of the QPC and the phenomenological relaxation term. In the basis {e, ni ,g, ni} of the coupled NAMR-DQD system, we obtain the following evolution equations for the reduced density matrix elements: ρen,en = γ+ρgn,gn − (γ− + γd) ρen,en, ρgn,gn = −γ+ρgn,gn + (γ− + γd) ρen,en, 1 2 ρen,gn = −iδnρen,gn − γ1 (ρen,gn − ρgn,en) − (21) (22) γdρen,gn, (23) 1 2 ρgn,en = iδnρgn,en + γ1 (ρen,gn − ρgn,en) − γdρgn,en. (24) Assuming eVQPC > δn > 0, the QPC-induced relaxation and excitation rates between the ground state and the ex- cited state of the DQD are defined as γ+ = γ1 (1 − λn) , γ− = γ1 (1 + λn) , where γ1 = 2πgSgDχ2eVQPC and λn = δn/eVQPC. Since the decay rate of the NAMR is much smaller than that of the DQD, dissipations of the NAMR are neglected (see further discussions be- low). In Eqs. (21)−(24), the reduced density matrix ele- ment ρin,in(i = g, e) gives the occupation probability of state i, ni of the coupled NAMR-DQD system, whereas ρin,jn(i 6= j) describes the quantum coherence between states i, ni and j, ni. Equations of motion for other el- ements, e.g., ρin,jn′ (n 6= n′), which are decoupled from those considered here, are not shown. Using the normal- ization condition pn = ρgn,gn + ρen,en, the solutions to the equations above are obtained as pn, (25) γ+ 2γ0 γ+ 2γ0 ρen,en (t) =(cid:20)ρen,en (0) − ρgn,gn (t) =(cid:20)ρgn,gn (0) − ρen,gn (t) = e−γ0th cos (νnt) ρen,gn (0) + sin (νnt) pn(cid:21) e−2γ0t + pn(cid:21) e−2γ0t+ γ− + γd 2γ0 2γ0 γ− + γd (26) γ1ρgn,en (0) − iδnρen,gn (0) (27) ρgn,en (t) = e−γ0th cos (νnt) ρgn,en (0) + sin (νnt) γ1ρen,gn (0) + iδnρgn,en (0) × × pδ2 n − γ2 1 pδ2 n − γ2 1 i, i, (28) Here, S0 = 2eI0 is the current-noise background. From Eq. (34), one sees that the current correlation spectrum of the QPC consists of peaks at resonance frequencies ω = ±νn. These peaks have width γ0 and heights in- creasing with the probability pn. In particular, for small backaction from the QPC, i.e., γ1 ≪ δn, peaks are lo- cated at the resonance point ω = δn = 2Ω + (2n + 1)χ, admitting a shift (2n + 1)χ inherited from the phonon- number-dependent frequency shift in the DQD as ex- plained above. Thus, one can read out the phonon- number state of the NAMR from these peak shifts in the current correlation spectrum of the QPC. A. Verification of ground state cooling of the NAMR The observation of quantum mechanical phenomena requires a high frequency and a low temperature for the NAMR (see, e.g., Refs. 5 -- 7) so that Nth ≡ kBTm/ωm < 1 or hni < 0.582, where Nth is the thermal occupa- tion number and kB is the Boltzmann constant. We now assume thermal equilibrium of the NAMR with a probability19 pn ∝ e−hnHNAMRni/kBT for a state ni so that pn = hnin / (1 + hni)n+1 . In general, a state with the average phonon number hni ≪ 1 (e.g., hni = 0.01) (a) no NAMR <n>=0.02 0.40 0.41 0.42 0.43 1.10 1.05 1.00 0 S / ) ( S m u r t c e p S n o i t l a e r r o C t n e r r u C 1.08 (b) <n>=0.3 1.04 1.00 1.04 1.02 1.00 1.002 1.001 1.000 0.40 0.41 0.42 0.43 (c) <n>=0.6 0.40 0.41 0.42 0.43 (d) <n>=6 0.40 0.41 0.42 /2 (GHz) 0.43 n p y t i l i b a b o r P 1.0 0.8 0.6 0.4 0.2 0.0 1.0 0.8 0.6 0.4 0.2 0.0 1.0 0.8 0.6 0.4 0.2 0.0 1.0 0.8 0.6 0.4 0.2 0.0 <n>=0.02 0 1 2 3 4 5 6 <n>=0.3 0 1 2 3 4 5 6 <n>=0.6 0 1 2 3 4 5 6 <n>=6 0 1 2 3 n 4 5 6 (Color online) Left panel: Current correlation FIG. 2: spectrum of the QPC when the average phonon numbers in the NAMR are (a) hni = 0.02, (b) 0.3, (c) 0.6, and (d) 6, respectively, given by the thermal distribution, i.e., pn = hnin/(1 + hni)n+1. Right panel: the corresponding probability of the NAMR in state ni. The coupling strength between the NAMR and the DQD is chosen as g = 0.3 ωm. The other parameters are ωm = 2π × 100 MHz, Ω = 2ωm, γ2 = 0.01ωm, γ1 = 0.2γ2, γd = 2γ2, and ζ/T = 0.044. 5 implying p0 ≈ 1 is considered as a quantum ground state. By using typical parameters18,21,23 -- 25 ωm = 2π × 100, Ω = 2π × 200 MHz, g = 0.3ωm, ζ/T = 0.044, γ2 = 0.01ωm, γ1 = 0.2γ2, and γd = 2γ2, the current cor- relation spectrum of the QPC is calculated and is pre- sented in Fig. 2 where only the positive frequency regime is shown. For hni = 0.02 ≪ 0.582 in the quantum regime [see Fig. 2(a)], there is only a single peak in the spec- trum corresponding to the transition frequency between the two eigenstates of the DQD. From Eq. (34), the peak is located at the resonance frequency ν0 given in Eq. (30). The corresponding probability distribution function pn is also shown in the right panel of the figures showing the probability of finding the NAMR in state ni. This spec- trum is nearly indistinguishable from the pure ground state with hni = 0. At a higher temperature, other peaks begin to appear in the current correlation spec- trum [Fig. 2(b)]. The peak position as given in Eqs. (30) and (34) admits an NAMR-induced shift analogous to an ac Stark shift. In the regime with, e.g., hni = 0.6 [Fig. 2(c)] and 6 [Fig. 2(d)], multiple peaks are obtained. As each resonance peak in the spectrum corresponds to a phonon-number state of the NAMR, the relative area under each peak could be used, in principle, to calculate the phonon statistics of the NAMR. To observe multiple peaks in the correlation spectrum, the separation between two adjacent peaks must be larger than the intrinsic peak width, i.e., 2χ > γ0. The ensem- ble can then be individually resolved, which allows us to detect the phonon number to verify the cooling efficiency of the NAMR. On the contrary, the phonon-number state of the NAMR cannot be measured when 2χ < γ0. Indeed, a relatively strong coupling between an NAMR and a quantum dot has been recently demonstrated.26 Thus, the regime with 2χ > γ0 could be achievable. Also, the ground-state cooling of an NAMR coupled to a DQD was proposed.17 One can then apply the proposed cou- pled NAMR-DQD system to verify the ground-state cool- ing of the NAMR. The frequency shifts in the DQD en- ergy levels are also different for the ground and excited states of the NAMR and can then be used to read out the phonon state of the NAMR, which can be different from the phonon statistics in the thermal state discussed above. Figure 3(a) shows the current correlation spectrum of the QPC when the NAMR is in its ground state with hni = 0.02. If the NAMR and the DQD are decou- pled, i.e., g = 0, the frequency corresponding to the peak position is about 2π × 0.4 GHz, which is the transition frequency 2Ω between the two eigenstates of the DQD. By increasing the coupling strength g, we find that the peak shifts to the right, while the linewidth as well as the amplitude are unchanged. This suggests that the en- ergy levels of the DQD are shifted so that the energy splitting is widened under the effect of the NAMR. How- ever, these changes do not involve the absorption of any NAMR phonon. As demonstrated in Fig. 3(b), the fre- quency shift increases with the square of the coupling strength between the DQD and the NAMR, consistent with χ = g2/δ in Eq. (10). B. QPC-induced backaction in the current correlation spectrum of the QPC The backaction on the DQD due to measurement by the QPC is illustrated in Fig. 4 when the NAMR is prac- tically in its ground state with hni = 0.02. There is no backaction effect when the bias voltage VQPC across the QPC is less than the energy difference between the two DQD eigenstates,21 i.e., eVQPC < δ0. At eVQPC = 2π × 0.5 GHz > δ0, for example, a single peak located at ω ∼ 2π × 0.4 GHz appears. When VQPC is increased, we find that the linewidth of the spectrum becomes broad- ened, which results from γ0 = γ1 + γd/2 where γ1 is proportional to the bias voltage. Physically, the broad- ening results from more frequent state transitions in the DQD induced by the backaction from the QPC when a larger bias voltage is applied across the QPC. Dissipations in the NAMR due to the environment 6 have been neglected in our analysis. Dissipation in an NAMR (see Ref. 27) can be expressed as γm = ωm/Q with a quality factor Q. However, even for the NAMR- DQD coupling discussed above (e.g., 2π × 30 MHz), the dissipation of the NAMR is still very small: γm/g ∼ 10−4 for an experimentally achievable quality factor26 Q = 105. This justifies neglecting the dissipations of the NAMR due to other environmental effects in our calcu- lations. For a QPC, a Kondo-like model was proposed in Ref. 28, which is similar to the Kondo problem in a single quantum dot coupled to two leads where the spin degree of freedom plays an important role. Here, as in Refs. 22 and 29, the QPC we used is simply modeled as a tun- neling junction, and the spin degree of freedom does not affect its performance. In addition, it should be noted that the Kondo effect in the DQD can be avoided here. In fact, the present setup involves no reservoirs (leads) coupled to the DQD because the coupling between the DQD and the reservoirs is tuned to zero or is negligibly small. Nevertheless, the Kondo effect in a DQD needs a strong coupling between the DQD and the reservoirs in addition to other requirements. In practice, there are finite cross-capacitive couplings among various gate electrodes, which affect the whole system. However, because the coupling between the NAMR and the DQD is in the dispersive regime, the ef- fect of the NAMR on varying the parameters of the DQD is small. As for the cross-capacitive couplings in the DQD system, the experiment in Ref. 30 showed that the effect of the cross-capacitive couplings can be canceled by ad- justing the plunger voltage of the detector during sweeps of the DQD plunger voltages. In our proposed setup in Fig. 1(b), more gate electrodes are introduced. This will enhance the tunability of the DQD system to achieve the needed parameters of the system. eVQPC/2 =0.5GHz eVQPC/2 =2.5GHz eVQPC/2 =7.5GHz 0 S / ) ( 1.12 S m u r t c e p S n o i t l a e r r o C t n e r r u C 1.08 1.04 1.00 FIG. 3: (Color online) (a) Current correlation spectrum of the QPC when the NAMR is in the ground state and the coupling strengths g between the NAMR and the DQD are 0 (black squares), 10 MHz (red circles), 30 MHz (blue upper triangles), and 50 MHz (olive lower triangles), respectively. (b) Frequency shift δg as a function of the coupling strength g when the NAMR is in the ground state. Other parameters are the same as those in Fig. 2. 0.395 0.400 0.405 /2 (GHz) 0.410 0.415 FIG. 4: (Color online) Current correlation spectrum of the QPC at various voltage biases VQPC of the QPC when the NAMR is approximately in the ground state, i.e., hni = 0.02. Other parameters are the same as those in Fig. 2. VI. CONCLUSIONS We have proposed an approach to study quantum be- haviors of an NAMR by coupling it indirectly to a QPC as a charge detector via a DQD serving as a quantum trans- ducer. By detecting the current correlation spectrum of the charge detector, quantum behaviors of the NAMR can be observed. It provides interesting insight on the quantum system as well as dynamics of these backaction effects induced by an act of measurement, which neces- sarily perturbs the system being measured. More impor- tantly, the cooling of the NAMR down to the quantum regime can be verified. In the quantum regime, NAMR- phonon-induced shifts (an analog to the Stark shift) of 7 DQD energy levels as well as their relations with coupling strength between the NAMR and the DQD are demon- strated. Backaction effects from the charge detector are also explained. Acknowledgements This work was supported by the National Basic Re- search Program of China Grant No. 2009CB929300, the National Natural Science Foundation of China Grant No. 91121015, and the Research Grant Council of Hong Kong SAR under Project No. 5009/08P. ∗ Electronic address: jqyou@fudan.edu.cn 1 M. P. Blencowe, Phys. Rep. 395, 159 (2004). 2 K. C. Schwab and M. L. Roukes, Phys. Today 58(7), 36 16 L. F. Wei, Y. X. Liu, C. P. Sun, and F. Nori, Phys. Rev. Lett. 97, 237201 (2006). 17 S. H. Ouyang, J. Q. You, and F. Nori, Phys. Rev. B 79, (2005). 075304 (2009). 3 A. Cho, Science 299, 36 (2003). 4 C. A. Regal and K. W. Lehnert, J. Phys.: Conf. Ser. 264, 18 N. Lambert and F. Nori, Phys. Rev. B 78, 214302 (2008). 19 M. O. Scully and M. S. Zubairy, Quantum Optics (Cam- 012025 (2011). 5 A. Gaidarzhy, G. Zolfagharkhani, R. L. Badzey, and P. Mohanty, Phys. Rev. Lett. 94, 030402 (2005); 95, 248902 (2005); K. C. Schwab, M. P. Blencowe, M. L. Roukes, A. N. Cleland, S. M. Girvin, G. J. Milburn, and K. L. Ekinci, ibid. 95, 248901 (2005). 6 R. L. Badzey and P. Mohanty, Nature (London) 437, 995 (2005). 7 W. K. Hensinger, D. W. Utami, H. S. Goan, K. Schwab, C. Monroe, and G. J. Milburn, Phys. Rev. A 72, 041405(R) (2005). 8 M. D. LaHaye, O. Buu, B. Camarota, and K. C. Schwab, Science 304, 74 (2004). 9 X. M. H. Huang, C. A. Zorman, M. Mehregany, and M. L. Roukes, Nature (London) 421, 496 (2003). 10 R. G. Knobel and A. N. Cleland, Nature (London) 424, 291 (2003). 11 A. N. Cleland, J. S. Aldridge, D. C. Driscoll, and A. C. Gossard, Appl. Phys. Lett. 81, 1699 (2002). bridge University Press, Cambridge, U.K., 1997). 20 S. Gustavsson, M. Studer, R. Leturcq, T. Ihn, K. Ensslin, D. C. Driscoll, and A. C. Gossard, Phys. Rev. Lett. 99, 206804 (2007). 21 S. H. Ouyang, C. H. Lam, and J. Q. You, Phys. Rev. B 81, 075301 (2010). 22 S. A. Gurvitz, Phys. Rev. B 56, 15215 (1997). 23 W. G. van der Wiel, S. De Franceschi, J. M. Elzerman, T. Fujisawa, S. Tarucha, and L. P. Kouwenhoven, Rev. Mod. Phys. 75, 1 (2003). 24 T. F. Li, Y. A. Pashkin, O. Astafiev, Y. Nakamura, J. S. Tsai, and H. Im, Appl. Phys. Lett. 92, 043112 (2008). 25 L. M. K. Vandersypen, J. M. Elzerman, R. N. Schouten, L. H. W. van Beveren, R. Hanson, and L. P. Kouwenhoven, Appl. Phys. Lett. 85, 4394 (2004). 26 A. K. Huttel, G. A. Steele, B. Witkamp, M. Poot, L. P. Kouwenhoven, and H. S. J. van der Zant, Nano Lett. 9, 2547 (2009). 27 J. Tamayo, J. Appl. Phys. 97, 044903 (2005); A. N. Cleland 12 W. Marshall, C. Simon, R. Penrose, and D. Bouwmeester, and M. L. Roukes, ibid. 92, 2758 (2002). Phys. Rev. Lett. 91, 130401 (2003). 28 Y. Meir, K. Hirose, and N. S.Wingreen, Phys. Rev. Lett. 13 M. F. Bocko and R. Onofrio, Rev. Mod. Phys. 68, 755 89, 196802 (2002). (1996). 14 J. D. Teufel, T. Donner, Dale Li, J. W. Harlow, M. S. Allman, K. Cicak, A. J. Sirois, J. D. Whittaker, K. W. Lehnert, and R. W. Simmonds, Nature (London) 475, 359 (2011). 15 M. R. Geller and A. N. Cleland, Phys. Rev. A 71, 032311 (2005). 29 H. A. Engel, V. N. Golovach, D. Loss, L. M. K. Vander- sypen, J. M. Elzerman, R. Hanson, and L. P. Kouwen- hoven, Phys. Rev. Lett. 93, 106804 (2004). 30 Y. Hu, H. O. H. Churchill, D. J. Reilly, J. Xiang, C. M. Lieber, and C. M. Marcus, Nature Nanotech. 2, 622 (2007). 

1310.1704

1

1310

Quantum calculations of the carrier mobility in thin films: Methodology, Matthiessen's rule and comparison with semi-classical approaches

We discuss the calculation of the carrier mobility in silicon films within the quantum Non-Equilibrium Green's Functions (NEGF) framework. We introduce a new method for the extraction of the carrier mobility that is free from contact resistance contamination, and provides accurate mobilities at a reasonable cost, with minimal needs for ensemble averages. We then introduce a new paradigm for the definition of the partial mobility $\mu_{M}$ associated with a given elastic scattering mechanism "M", taking phonons (PH) as a reference ($\mu_{M}^{-1}=\mu_{PH+M}^{-1}-\mu_{PH}^{-1}$). We argue that this definition makes better sense in a quantum transport framework as it is free from long range interference effects that can appear in purely ballistic calculations. As a matter of fact, these mobilities satisfy Matthiessen's rule for three mechanisms [surface roughness (SR), remote Coulomb scattering (RCS) and phonons] much better than the usual, single mechanism calculations. We also discuss the problems raised by the long range spatial correlations in the RCS disorder. Finally, we compare semi-classical Kubo-Greenwood (KG) and quantum NEGF calculations. We show that KG and NEGF are in reasonable agreement for phonon and RCS, yet not for SR. We point to possible deficiencies in the treatment of SR scattering in KG, opening the way for further improvements.

cond-mat.mes-hall

cond-mat

Quantum calculations of the carrier mobility in thin films: Methodology, Matthiessen's rule and comparison with semi-classical approaches Yann-Michel Niquet,1 , a) Viet-Hung Nguyen,1 Fran¸cois Triozon,2 Ivan Duchemin,1 Olivier Nier,3 and Denis Rideau3 1)L Sim, SP2M, UMR-E CEA/UJF-Grenoble 1, INAC, Grenoble, France 2)CEA, LETI-MINATEC, Grenoble, France 3)ST Microelectronics, Crolles, France We discuss the calculation of the carrier mobility in silicon films within the quan- tum Non-Equilibrium Green's Functions (NEGF) framework. We introduce a new method for the extraction of the carrier mobility that is free from contact resistance contamination, and provides accurate mobilities at a reasonable cost, with minimal needs for ensemble averages. We then introduce a new paradigm for the definition of the partial mobility µM associated with a given elastic scattering mechanism "M", taking phonons (PH) as a reference (µ−1 PH). We argue that this def- inition makes better sense in a quantum transport framework as it is free from long M = µ−1 PH+M − µ−1 range interference effects that can appear in purely ballistic calculations. As a mat- ter of fact, these mobilities satisfy Matthiessen's rule for three mechanisms [surface roughness (SR), remote Coulomb scattering (RCS) and phonons] much better than the usual, single mechanism calculations. We also discuss the problems raised by the long range spatial correlations in the RCS disorder. Finally, we compare semi- classical Kubo-Greenwood (KG) and quantum NEGF calculations. We show that KG and NEGF are in reasonable agreement for phonon and RCS, yet not for SR. We point to possible deficiencies in the treatment of SR scattering in KG, opening the way for further improvements. a)Electronic mail: yniquet@cea.fr 1 I. INTRODUCTION Device scaling has been a major trend in micro-electronics for almost fifty years, allowing for continuous performance and functionality enhancements. Complementary Metal-Oxide- Semiconductor1 (CMOS) transistors with gate lengths Lg in the 20 nm range are now man- ufactured at the industrial level thanks to the breakthroughs made in material and device processing. Micro-electronics is now facing new challenges.2 In particular, extreme scaling calls for innovative device architectures, such as fully-depleted silicon-on-insulator (FDSOI) transistors based on thin films,3,4 or multi-gate, short channel nanoscale transistors.5,6 In the sub-10 nm scale, the distinction between material and device modeling is getting increasingly blurred: the electronic and transport properties of the system sharply depart from those of bulk materials and become strongly dependent on the detailed device geometry. Quantum corrections start to be significant with strong sub-band quantization even in the weak in- version regime and source-to-drain tunneling. Confinement also enhances the interactions between, e.g., the silicon channel and the surrounding gate stack, affecting carrier mobilities ever more.7 -- 18 Modeling and simulation can play a prominent role in the design and understanding of these devices. The standard simulation toolbox is still, however, mostly based on classical (e.g., drift-diffusion19) and semi-classical simulation methods (Kubo-Greenwood,20,21 Monte- Carlo22 or deterministic23 solution of Boltzmann transport equation). This toolbox shall, therefore, be complemented with quantum transport methods, in order to assess the im- portance of quantum corrections, and in order to work out a multi-scale framework able to address problems at various scales with different levels of approximations. Non-Equilibrium Green's Functions (NEGF) is one of the most versatile approach for that purpose.24 In partic- ular, it can deal with quantum confinement, elastic scattering (surface roughness, impurities, ...) and inelastic scattering (phonons) in a seamless way. Although computationally expen- sive, Green's functions methods have benefited from recent advances in numerical methods and algorithms25 -- 28 and from the increasing availability of high performance computing infrastructures.29 They can, therefore, be applied to more and more realistic devices, and will certainly take an important place in the design of the ultimate technology nodes. Although NEGF is primarily intended for large bias (out-of-equilibrium) calculations,30 -- 32 it can also be used to compute carrier mobilities in a quantum framework. There is already 2 FIG. 1. Schematics of the simulated FDSOI devices. a lot of literature about carrier mobilities within the Green's functions framework, mostly on nanowires.8 -- 16 The carrier mobility is usually extracted from gate length scaling analyses,13 -- 15 or from a separation between the ballistic33,34 and diffusive components of the current.9 -- 12 In this work, we discuss the strengths and weaknesses of these approaches, and propose a new method that gives very accurate results at a reasonable cost (Sections II and III). We apply this methodology to thin silicon films (thickness in the 2-10 nm range), on devices with unprecedented size (length up to 95 nm and width up to 30 nm). We introduce a new paradigm for the definition of the partial mobility µM associated with a single elastic mechanism "M", taking phonons (PH) as a reference (µ−1 PH). We argue that this definition makes better sense in a quantum transport framework, as it mitigates M = µ−1 PH+M − µ−1 long range interference effects that can appear in ballistic (no phonons) calculations. We also discuss the problems raised by long range spatial correlations appearing in, e.g., remote Coulomb scattering (RCS, section IV) We show that our partial mobilities actually satisfy Matthiessen's rule much better than the usual, single mechanism calculations (Section V). Finally, we compare NEGF mobilities with KG calculations (Section VI). We show that NEGF and KG agree reasonably well for phonons and RCS scattering, yet not for SR. We discuss possible reasons for these discrepancies. 3 II. DEVICES AND METHODOLOGIES The devices considered in this work (Fig. 1) are undoped, (100) FDSOI films with thickness ranging from tSi = 2.5 nm to tSi = 10 nm. The buried oxide (BOX) below the film is 25 nm thick, and the n-doped (Nd = 1018 cm−3) silicon substrate acting as a back gate is grounded. The front gate (FOX) stack is made up of a layer of SiO2 (with thickness 1 < tSiO2 < 4 nm and dielectric constant ε = 3.9), and of a layer of HfO2 (with thickness tHfO2 = 2 nm and dielectric constant ε = 22). The current is computed in a self-consistent NEGF framework,24 on top of the effective mass approximation (EMA).35 The longitudinal mass is m∗l = 0.916 m0 and the transverse mass m∗t = 0.191 m0 in silicon. The effective mass in SiO2 is m∗ = 0.5 m0, and the Si/SiO2 barrier is 3.15 eV high. The NEGF equations are solved on a finite differences grid, in a fully coupled mode space approach.36 Details can be found in Appendix A. NEGF can deal with phonons,37 surface roughness9 (SR) and remote Coulomb scattering11 (RCS) in a seamless way. We account for intra-valley acoustic phonon scattering (defor- mation potential38 Dac = 14.6 eV), and for inter-valleys scattering by the 3 f -type and 3 g-type processes of Ref. 22. Random SR profiles are generated as in Ref. 9, with an exponential auto-covariance function:39 FSR(r) = hδh(R)δh(R + r)i = ∆2e−√2r/ℓc , (1) where δh(r) is the variation of the surface height at point r. The typical rms is ∆ = 0.47 nm and the correlation length ℓc = 1.3 nm (same parameters as in Ref. 40). The FOX Si/SiO2, SiO2/HfO2 and HfO2/gate interfaces are conformal, but the FOX and BOX interfaces are uncorrelated. As for RCS, we generate random distributions of charges at the SiO2/HfO2 interface (with a given density nRCS), then solve Poisson's equation for the RCS potential. The caveats of this solution will be discussed in section IV. The semi-classical Kubo-Greenwood calculations discussed in this work have been per- formed with the commercial "Sentaurus Device" solver of Synopsys.41 4 III. EXTRACTING MOBILITIES FROM NEGF CALCULATIONS A. Definitions and problems In general, the low-field resistance of a channel with length L can be written: R(L) = V I = Rc + R0 + L n1dµe , (2) where V is the (small) drain-source bias, I is the current, n1d is the carrier density per unit length, µ is the carrier mobility and e is the electron charge. The first term, Rc, is a "contact" resistance accounting for backscattering in the access areas and/or at the interface between the access areas and the channel. It is, therefore, extrinsic to the channel. The second term is the so-called "ballistic" resistance of the channel,33,34 while the third, ∝ L term is the classical "diffusive" resistance. At zero temperature, the resistance of a purely ballistic channel, R0 = 1/(G0Nm) = 12.9kΩ/Nm, is limited by the number of modes (1D sub-bands) Nm carrying current. At finite temperature T , assuming Maxwell-Boltzmann statistics and a single transport mass m∗ for all sub-bands:15 1 R0 = − 2e2 h Z dE(cid:18) ∂f ∂E(cid:19) t(E) = n1de2 √2πm∗kT , (3) where f (E) = exp[(E − µ)/kT ] is the distribution function, and t(E) the transmission function, which is equal to the number of 1D sub-bands at energy E. Although the above assumptions may not hold in general, Eq. (3) nicely illustrates the main trends followed by the ballistic resistance. In particular, both the diffusive and ballistic resistance decrease with the width W of 2D devices, since n1d = n2dW , where n2d is the sheet density in the channel. We emphasize, at this point, that Eq. (2) is only valid in "long enough" channels. L shall not only be larger than the mean free path ℓe to reach the diffusive regime; It must also be much larger than the typical correlation length ℓc of the disorder so that the carriers sample a representative set of configurations along their way from source to drain ("self-averaging"). Variability around Eq. (2) increases with decreasing L, and the mobility must primarily be understood as a long channel concept, or as an average figure for channels shorter than a few ℓc's. As discussed in paragraph IV, ℓc can range from a 1 − 2 nanometers (e.g., SR) to ≈ 10 nanometers (e.g., RCS). 5 In principle, the carrier mobility can be computed in different ways with a real space NEGF code. The ballistic resistance R0 can be obtained independently from a purely "bal- listic" calculation (no scattering), and the mobility extracted from the data for a single length L using Eq. (2).9 -- 12 This, however, presumes that the contact resistance Rc is negli- gible, and that the length L of the channel is perfectly well defined (where does the channel really start and end in the device ?). Alternatively, the current can be computed in channels with various lengths L, then Rc + R0 and µ fitted to the R(L) data (gate length scaling analysis).13 -- 15 This method is a priori immune to contact resistance contamination and to channel length misestimates. Indeed, a systematic error ∆L on the channel length will not change the slope of the R(L) data (hence the mobility), but only the apparent contact and ballistic resistance Rc + R0. Yet the R(L) data can be very noisy, since different L usually correspond to different realizations of the disorder (with different diffusive resistances but also possibly different Rc's). Therefore, accurate mobilities practically call for averages over large numbers of samples, especially if L can not be made very long. Also, as shown by Eqs. (2) and (3), the resistance of the channel has a prevalent 1/n1d behavior. It is, therefore, essential to compare data computed at the same carrier density (whatever the method). This raises at once the question "What is the carrier density in the channel ?", which is not trivial. In a disordered (e.g., rough) channel, the carrier density is intrinsically non uniform, hence not univocally defined. We therefore need to design a method that is free from the above limitations to the largest possible extent, i.e. i) free from contact resistance contamination; ii) free from channel length misestimates; iii) with minimal needs for ensemble averages, and iv) with a well defined prescription for the density. B. Methodology and example We have set-up the following methodology to deal with the above issues: We prepare a sample of disorder (e.g., surface roughness) with periodic boundary conditions over length Ls and width Ws. We then build devices with lengths L(N) = 2Lc + NLs made of this sample repeated N times and connected to access areas with length Lc on both source and drain sides. The resistance of these devices is therefore expected to follow an arithmetic progression R(N) = Rc + R0 + NRs, representative of a series of N segments with ballistic resistance 6 FIG. 2. (a) A periodic Ws = 20 nm × Ls = 30 nm sample of surface roughness. (b) The spatial (solid blue line) and target (dotted red line) auto-correlation functions of this sample, computed along the z (transport) axis. R0 and diffusive resistance Rs, connected to access areas with total resistance Rc. The noise on the R(N) data shall be minimal since all segments are identical, allowing for an accurate extraction of the sample resistance Rs and mobility µ = Ls/(n1deRs) according to Eq. (2). In particular, this method is completely free from contact resistance contamination as Rc shall be the same whatever N, all devices showing, by design, exactly the same interfaces between source/drain and channel. It is also immune to channel length misestimates, as the period Ls of the disorder is perfectly well defined. This shall decrease the number of samples needed to converge ensemble averages. For the sake of illustration, we focus on phonons+surface roughness (PH+SR) scattering in thin film devices -- though the methodology is valid in 1D trigate or nanowire devices as 7 FIG. 3. Carrier density in the N = 2 (L = 63.6 nm long) device at gate voltage Vgs = 1.6 V. From bottom to top, the BOX, Si film, SiO2, HfO2 and gate layers are clearly visible, as well as the 30 nm long SR sample, repeated twice. [tSi = 4 nm, tSiO2 = 2 nm] well. A typical, periodic Ws = 20 nm × Ls = 30 nm sample of SR is shown in Fig. 2, along with the target [Eq. (1)] and calculated spatial auto-correlation functions: F (r) = 1 S ZS d2R δh(R)δh(R + r) , (4) where S is the Si/SiO2 interface with surface S. We will come back in section IV to the importance of this function. We build devices made of this sample repeated once [L(1) = 33.6 nm], twice [L(2) = 63.6 nm, see Fig. 3] and up to three times [L(3) = 93.6 nm]. There is a Lc = 1.8 nm long, undisordered "contact" region on each side of the device, which has no sizable influence on the extracted Rs (though it has on Rc). Note that the devices are fully gated from left to right. The density of carriers is thus controlled by the electrochemical potential µs in the source, the doping density and the gate voltages. µs was set as the electrochemical potential in a (remote) source with n-type doping Nd = 5 × 1018 cm−3. The choice of µs is pretty irrelevant in single gate devices (this merely shifts the I(Vgs) characteristics along the front gate Vgs axis), but can have some impact on the mobility in double gate devices, due to the interplay between the front and back gate electric fields. As for the definition of n1d, the average density in the device practically yields the best (most linear) fits to the R(N) data. We therefore stick to this definition, which makes sense as the variations around the average density can be interpreted as the response (screening) 8 FIG. 4. Phonons+SR resistance (times device width) as a function of length for a sample like Fig. 2. The dotted red line is a linear regression with Eq. (2), which yields (Rc + R0)W = 38.04Ω.µm and µPH+SR = 303 cm2/V/s (carrier density n2d = 1013 cm−2). The shaded gray area is the 95% confidence interval. [tSi = 4 nm, tSiO2 = 2 nm] of the carriers to the disorder. Note that n1d is very little sensitive to the the length Lc of the contact regions, since there is no junction between a highly doped source/drain and the channel in the simulation box. We sweep the gate voltage from Vgs = 0 V to Vgs = 1.8 V and monitor the current, average density and effective electric field in the devices. The drain-source voltage is Vds(1) = 2 mV, Vds(2) = 4 mV, and Vds(3) = 6 mV, in order to extract the mobility at low, but constant longitudinal electric field. The average density and effective electric field are weakly dependent on the device length L (within ±1%); However, as discussed in section III A, the current has a strong dependence on n1d, so that it is best to compare resistances computed at the same density. Therefore, we first fit n1dR(n1d) with a spline (for each N), then interpolate R(n1d) on a grid of target densities. This slightly improves the quality of the linear regression on the R(N) data. As an example, the R(N) data computed for a specific SR sample are plotted in Fig. 4 (carrier density n2d = 1013 cm−2). As expected, the data lie on a straight line, allowing for an unambiguous fit (Rc + R0)W = 38.04 Ω.µm and µPH+SR = 303 cm2/V/s (±2.25% with 95% confidence). For practical purposes, we most often compute the N = 1 and N = 2 devices only, and check the N = 3 data for a few, critical cases. The calculated phonons+SR mobility is plotted as a function of carrier density in Fig. 5, for Ns = 3 different SR samples, 9 FIG. 5. The phonon+SR limited mobility as a function of carrier density, for three different SR samples with size Ws = 20 nm× Ls = 30 nm, along with the average mobility µ−1 µ−1 3 )/3. [tSi = 4 nm, tSiO2 = 2 nm] avg = (µ−1 1 + µ−1 2 + along with the average: µ−1 avg = 1 Ns Ns Xi=1 µ−1 i , (5) where µi is the mobility extracted from the ith sample. There is little variability -- hence practically no need for ensemble averages, as the samples are longer than the mean free path and much longer than the correlation length (ℓc = 1.3 nm) of the disorder. Using the relations µ = eτ /m∗, where τ is the average scattering time, and ℓe = vτ , where v is the thermal velocity 1 2 m∗v2 = kT , we indeed estimate mean free paths in the 2 − 19 nm range for mobilities 100 ≤ µ ≤ 800 cm2/V/s. Convergence with respect to the sample size will be discussed in more detail in the next subsection. The ballistic resistance R0W = 34.44 Ω.µm of the the same device has also been computed independently (switching off all scattering mechanisms). It is, as expected, significantly lower than the R(L = 0)W = 38.04 Ω.µm extrapolation of the NEGF data. This is due, primarily, to the resistance Rc associated with the two Lc = 1.8 nm long contacts on both sides of the disordered channel, and with the backscattering at the contact/channel interfaces. This might also be due to the mismatch between the ballistic resistance of the purely ballistic channel and the "ballistic" resistance of the rough channel, which will be limited by the thinnest parts of the device, and is therefore expected to be slightly larger. Since the above methodology requires at least two calculations on "long" devices, it is 10 FIG. 6. (a) The phonon+SR limited mobility extracted from our methodology and from single length calculations with different L's [Eq. (6)], and (b) the relative difference between our method- ology and the single length calculations. [tSi = 4 nm, tSiO2 = 2 nm] worth comparing the results with mobilities extracted from (cheaper) single length data as: µ(L) = L n2de[R(L) − Rc − R0]W . (6) Using the data of Fig. 4 with the extrapolated (Rc + R0)W = 38.04Ω.µm would of course yield exactly the same mobility µPH+SR = 303 cm2/V/s whatever L but presumes that we have computed at least two long devices already. Instead, we use our "best" simple estimate for Rc + R0, that is the phonon limited resistance of a very short 3.6 nm long channel (mimicking the two access areas on the source and drain sides). Accordingly, we use the length L = NLs of the channel in Eq. (6). The mobilities extracted for different L's are compared with our methodology on Fig. 6. As expected from Eq. (6), any error on Rc + R0 results in a ∝ 1/L correction on 1/µ, leading to an apparent length dependent mobility.10 11 In that particular case, the error is as large as 12% on the 30 nm long device, and is still ≈ 4% on the 90 nm long device. The error would be up to 25% in the 30 nm long device if we had used the ballistic resistance R0W = 34.44 Ω.µm as an approximation for Rc + R0. We therefore conclude that our methodology yields the best balance between accuracy and efficiency for NEGF mobility calculations. C. Phonons as a reference The above methodology yields the phonons+SR limited mobility µPH+SR. What about the phonons and SR limited mobilities ? The phonon limited mobility µPH can easily be extracted in the same way. The phonon limited resistance is, actually, strictly proportional to L, since the electron-phonon inter- action is "intrinsic" to the material, at variance with the "extrinsic" disorders such as SR and RCS.13,14 We might, arguably, compute the SR limited mobility along the same lines (simply switching off phonons in the above calculation). There are, however, two difficul- ties with this procedure. The first one is computational: Calculations are actually faster with than without phonons. Of course, one has to achieve both Poisson and "Born" (self- energies) self-consistency in NEGF calculations with phonons, which costs extra iterations (see Appendix A). On the other hand, inelastic scattering smooths Van Hove singularities in the spectral functions (local density of states, etc...), so that the number of energy points needed to integrate these spectral functions [e.g., Eq. (A3)] can be up to 10× smaller than in a "ballistic" (no phonons) calculation. Hence, with an efficient preconditioning scheme for the self-consistent loops, the computational cost of a calculation with phonons can be much lower than the one of a ballistic calculation. The second issue is about physics. Phonons do break phase coherence and mitigate localization and other long range interference effects.42 NEGF (or other quantum) calculations without phonons do not, therefore, necessarily give an accurate picture of the physics of the devices at room temperature. This is especially sensitive in 1D devices such as nanowires, but can also affect 2D devices. As hinted above, the electron-phonon interaction plays a particular role in the physics of the devices. It is the only inelastic and "intrinsic" scattering mechanism. Its interactions with elastic mechanisms (leading, in particular, to decoherence) suggest that we may choose it as a reference frame in quantum calculations. We therefore define an effective SR limited 12 FIG. 7. The phonon limited and phonon+SR limited mobilities computed with NEGF, and the effective SR limited mobility obtained from Eq. (7). [tSi = 4 nm, tSiO2 = 2 nm] mobility from Matthiessen's rule: µ−1 SR,eff = µ−1 PH+SR − µ−1 PH , (7) where µPH is the phonon limited mobility and µPH+SR the phonon+SR limited mobility, both computed with the above NEGF methodology. We stress that µPH and µPH+SR are unambiguously given by NEGF. Although the above µSR,eff might differ from a direct calculation, the combination of µSR,eff with the phonon limited mobility µPH yields the "exact" total mobility µPH+SR, which is the only important figure (if there are no other scattering mechanisms such as Coulomb traps). The validity of Matthiessen's rule has actually been much debated in the literature.43 -- 49 Yet, as shown in section V, µPH and the effective mobilities µSR,eff and µRCS,eff defined with respect to phonons can be combined with very good accuracy using Matthiessen's rule. This makes a clear case for Eq. (7) as a definition of the single mechanism mobility. As for surface roughness, the effective mobility obtained from Eq. (7) is plotted in Fig. 7 for a particular SR sample. The SR limited mobility follows the expected trend as a function of carrier density,38,40 with a plateau at small density (limited by both FOX and BOX SR), and a fast decrease at large density (due to strong inversion at the FOX interface). The mobilities computed on other samples with the same or different sizes are shown in Fig. 8. As expected from Fig. 5, there is little sample to sample variability, except possibly at low density/effective field where SR is not the dominant mechanism. The SR limited mobility 13 FIG. 8. The SR limited mobility as a function of effective field, computed in samples with different lengths Ls and widths Ws. [tSi = 4 nm, tSiO2 = 2 nm] FIG. 9. The SR limited mobility extracted from Eq. (7) for "normal" and "low" (√2× weaker deformation potentials) phonons. [tSi = 4 nm, tSiO2 = 2 nm] is converged in Ws = 20 nm × Ls = 30 nm samples. To demonstrate that Eq. (7) is meaningful, we have recomputed the effective SR mobility using deformation potentials √2 times weaker (that is a phonon limited mobility two times larger). As shown in Fig. 9, the resulting µSR is little affected (< 10%), showing that the methodology is robust. We will further discuss the applicability of Matthiessen's rule in section V. 14 FIG. 10. The RCS limited mobility as a function of carrier density, computed for different distri- butions of charges at the SiO2/HfO2 interface (positive charges with density nRCS = 1013 cm−2). [tSi = 4 nm, tSiO2 = 1 nm] IV. THE IMPORTANCE OF THE AUTO-CORRELATION FUNCTIONS: THE CASE OF REMOTE COULOMB SCATTERING In this section, we discuss the problems arising when the samples can not be much longer than the correlation length of the disorder, and possible workarounds. Remote Coulomb Scattering (RCS) is the scattering of carriers in the channel by remote charges at the SiO2/HfO2 interface.50 -- 54 It is believed to be a major limiting mechanism in high-κ gate stacks. It is modeled in Kubo-Greenwood solvers as the scattering by indepen- dent, uncorrelated point charges at the interface. It is, therefore, tempting (and straight- forward) to mimic the same assumptions in NEGF by randomly distributing test charges at the interface between SiO2 and HfO2, then solve Poisson's equation for the scattering potential Vtest(r). As in the case of SR, we generate periodic samples of RCS disorder in Ws = 20 nm × Ls = 30 nm supercells. The effective RCS mobilities extracted on different samples are plotted as a function of the carrier density in Fig. 10. In these devices, the interfaces are smooth, the SiO2 FOX oxide is only 1 nm thick (to enhance RCS) and the density of charges at the SiO2/HfO2 interface is nRCS = 1013 cm−2 (60 positive charges in the sample). The results of a Kubo- Greenwood calculation have been added for comparison [the same methodology, Eq. (7), was used to extract the mobility for consistency]. Although the trends are well reproduced 15 by NEGF, the sample to sample variability is impressive (> one order of magnitude). These data are clearly representative of the large, Coulomb-induced variability expected in short channel devices,55 yet they do not give any clue about the long channel mobility. These deficiencies can be understood from second-order perturbation theory (which un- derlies Kubo-Greenwood calculations and is the leading order of NEGF calculations). For the sake of simplicity, we discard electron-phonon interactions here. In the presence of a scalar random potential potential V (r), the Green's function of the film, G(E), can be expanded to second order in V and the unperturbed Green's function G0(E) as:16,56 G(E) = G0(E) + G0(E)V G0(E) + G0(E)V G0(E)V G0(E) + ... . (8) The ensemble averaged Green's function hG(E)i therefore reads: hG(E)i = G0(E) + G0(E)hV iG0(E) + G0(E)hV G0(E)V iG0(E) + ... , (9) where h...i is an average over different samples. Hamiltonian H0 and Green's function G0 = [E − H0]−1, we might as well write: Including hV i(r) into the unperturbed hG(E)i = G0(E) + G0(E)hδV G0(E)δV iG0(E) + ... , where δV (r) = V (r) − hV i(r). The above equation can be cast in the form: where: hG(E)i = G0(E) + G0(E)Σ(E)G0(E) + ... , Σ(r, r′, E) = G0(r, r′, E) hδV (r)δV (r′)i (10) (11) (12) is the self-energy associated with the random potential V (r). The above equations are nothing else than a non self-consistent, yet conserving Born approximation.57 At that level, the disorder can be completely characterized by the ensemble averaged auto-covariance function of the potential: In the case of RCS, F (r, r′) = hδV (r)δV (r′)i . ν(r − Ri) V (r) = N Xi=1 16 (13) (14) where ν(r − Ri) is the potential created by a single charge at position Ri, and N is the number of RCS charges in the sample. Assuming uncorrelated charge positions, N F (r, r′) = = Xi,j=1 Xi=1 N hδν(r − Ri)δν(r′ − Rj)i hδν(r − Ri)δν(r′ − Ri)i . If all positions Ri at the interface are equiprobable, F (r, r′) = N S ZS d2R δν(r − R)δν(r′ − R) , (15) (16) where S is the SiO2/HfO2 interface with area S → ∞. As expected, the scattering strength is proportional to the RCS charge density nRCS = N/S. We can finally introduce the in- k−rk, x, x′), plane coordinate rk and out-of-plane coordinate x, and write F (r, r′) = NFRCS(r′ where: FRCS(rk, x, x′) = 1 S ZS d2Rk δν(Rk, x)δν(Rk + rk, x′) (17) is the spatial auto-correlation function of a single RCS charge. The second-order Green's function hGi(E) [Eq. (11)] embeds all information about the average density and linear response current in an ensemble of many different realizations of the disorder. We do expect, however, fluctuations to vanish when S → ∞ (thermodynamic limit), so that the probability to find a sample that departs from the average goes to zero. Fig. 10 simply shows that the present Ws = 20 nm× Ls = 30 nm devices are far too small to kill those fluctuations, so that the number of samples Ns needed to converge the ensemble averaged mobility [Eq. (5)] is still very large at this scale. We can, unfortunately, neither simulate devices with widths and lengths in the µm range using NEGF, nor accumulate statistics on hundreds of samples. We might, therefore, attempt to design a minimal set of test potentials Vtest(r) that reproduces the effects of the leading second-order self-energy Σ(r, r′, E). This set of test potentials must satisfy: hδVtest(rk, x)i = 0 hδVtest(rk, x)δVtest(r′ k , x′)i = NFRCS(r′ k − rk, x, x′) . (18a) (18b) 17 Taking the in-plane Fourier transform on both sides, hδVtest(Kk, x)i = 0 hδV ∗test(Kk, x)δVtest(K′ k , x′)i = NFRCS(Kk, x, x′)δKk,K′ k = Nδν∗(Kk, x)δν(Kk, x′)δKk,K′ k , (19a) (19b) (19c) where: S ZS Note that the diagonal elements Kk = K′ k scattering strength. Hence we shall hopefully expedite convergence of the ensemble average d2rk Vtest(rk, x)e−iKk·rk . in Eq. (19c) play a key role in defining the Vtest(Kk, x) = (20) 1 by choosing test potentials that satisfy: hδVtest(Kk, x)i = 0 hδV ∗test(Kk, x)δVtest(K′ δV ∗test(Kk, x)δVtest(Kk, x′) = Nδν∗(Kk, x)δν(Kk, x′) , k, x′)i = 0 if Kk 6= K′ k The backward Fourier transform of the last equation reads: Ftest(rk, x, x′) = d2R δVtest(R, x)δVtest(R + rk, x′) 1 S ZS = NFRCS(rk, x, x′) . (21a) (21b) (21c) (22) In other words, the spatial auto-correlation function of each test potential, Ftest, shall match the ensemble averaged auto-covariance function of the disorder. As a matter of fact, this strategy is already widely used for surface roughness.9 The SR samples are generated so that the spatial auto-correlation function, F (rk) [Eq. (4)] matches the ensemble averaged auto-covariance function FSR(rk) [Eq. (1)]. In reciprocal space, FSR(Kk) = π∆2ℓ2 c ℓ2 c/2(cid:17)3/2 S(cid:16)1 + K2 = F (Kk) = δh(Kk)2. k (23) We can therefore choose δh(Kk) = eiϕ(Kk)pFSR(Kk), where ϕ(Kk) is a random phase, and transform back to real space to find a suitable SR profile δh(rk). How much can the spatial auto-correlation function of a random charge distribution, Ftest, be different from the target Ftarget = NFRCS ? For the purpose of comparisons, we define 18 FIG. 11. Spatial auto-correlation function ¯Ftest computed along the z axis for 10 test distributions of 60 RCS charges in a Ws = 20 nm × Ls = 30 nm sample. The target auto-covariance function ¯Ftarget is plotted for comparison. [tSi = 4 nm, tSiO2 = 1 nm] the x-averaged auto-correlation function: ¯F (rk) = 1 L2 x Z x1 x0 dxZ x1 x0 dx′ F (rk, x, x′) , (24) where Lx = x1 − x0 and the integration range, [x0, x1], is typically the upper half of the ¯Ftest is plotted for different random, film (which matters most in the inversion regime). unscreened test distributions on Fig. 11, and compared to ¯Ftarget. The size of the unit cell is Ws = 20 nm× Ls = 30 nm, and the density of RCS charges is nRCS = 1013 cm−2 (N = 60 positive charges). The auto-correlation of the test and target potentials can indeed be very different, which explains the huge variability seen in Fig. 10. This is due to the fact that the size of the sample is not very much larger than the decay length of the auto-correlation function (at variance with the SR profiles discussed in paragraph III B). A possible set of test potentials that satisfy Eqs. (21) is given by: δVtest(Kk, x) = √N eiϕ(Kk)δν(Kk, x) , (25) where ϕ(Kk) is a random phase shift. However, the backward Fourier transform δVtest(rk, x) of Eq. (25) solutions is not, in general, the potential created by a distribution of point charges at the SiO2/HfO2 interface. It might, hence, lead to inappropriate electronic response and electrostatics. In general, there is no distribution of point charges at the SiO2/HfO2 interface that fulfills Eq. (21c) exactly for all K, x and x′. To rank tentative charge distributions, we 19 FIG. 12. The RCS limited mobility as a function of carrier density, computed in samples with different lengths Ls and widths Ws, using optimized test charge distributions at the SiO2/HfO2 interface (positive charges with density nRCS = 1013 cm−2). [tSi = 4 nm, tSiO2 = 1 nm] therefore introduce the least square deviation σ2 between ¯Ftest and ¯Ftarget: σ2 = XKk (cid:0) ¯Ftest(Kk) − ¯Ftarget(Kk)(cid:1)2 , (26) where ¯Ftest(Kk) and ¯Ftarget(Kk) are the in-plane Fourier transforms of ¯Ftest and ¯Ftarget, com- puted for the unscreened potentials using Fast Fourier Transform on the finite differences grid. We then sample (in parallel) a large number (typically 8192) of random charge distri- butions, and select the one that minimizes σ2. As an example, the mobilities computed using different optimized test charge distributions are plotted in Fig. 12 (same film as in Fig. 10). The variability is definitely improved, and the NEGF data are much closer to the KG results (a more detailed comparison will be made in paragraph VI B). This figure also shows the influence of the width Ws of the samples. The sample must be at least 20 nm wide to get reliable results. The auto-correlation functions of the four Ws = 20 nm samples of Fig. 12 are given in Fig. 13, along with a test RCS potential plotted 1 nm below the Si/SiO2 interface. The target auto-correlation function is, indeed, well reproduced by the four samples. The above methodology also applies to nanowires and other 1D structures (the Kk vector then runs along the nanowire axis). When both surface roughness and RCS come into play, the film or wire is not homogeneous and the potential ν(r − Ri) created by a single RCS charge depends on its position Ri. In that case, we first average ¯Ftarget(Kk) over (typically 20 FIG. 13. (a) A test RCS potential, plotted 1 nm below the Si/SiO2 interface. (b) Spatial auto- correlation function ¯Ftest computed along the z axis for 4 optimized distributions of 60 RCS charges in a Ws = 20 nm × Ls = 30 nm sample. Each optimal distribution was selected out of 8192 configurations, generated with a different random seed. The target auto-correlation function ¯Ftarget is plotted for comparison. [tSi = 4 nm, tSiO2 = 1 nm] 8192) random charge positions at the interface. To conclude, we have emphasized the importance of having well characterized auto- correlation functions for the disorder in NEGF mobility calculations. This is particularly critical for mechanisms such as RCS, which exhibit long-range correlations. 21 FIG. 14. (a) Total NEGF mobility µNEGF tot = µNEGF PH+SR+RCS including phonons, SR and RCS (nRCS = 2.5 1013 cm−2), compared with Matthiessen's law µM tot on the effective mobilities [Eq. (28)]. Matthiessen's laws on phonons+SR, phonons+RCS and SR+RCS are also plotted for com- pleteness (they are, by design, equivalent to the corresponding NEGF data in the first two cases). The data for µSR,eff , µRCS,eff and µNEGF tot were averaged over 4 configurations to reduce the im- pact of residual variability, shown as a shaded area around each curve. (b) The error made by Matthiessen's law on the total mobility. [tSi = 4 nm, tSiO2 = 2 nm] V. ABOUT MATTHIESSEN'S RULE Matthiessen's rule states that the inverse of the total mobility µM tot is simply the sum of the inverse of the partial mobilities computed for each different scattering mechanism (e.g. µPH for phonons, µSR for SR and µRCS for RCS): 1 µM tot = 1 µPH + 1 µSR + 1 µRCS . 22 (27) As discussed in section III C, the validity of Matthiessen's rule has been questioned in the literature.43 -- 49 Indeed, the mobility µtot computed with all scattering mechanisms at once can depart significantly from µM tot. As a matter of fact, Eq. (27) is exact in a Kubo-Greenwood framework only if i) the scattering rates of all mechanisms have the same dependence on energy, and ii) these mechanisms are independent one from each other (no spatial correla- tions). Although the second condition usually holds for phonons, SR and RCS, the first one does not. Esseni and Driussi48 have reported errors > 50% for phonons+Coulomb scatter- ing (at low density) and around 10 − 15% for phonons+SR (at high density) in bulk MOS transistors. We argue, however, that the effective SR and RCS mobilities defined by Eq. (7) follow Matthiessen's rule much better than the usual, single mechanism partial mobilities. This is illustrated in Fig. 14, which compares Matthiessen's rule on the effective mobilities: 1 µM tot = 1 µPH + 1 µSR,eff + 1 µRCS,eff (28) with a NEGF calculation including all mechanisms at once. To reduce the impact of residual variability, the data were averaged over 4 different configurations for each single mechanism (SR, RCS) and for the total NEGF mobility µtot. Mathiessen's holds within 3% in the whole density range. Similar agreement was obtained on a 7.5 nm thick film, even though the current shows a stronger multi-valley and multi-band character (a condition known to hinder Matthiessen's rule47,48), and on a 10 × 10 nm square nanowire in a trigate configuration. Of course, one expects Eq. (28) to be valid whenever one of the two elastic mechanisms dominates the other. In the high density range for example, where RCS is negligible, µSR,eff reproduces, by construction, the total mobility µtot ≃ µPH+SR. However, Fig. 14 shows that Matthiessen's rule also holds when SR and RCS have comparable strengths. Inelastic scattering by phonons (included in all calculations) indeed smooths spectral quantities such as the local density of states and the current density, hence reducing the mismatch between different mechanisms that otherwise plagues Matthiessen's rule. This further motivates the use of Eq. (7) as a definition of the contribution of a given mechanism to the mobility, and opens the way for a more accurate modeling of the mobility in Technology Computer Aided Design (TCAD) tools. 23 FIG. 15. The phonon limited mobility as a function of film thickness, at three effective fields, computed within NEGF (lines) and Kubo-Greenwood (symbols). [tSiO2 = 2 nm] VI. COMPARISONS BETWEEN KUBO-GREENWOOD AND NEGF CALCULATIONS We conclude this paper with a comparison between Kubo-Greenwood and NEGF calcu- lations. We successively discuss phonons, RCS and SR limited mobilities. A. Phonons KG and NEGF phonon limited mobilities are plotted as a function of Si film thickness tSi in Fig. 15, for different effective electric fields. They have been computed in the same devices, with the same effective mass model and the same material parameters. As already discussed in Ref. 18, KG and NEGF are in very good agreement about electron-phonon scattering down to very thin films or wires. In particular, the mobility overshoot around tSi = 3 nm, which is due to the depletion of the heavy ∆′ valleys off-Gamma into the light ∆ valleys at Γ, is consistent in both approaches. B. Remote Coulomb scattering As shown in Fig. 12, the KG model of Sentaurus Device is in reasonable agreement with NEGF. In particular, the slope of the µRCS(n2d) curves, which characterizes how the RCS potential is screened, is almost the same in both methods. The NEGF mobility is, however, 24 FIG. 16. The RCS limited mobility as a function of carrier density, for different tSi, computed within NEGF and Kubo-Greenwood. [tSiO2 = 1 nm] systematically larger than the KG mobility (whatever the distribution of RCS charges at the interface). The average difference is around 30% over the whole density range. Nonetheless, KG and NEGF show comparable quantitative trends as a function of tSi (and tSiO2), as evidenced in Fig. 16. Overall, the present KG and NEGF simulations are in line with Ref. 54, but tend to indicate that the impact of RCS on the mobility is much lower than anticipated from the data of Ref. 53.58 C. Surface roughness The NEGF and KG surface roughness limited mobilities are plotted as a function of the effective field in Fig. 17, for different film thicknesses tSi (∆ = 0.47 nm and ℓc = 1.3 nm). As for RCS, the effective KG mobility is computed like the NEGF mobility, from Eq. (7), but the same conclusions can be reached from a direct KG calculation. It is clear that the NEGF mobility is significantly smaller than the KG mobility, and that the NEGF mobility decreases much faster than the KG mobility when thinning the film. In this respect, the trends followed by the NEGF mobility are much closer to the experimental data of Uchida et al..59,60 This strong decrease of the mobility is due to the scattering by film thickness fluctuations. The disagreement between KG and NEGF might result from i) the semi-classical nature 25 FIG. 17. The SR limited mobility as a function of effective field, for different tSi, computed within NEGF (lines) and Kubo-Greenwood (symbols). [tSiO2 = 2 nm, ∆ = 0.47 nm, ℓc = 1.3 nm] FIG. 18. The NEGF SR limited mobility as a function of the effective field, computed for ∆ = 0.47 nm and ∆ = 0.47/√2 = 0.33 nm [ℓc = 1.3 nm, tSi = 4 nm, tSiO2 = 2 nm]. The mobility shows an almost perfect ∝ 1/∆2 scaling, as evidenced by the red dashed line. of Boltzmann transport equation and distribution function; ii) the second-order perturba- tion theory (Fermi Golden Rule) behind KG; and iii) the approximations made in the SR Hamiltonian in KG. We now examine each of these possibilities. The SR, RCS and electron-phonon scattering rates show, admittedly, very different energy and wave vector dependences, but this hardly explains why Boltzmann transport equation reproduces the phonon and RCS limited mobility, but not the SR limited mobility. We therefore presume that the semi-classical nature of this equation is not the primary source 26 of discrepancies with NEGF. As for perturbation theory, Fermi Golden Rule predicts a characteristic, 1/∆2 behavior for the SR limited mobility, which does not hold at higher order. Interestingly, the NEGF data also show an almost perfect 1/∆2 scaling around the investigated rms ∆ = 0.47 nm (see Fig. 18). Therefore, Fermi Golden Rule seems to be valid for SR scattering. Hence, approximations on the SR Hamiltonian are most likely responsible for the differences between KG and NEGF. Indeed, SR is explicit in NEGF calculations, and no approximations are made on the SR Hamiltonian and screening (beyond the mean field approximation for Coulomb interactions also made in KG). In KG calculations, SR and screening are "implicit" in the sense that they act on a smooth reference device. There are a few approximations in the SR potential (see, e.g., the very detailed derivation of Ref. 40), whose quantitative accuracy shall be assessed in detail. We suggest that this is the primary target to improve the KG results. VII. CONCLUSIONS We have proposed a new method for the calculation of carrier mobilities in a real space NEGF framework that provides accurate results at a reasonable cost. We have also intro- duced a new paradigm for the definition of the partial mobility associated with a given elastic scattering mechanism, based on Matthiessen's rule with phonons as a reference. We argue that this definition makes better sense in a quantum transport framework, as it mitigates long range interference effects that appear in purely ballistic calculations. As a matter of fact, the partial mobilities obtained that way satisfy Matthiessen's rule for two and three mechanisms much better than the usual, direct single mechanism calculations. We have also emphasized the need for well characterized auto-correlation functions for the disorder in NEGF, in order to mimic the thermodynamic (large samples) limit as best as possible. Finally, we have compared our NEGF mobilities with KG calculations. NEGF and KG are in good agreement for the phonons and RCS, yet not for the SR limited mobilities. We suggest that the weaknesses of KG lie in the approximations made in the treatment of the SR potential and screening, opening the way for further improvements. We thank S. Barraud for fruitful discussions about RCS. This work was supported by the French National Research Agency (ANR project Quasanova). The NEGF calculations were run at the TGCC/Curie machine using allocations from PRACE and GENCI,61 and on the 27 "Froggy" platform of the CIMENT infrastructure in Grenoble.62 Appendix A: Details about the NEGF implementation In this appendix, we give details about our implementation of the NEGF equations. 1. NEGF equations and implementation The lesser Green's function Gv<(E), greater Green's function Gv>(E) and retarded Green's function Gvr(E) of each valley v embed all information about the charge and current density at energy E.24 On the finite differences grid, they become matrices that satisfy: [E − H v + eV − Σvr Gv≷(E) = Gvr(E)(cid:2)Σv≷ b (E) − Σvr b (E) + Σv≷ s (E)] Gvr(E) = I s (E)(cid:3) Gvr(E)† , (A1a) (A1b) where H is the EMA Hamiltonian, V is the electrostatic potential, Σb(E) is the "boundary" self-energy describing the source and drain contacts, and Σs(E) is the "scattering" self- energy accounting for inelastic electron-phonon interactions. The electrostatic potential V (r) satisfies Poisson's equation: ∇rε(r)∇rV (r) = −4πn(r) , (A2) where n(r) is the charge density. The electronic charge Qi at each point Ri of the finite differences grid can be computed from the diagonal elements Gv< ii (E) of the lesser Green's function: Qi = −e ImXv Z +∞ −∞ dE 2π Gv< ii (E) . (A3) The current in the device can be computed along the same lines.24 Only 1 nm of SiO2 is included in the EMA hamiltonian, on each side of the film. The mesh is homogeneous in this subdomain, with 2A step, and non homogeneous outside (where only Poisson's equation is solved and the wave functions are assumed to be zero). Electrons-phonons, surface roughness and remote Coulomb scattering can be included in the calculations. As for phonons, we use the usual diagonal approximation37 for the self-energy Σs. For intra-valley acoustic phonons, Σv≷ s,ii(E) = 1 Ωi kT D2 ac ρv2 s Gv≷ ii (E) , 28 (A4) where kT is the thermal energy, Dac = 14.6 eV is the acoustic deformation potential,38 ρ = 2.33 g/cm3 is the density of silicon, vs = 9000 m/s is the longitudinal sound velocity,22 and Ωi is the elementary volume around point Ri of the finite differences grid. For an inter-valleys acoustic or optical phonon, Σv≷ s,ii(E) = 1 Ωi op )2 (Dvv′ 2ρω hhNiGv′≷ ii (E ± ω) + (hNi + 1)Gv′≷ ii (E ∓ ω)i , (A5) where ω is the energy of the phonon mode, hNi is the average number of phonons in this mode, and Dvv′ op is a deformation potential. We account for the 3 f -type and for the 3 g-type inter-valleys processes of Ref. 22. We approximate the retarded electron-phonon self-energy Σvr s as the antihermitic part of (Σv> Born-von-Karman (periodic) boundary conditions are applied in the transverse y direc- s − Σv< s )/2.63 tion. The Green's functions can therefore be written: Gij(E) = Xky Gij(ky, E)eiky(yj−yi) , (A6) where ky is the transverse wave vector and G(ky, E) has the periodicity of the "supercell" used for the calculation. The latter is typically Ws = 20 nm wide, and the first Brillouin zone is sampled with 3 ky points. While H must be replaced with the Bloch Hamiltonian H(ky) in Eqs. (A1), the scattering self-energy Σs remains independent of ky. Dirichlet boundary conditions (constant potential) are applied on the gates, while standard, Neumann boundary conditions (zero normal electric field) are applied along the transport direction.64 Equations (A1) are solved with a standard Recursive Green's Functions method65 in a fully coupled mode space approach36 (192 modes for each valley) on CPUs or graphics cards units (GPU). The latter are highly specialized, parallel units that can process linear algebra operations much faster than traditional CPU cores, thus enabling significant speed-ups.66,67 that the scattering self-energy Σ< Note that the potential V depends on the charge density in the device, hence on G<, and s also depends on G<. One therefore needs to achieve s . In particular, failure to achieve self-consistency on Σ< s breaks current conservation (source and drain currents are different).68,69 It is a common self-consistency on both V and Σ< practice to reach self-consistency on Σ< s for a given V before making any change to the potential,30,31 so that the Green's functions used to update V are conserving. However, this strategy, which alternates updates on Σ< s and V , considerably increases the number of iterations needed to achieve global self-consistency (typically > 50). We find that updating 29 Σ< s and V at each iteration -- even if the Green's functions used to compute the density are not yet conserving -- expedites convergence to the same fixed point in only 5 to 25 iterations depending on the bias conditions. The Green's function G< computed at a given iteration is directly used as input for the self-energy Σ< s of the next iteration, while the variations of the potential V are damped with a Newton-Raphson-like correction (see below).64 The modes are also updated at each iteration to account for the changes in the potential. The convergence criteria are i) variations of the charge density < 0.01% for three consecutive iterations, ii) variations of the average current < 0.1%, and iii) current conserved within 1% along the device. Integrations such as Eq. (A3) are performed on a regular grid of 256 energy points extending from µs − 0.2 eV to µs + 16kT , where µs is the chemical potential of the source. With these parameters, we estimate the error on the current to be < 0.25%. The code is parallelized over the loops on ky points and energies. 2. Newton-Raphson-like correction to the potential Poisson's equation can formally be written:64 ∇rε(r)∇rV (r) = −4πn[V ](r) , (A7) where we have emphasized that n(r) is a functional of the potential V (r). This non-linear problem can in principle be solved with the Newton-Raphson method. Starting from an arbitrary potential V0(r), we look for a correction δV (r) = V (r) − V0(r) such that: ∇rε(r)∇rδV (r) = −4πn[V0 + δV ](r) − ∇rε(r)∇rV0(r) . We then linearize the right-hand side: n[V0 + δV ](r) = n[V0](r) +Z d3r′D[V0](r, r′)δV (r′) , (A8) (A9) where D[V ](r, r′) = δn[V ](r)/δV (r′) is the functional derivative of n[V ](r) with respect to V (r′). The resulting equation can be solved on the finite differences grid with standard linear algebra routines. We next iterate from the new solution V1(r) = V0(r) + δV (r), until convergence. The Newton-Raphson method converges in principle much faster than straightforward fixed-point iteration (that is, solving ∇rε(r)∇rV1(r) = −4πn[V0](r) and iterating until self-consistency). 30 We do not know, however, the explicit form of the functional n[V ] and of its derivatives. Yet we might design approximations for D[V ](r, r′), solve Eqs. (A8) and (A9) for the input potential Vm(r) and the output density n[Vm](r) of the mth NEGF iteration, and use the resulting Vm+1(r) = Vm(r) + δV (r) as input for the (m + 1)th iteration. Since Vm+1(r) anticipates over the response of the density through the D[V ](r, r′) kernel, we shall hopefully achieve self-consistency much faster. Practically, we make a local density approximation for D(r, r′): where: D(r, r′) = e dn0(r, µ(r)) dµ δ(r − r′) , n0(r, µ) = −eZ dE ρ0(r, E)fFD(E − µ) , (A10) (A11) ρ0(r, E) is the local density of states and fFD is the Fermi-Dirac distribution function. In Eq. (A10), the local chemical potential µ(r) is computed so that n0(r, µ(r)) matches n[Vm](r) at each point r. We further approximate ρ0(r, E) as ρs(x, y, E), the local density of states in the source, in the absence of external potential, computed once for all at the beginning.70 This approximation for D(r, r′) takes quantum confinement into account and is therefore more accurate than a Fermi integral formula for n0.64 It makes an excellent preconditioner for the self-consistent Poisson iteration, even far out of equilibrium.15 REFERENCES 1T. Ytterdal, C. Y. Hua, and T. A. Fjeldly, Device Modeling for Analog and RF CMOS Circuit Design (Wiley, Chichester, 2003). 2"The international technology roadmap for semiconductors (itrs)," http://www.itrs.net/. 3O. Faynot, F. Andrieu, O. Weber, C. Fenouillet-Beranger, P. Perreau, J. Mazurier, T. Benoist, O. Rozeau, T. Poiroux, M. Vinet, L. Grenouillet, J.-P. Noel, N. Posseme, S. Barnola, F. Martin, C. Lapeyre, M. Casse, X. Garros, M. A. Jaud, O. Thomas, G. Cibrario, L. Tosti, L. Brevard, C. Tabone, P. Gaud, S. Barraud, T. Ernst, and S. Deleonibus, in Electron Devices Meeting (IEDM), 2010 IEEE International (2010) pp. 3.2.1 -- 3.2.4. 31 4N. Planes, O. Weber, V. Barral, S. Haendler, D. Noblet, D. Croain, M. Bocat, P. Sassoulas, X. Federspiel, A. Cros, A. Bajolet, E. Richard, B. Dumont, P. Perreau, D. Petit, D. Golan- ski, C. Fenouillet-Beranger, N. Guillot, M. Rafik, V. Huard, S. Puget, X. Montagner, M. A. Jaud, O. Rozeau, O. Saxod, F. Wacquant, F. Monsieur, D. Barge, L. Pinzelli, M. Mellier, F. Boeuf, F. Arnaud, and M. Haond, in VLSI Technology (VLSIT), 2012 Symposium on (2012) pp. 133 -- 134. 5J.-P. Colinge, Solid-State Electronics 48, 897 (2004). 6S. Barraud, R. Coquand, M. Casse, M. Koyama, J. Hartmann, V. Maffini-Alvaro, C. Com- boroure, C. Vizioz, F. Aussenac, O. Faynot, and T. Poiroux, Electron Device Letters, IEEE 33, 1526 (2012). 7S. Jin, M. V. Fischetti, and T. wei Tang, J. Appl. Phys. 102, 083715 (2007). 8S. Poli, M. Pala, T. Poiroux, S. Deleonibus, and G. Baccarani, Electron Devices, IEEE Transactions on 55, 2968 (2008). 9C. Buran, M. Pala, M. Bescond, M. Dubois, and M. Mouis, Electron Devices, IEEE Transactions on 56, 2186 (2009). 10S. Poli and M. Pala, Electron Device Letters, IEEE 30, 1212 (2009). 11S. Poli, M. Pala, and T. Poiroux, Electron Devices, IEEE Transactions on 56, 1191 (2009). 12M. P. Persson, H. Mera, Y.-M. Niquet, C. Delerue, and M. Diarra, Phys. Rev. B 82, 115318 (2010). 13M. Aldegunde, A. Martinez, and A. Asenov, Journal of Applied Physics 110, 094518 (2011). 14M. Luisier, Appl. Phys. Lett. 98, 032111 (2011). 15V.-H. Nguyen, F. Triozon, F. Bonnet, and Y.-M. Niquet, Electron Devices, IEEE Trans- actions on 60, 1506 (2013). 16J. H. Oh, S.-H. Lee, and M. Shin, Journal of Applied Physics 113, 233706 (2013). 17N. Neophytou and H. Kosina, Physical Review B 84, 085313 (2011). 18Y.-M. Niquet, C. Delerue, D. Rideau, and B. Videau, Electron Devices, IEEE Transactions on 59, 1480 (2012). 19R. Granzner, V. Polyakov, F. Schwierz, M. Kittler, R. Luyken, W. Rosner, and M. Stadele, Microelectronic Engineering 83, 241 (2006). 20R. Kubo, Journal of the Physical Society of Japan 12, 570 (1957). 21D. A. Greenwood, Proceedings of the Physical Society 71, 585 (1958). 32 22C. Jacoboni and L. Reggiani, Reviews of Modern Physics 55, 645 (1983). 23S.-M. Hong and C. Jungemann, Journal of Computational Electronics 8, 225 (2009). 24M. P. Anantram, M. S. Lundstrom, and D. E. Nikonov, Proceedings of the IEEE 96, 1511 (2008). 25M. Luisier, A. Schenk, W. Fichtner, and G. Klimeck, Phys. Rev. B 74, 205323 (Nov 2006). 26S. Li, S. Ahmed, and E. Darve, Journal of Computational Electronics 6, 187 (2007). 27K. Kazymyrenko and X. Waintal, Phys. Rev. B 77, 115119 (2008). 28S. Cauley, M. Luisier, V. Balakrishnan, G. Klimeck, and C.-K. Koh, Journal of Applied Physics 110, 043713 (2011). 29M. Luisier and G. Klimeck, in High Performance Computing, Networking, Storage and Analysis, 2008. SC 2008. International Conference for (2008) pp. 1 -- 10. 30M. Luisier and G. Klimeck, Phys. Rev. B 80, 155430 (2009). 31N. Cavassilas, F. Michelini, and M. Bescond, Journal of Applied Physics 109, 073706 (2011). 32N. Dehdashti Akhavan, I. Ferain, R. Yu, P. Razavi, and J.-P. Colinge, Journal of Compu- tational Electronics 11, 249 (2012). 33S. Datta, F. Assad, and M. Lundstrom, Superlattices and Microstructures 23, 771 (1998). 34M. Shur, Electron Device Letters, IEEE 23, 511 (2002). 35G. Bastard, Wave mechanics applied to semiconductor heterostructures, Monographies de physique (Les ´Editions de Physique, 1988). 36J. Wang, E. Polizzi, and M. Lundstrom, Journal of Applied Physics 96, 2192 (2004). 37S. Jin, Y. J. Park, and H. S. Min, Journal of Applied Physics 99, 123719 (2006). 38D. Esseni, A. Abramo, L. Selmi, and E. Sangiorgi, Electron Devices, IEEE Transactions on 50 (2003). 39S. M. Goodnick, D. K. Ferry, C. W. Wilmsen, Z. Liliental, D. Fathy, and O. L. Krivanek, Physical Review B 32, 8171 (1985). 40S. Jin, M. Fischetti, and T.-W. Tang, Electron Devices, IEEE Transactions on 54, 2191 (2007). 41"Sentaurus device," http://www.synopsys.com/Tools/TCAD/DeviceSimulation/Pages/SentaurusDevice.aspx. 42B. Kramer and A. MacKinnon, Reports on Progress in Physics 56, 1469 (1993). 43W. Walukiewicz, L. Lagowski, L. Jastrzebski, M. Lichtensteiger, and H. C. Gatos, Journal of Applied Physics 50, 899 (1979). 33 44F. Stern, Phys. Rev. Lett. 44, 1469 (1980). 45Y. Takeda and T. Pearsall, Electronics Letters 17, 573 (1981). 46A. K. Saxena and M. A. L. Mudares, Journal of Applied Physics 58, 2795 (1985). 47M. V. Fischetti, F. Gamiz, and W. Hansch, Journal of Applied Physics 92, 7320 (2002). 48D. Esseni and F. Driussi, Electron Devices, IEEE Transactions on 58, 2415 (2011). 49M.-J. Chen, W.-H. Lee, and Y.-H. Huang, Electron Devices, IEEE Transactions on 60, 753 (2013). 50F. Gamiz, J. B. Roldan, J. E. Carceller, and P. Cartujo, Applied Physics Letters 82, 3251 (2003). 51D. Esseni and A. Abramo, Electron Devices, IEEE Transactions on 50, 1665 (2003). 52M. Casse, L. Thevenod, B. Guillaumot, L. Tosti, F. Martin, J. Mitard, O. Weber, F. An- drieu, T. Ernst, G. Reimbold, T. Billon, M. Mouis, and F. Boulanger, Electron Devices, IEEE Transactions on 53, 759 (2006). 53S. Barraud, O. Bonno, and M. Casse, Journal of Applied Physics 104, 073725 (2008). 54P. Toniutti, P. Palestri, D. Esseni, F. Driussi, M. D. Michielis, and L. Selmi, Journal of Applied Physics 112, 034502 (2012). 55A. Asenov, A. R. Brown, G. Roy, B. Cheng, C. Alexander, C. Riddet, U. Kovac, A. Mar- tinez, N. Seoane, and S. Roy, Journal of Computational Electronics 8, 349 (2009). 56G. D. Mahan, Many-Particle Physics, 2nd ed. (Plenum, New York, N.Y., 1993). 57H. Mera, M. Lannoo, C. Li, N. Cavassilas, and M. Bescond, Phys. Rev. B 86, 161404 (2012). 58There is possibly a factor 1/(4π2) missing in Eq. (35) [hence a factor 1/(2π) missing in Eq. (36)], and a factor gs = 2 accounting for spin degeneracy missing in Eq. (38), which would result in an underestimation of the RCS mobility by a factor 4π in Ref. 53. 59K. Uchida, J. Koga, R. Ohba, T. Numata, and S.-I. Takagi, in Electron Devices Meeting, 2001. IEDM '01. Technical Digest. International (2001) pp. 29.4.1 -- 29.4.4. 60K. Uchida and S. ichi Takagi, Applied Physics Letters 82, 2916 (2003). 61http://www-hpc.cea.fr/en/complexe/tgcc-curie.htm. 62http://ciment.ujf-grenoble.fr. 63A. Svizhenko and M. P. Anantram, Electron Devices, IEEE Transactions on 50, 1459 (2003). 64Z. Ren, R. Venugopal, S. Goasguen, S. Datta, and M. Lundstrom, Electron Devices, IEEE 34 Transactions on 50, 1914 (2003). 65A. Svizhenko, M. P. Anantram, T. R. Govindan, B. Biegel, and R. Venugopal, Journal of Applied Physics 91, 2343 (2002). 66R. Nath, S. Tomov, and J. Dongarra, International Journal of High Performance Comput- ing Applications 24, 511 (2010). 67S. Tomov, J. Dongarra, and M. Baboulin, Parallel Computing 36, 232 (2010). 68G. Baym and L. P. Kadanoff, Phys. Rev. 124, 287 (1961). 69G. Baym, Phys. Rev. 127, 1391 (1962). 70ρs(E) = −Im Gs(E + iη)/π is computed from the retarded bulk Green's function Gs of the source, a by-product of standard decimation routines for the contact self-energy. A large imaginary part η is added to the energy to smooth out rapid variations of the density of states. 35 

1807.03177

3

1807

Scanning Gate Microscopy in a Viscous Electron Fluid

We measure transport through a Ga[Al]As heterostructure at temperatures between 0.1 K and 30 K. Increasing the temperature enhances the electron-electron scattering rate and viscous effects in the two-dimensional electron gas arise. To probe this regime we measure so-called vicinity voltages and use a voltage-biased scanning tip to induce a movable local perturbation. We find that the scanning gate images differentiate reliably between the different regimes of electron transport. Our data are in good agreement with recent theories for interacting electron liquids in the ballistic and viscous regimes stimulated by measurements in graphene. However, the range of temperatures and densities where viscous effects are observable in Ga[Al]As are very distinct from the graphene material system.

cond-mat.mes-hall

cond-mat

Scanning Gate Microscopy in a Viscous Electron Fluid B. A. Braem,1, ∗ F. M. D. Pellegrino,2, 3 A. Principi,4 M. Roosli,1 C. Gold,1 S. Hennel,1 J. V. Koski,1 M. Berl,1 W. Dietsche,1 W. Wegscheider,1 M. Polini,5 T. Ihn,1 and K. Ensslin1 1ETH Zurich, Solid State Physics Laboratory, Otto-Stern-Weg 1, 8093 Zurich, Switzerland 2Dipartimento di Fisica e Astronomia, Universit`a di Catania, Via S. Sofia, 64, I-95123 Catania, Italy 3INFN, Sez. Catania, I-95123 Catania, Italy 4School of Physics and Astronomy, University of Manchester, Manchester, M13 9PL, United Kingdom 5Istituto Italiano di Tecnologia, Graphene Labs, Via Morego 30, I-16163 Genova, Italy (Dated: December 24, 2018) We measure transport through a Ga[Al]As heterostructure at temperatures between 32 mK and 30 K. Increasing the temperature enhances the electron-electron scattering rate and viscous effects in the two-dimensional electron gas arise. To probe this regime we measure so-called vicinity voltages and use a voltage-biased scanning tip to induce a movable local perturbation. We find that the scanning gate images differentiate reliably between the different regimes of electron transport. Our data are in good agreement with recent theories for interacting electron liquids in the ballistic and viscous regimes stimulated by measurements in graphene. However, the range of temperatures and densities where viscous effects are observable in Ga[Al]As are very distinct from the graphene material system. Inter-particle collisions dominate the behavior of flu- ids as described by hydrodynamic theory [1]. In de- generate, clean two-dimensional electron gases (2DEGs), e.g. realized in Ga[Al]As heterostructures or in graphene, hydrodynamic behavior may be expected if electron- electron interaction is the dominant scattering mecha- nism. At millikelvin temperatures, however, electron- impurity scattering dominates over electron-electron scattering. The latter produces only small corrections ac- counted for within Fermi-liquid theory, a description in- volving weakly interacting quasiparticles. The relevance of electron-electron scattering is enhanced by increasing the temperature, thus softening the Fermi surface. The electron-electron scattering length lee then reaches well below both the geometric device sizes and the momen- tum relaxation length. Early experiments realized this regime aiming at the identification of hydrodynamic ef- fects in Ga[Al]As 2DEGs [2, 3]. Very recently, exper- imental signatures of viscosity due to electron-electron interaction have been found in graphene [4, 5], Ga[Al]As [6], PdCoO2 [7], and WP2 [8], and related theories have been developed [9 -- 13]. Viscous flow gives rise to intricate spatial flow patterns occurring at length scales well below the Drude scattering length lD, beyond which the momentum of the electronic system is dispersed [9 -- 11]. Such spatial patterns in elec- tronic systems have been theoretically predicted, but so far not been imaged experimentally. This motivates us to perform scanning gate microscopy [14, 15] measure- ments on a 2DEG in a Ga[Al]As heterostructures with signatures of viscous charge carrier flow. We find that the scanning gate measurement distinguishes the ballis- tic and viscous regimes of transport with high sensitiv- ity. In the viscous regime, the scanning tip can locally revive ballistic contributions to the measured signals by introducing new and tunable length scales to the system geometry. Both a hydrodynamic and a ballistic model of electron transport guide us in interpreting the experi- mental data. Following the experiments by Bandurin et al. [4, 12] on graphene, we use vicinity voltage probes close to a local current injector to measure effects of viscosity. The concept of the measurement is sketched in Fig. 1(a). We pass a current I from the source contact through a 300 nm wide orifice into a 5 µm wide channel, which is connected to the drain contact at ground potential. The upper channel boundary has three additional openings to probe the vicinity voltages Vj at a distance dj from the current-injecting orifice with dj being 600 nm, 1200 nm, and 2400 nm respectively. The vicinity voltages Vj are measured with respect to the reference potential Vref at the right end of the channel. In this geometry one ex- pects positive vicinity voltages for diffusive and ballistic electron motion in the channel, and negative values if electron-electron interaction is dominant [4, 9, 13]. In the latter case back-flow currents are proposed [11] as indicated by the schematic flow pattern in Fig. 1(a). We use a Ga[Al]As heterostructure with a 2DEG buried 130 nm below the surface and a back-gate to tune the electron density n [16]. The supplemental material provides experimental details, e.g. measurement param- eters, and electron density as a function of back-gate voltage. Applying negative voltages to the top-gates de- fines the structure shown in Fig. 1(a) by locally depleting the 2DEG. To measure the vicinity voltages we use low- noise voltage amplifiers and standard lock-in techniques at 31.4 Hz. We cool the sample in a cryostat equipped with an atomic force microscope to create a local pertur- bation by scanning gate microscopy (SGM). We define the vicinity resistance as the ratio Rj = Vj/I of the measured quantities, without offset-subtraction. Figure 1(b) shows the vicinity resistances normalized to 2 Figure 2. (a) Schematic of transport regimes as a function of temperature and electron density. Viscous effects are ex- pected at a high ratio lD/lee. Red lines mark contours of lD/lee and show the increase with T and n. The green shade marks the ballistic regime where both lD and lee exceed the channel width. Dashed grey lines indicate Dν = dj. Three black dots mark the parameters of the SGM measurements in panels (b)-(d). The data shown in Fig. 1(b) is measured along the dashed black line, the blue shade indicates the tem- perature range of negative R1200. (b)-(d) Vicinity resistance R1200 as a function of SGM tip position x, y with white color marking the value in the absence of the tip: (b) At 32 mK we observe a V-shape of reduced R1200(x, y) along the white dashed lines, which mark the ballistic trajectory. Dotted lines mark the outlines of the gates, areas of green color indicate tip positions leading to I = 0 or disconnected voltage probe. (c) At 7.9 K the vicinity resistance R1200(x, y) shows a maximum instead of the V. (d) R1200(x, y) at 7.9 K at lower electron density. describing the behavior along this line. Their applicabil- ity depends on the ratio lee/dj. The regime lee < dj realized for T >∼ 6 K is described by the viscous theory [4, 9, 10]. Numerical calculations as in Ref. 4 based on the solution of the Navier-Stokes equa- tion result in the flow patterns shown in Fig. 1(c) for our sample geometry. The intrinsic length scale of the theory Dν = (cid:112)leelD/4 was chosen to match the experimental conditions at about 7 K. The theory predicts negative vicinity resistances of R600/ρ = −0.65, R1200/ρ = −0.11, and R2400/ρ = −0.015, which are in qualitative agree- ment with the measurements in Fig. 1(b). With increas- ing temperature or dj, Dν falls below dj and the vicinity voltage probes become insensitive to the quasi-local vis- cous effects. This is in accordance with Rj/ρ in Fig. 1(b) tending towards zero for high T . For lee > dj, i.e. T <∼ 4 K, diffusive transport between the injector and the voltage probe is not effective yet, and Figure 1. (Color online) (a) Top-gates (indicated by black lines) deplete the 2DEG to shape the sample to a channel with orifices to the top region, which serve as current injec- tor and voltage probes. The vicinity voltages Vj are mea- sured with respect to the channel potential Vref . Arrows in- dicate schematically the current distribution if back-flow oc- curs due to viscosity. The dashed rectangle marks the area where the tip of the scanning gate microscope is scanned. (b) Normalized vicinity resistances Rj/ρ := (Vj − Vref )/Iρ as a function of temperature in the absence of the SGM tip at n = 1.2 × 1011 cm−2. The inset shows the same data en- larged to highlight the minima at around 7 K. The vertical dashed lines mark the temperatures of the SGM measure- ments in Fig. 2. (c) Current distribution and potential from solving the hydrodynamic model with a length scale parame- ter Dν = 1.25 µm, which corresponds to n = 1.2 × 1011 cm−2 and T ≈ 7 K. The green lines mark equipotential surfaces forming the contacts to the channel. the 2DEG sheet resistance ρ as a function of tempera- ture T from 30 mK to 30 K. At the lowest temperature, all vicinity resistances are positive. With increasing T their signs change at around 3 K. The temperature of the zero-crossing increases with dj. Furthermore, the vicin- ity resistances have a minimum at around 7 K and tend towards zero with increasing T . This behavior is similar to recent experiments in bilayer graphene [4, 12]. To understand the behavior of the vicinity resistances as a function of temperature in Fig. 1(b) we consider the scattering lengths lee and lD of the 2DEG realized within the range of our experimental parameters. Fig- ure 2(a) displays red contour lines of the ratio lD/lee, where lee = vFτee was calculated from τee [17, 18] and the Drude scattering length lD was extracted from bulk resistance measurements (absolute values of lee and lD in supplemental material). One can see that lD/lee (cid:29) 1 in an extended region of the parameter space indicating where electron-electron interactions dominate. The hori- zontal dashed line marks the density of the measurement shown in Fig. 1(b). Two complementary theories exist (a)(b)R/rj0temperature (K)0102030V 600V 1200V 2400drainsourceV ref(c)IV refD=1250 nmnI(V -V)/Irref01-22-12mmV 600V 1200V 24005mmchannel-224680101020R600R1200R2400SGM-1-2(b)0100R (W)12001234123015050-50(c)012345123405R (W)12000-20-10-30(d)x (mm)012345y (mm)123405R (W)1200102030104011-2n = 1.2310cmT = 32 mK11-2n = 1.2310cmT = 7.9 K11-2n = 0.3310cmT = 7.9 Kx (mm)x (mm)y (mm)y (mm)0T (K)(a)00.51.05101511-2n (10 cm)(c)(b)1(b)(d)l/lDee3010viscousballistic561200 nm600 nmD= 2400 nmnsingle electron-electron scattering events will dominate the measured vicinity voltages. This regime is described by the theory of Shytov et al. [13]. They propose that the vicinity voltage response is negative with its strength increasing with the electron-electron scattering rate, i.e. with temperature. This is in qualitative agreement with the strongly decreasing Rj around 3 K in Fig. 1(b). At temperatures below 1.7 K, lee exceeds the width of the channel of our sample and both of the above men- tioned theories become inapplicable. An extended the- ory covering the full range of temperatures [12] proposes that the positive vicinity voltage observed in the experi- ment is caused by ballistic electron motion between the injector orifice and the voltage probe with intermittent reflection at the opposite channel boundary. This claim is supported by the SGM measurements presented below. We now scan the SGM tip at a fixed height of 40 nm above the GaAs surface in the area indicated by the dashed rectangle in Fig. 1(a). Applying a negative volt- age to the tip creates a disk of depleted 2DEG with a diameter of approximately 300 nm. We have taken scan- ning gate images for a range of back-gate voltages, con- tact configurations and channel widths, but in the inter- est of brevity we present data for the three selected, most significant regimes marked by the black dots in Fig. 2(a). Figure 2(b) shows the vicinity resistance R1200 as a function of the tip position x, y at T = 32 mK, in the ballistic regime where lD ≈ 36 µm and lee (cid:29) lD. White color presents R1200 as measured in the absence of the tip. Blue indicates a reduced, and red an increased value of R1200. The black contour at zero highlights the tip posi- tions of sign inversion. For orientation, black dotted lines mark the outlines of the top-gates. If the tip depletes the 2DEG in the source orifice or in the voltage probe open- ing, R1200 cannot be extracted and the position is colored green. The classical ballistic electron trajectory from the source to the voltage probe, that is once reflected by the channel gate, is indicated by white dashed lines. We ob- serve a V-shaped reduction of R1200 along the outline of this ballistic path. We interpret the result in the fol- lowing way: In the absence of the tip, some electrons are ballistically reflected by the channel gate into the voltage probe and we measure positive R1200. For tip positions along the V-shaped ballistic path, the tip potential de- flects ballistic trajectories and we observe a reduction of R1200(x, y). Conversely, a tip positioned outside the V guides additional trajectories into the voltage probe and thus increases R1200(x, y). Such a deflection of ballistic trajectories has been demonstrated by earlier SGM work [19 -- 21]. We change to the viscous regime by heating the cryo- stat temperature to 7.9 K such that lD ≈ 16 µm and lee ≈ 0.4 µm < dj, leading to a characteristic length scale Dν = 1.2 µm. Figure 2(c) shows the corresponding SGM measurement. The striking difference to Fig. 2(b) wit- nesses the change of the transport regime from ballistic 3 Figure 3. All three vicinity resistances at T = 7.9 K and n = 1.2 × 1011 cm−2 as a function of tip position: (a) R600, (b) R1200 as already shown in Fig. 2(c), and (d) R2400. As indicated by the dashed lines, we find a maximum of Rj when the tip forms an equilateral triangle with the source orifice and the voltage probe. to viscous. The V-shaped reduction of R1200 is no longer present. Consistent with the measurements in Fig. 1(b), R1200(x, y) is negative if the tip is far from source orifice or voltage probe, for example at x > 5 µm. In contrast to measurements at lower temperature, R1200(x, y) features a maximum at x ≈ y ≈ 2 µm. This distinguished posi- tion is approximately separated by d1200 from both the source orifice and the voltage probe. Here the tip forms a scattering site much closer than the lower channel edge at y ≈ −2 µm. We now reduce the electron density to n = 0.3 × 1011 cm−2 while keeping the temperature at 7.9 K (see the point labeled (d) in Fig. 2(a).) At this low den- sity, lD ≈ 1.6 µm and lee ≈ 70 nm (cid:28) dj, and the char- acteristic scale Dν = 170 nm has fallen well below dj. Therefore we do not observe the effects of viscosity but a positive vicinity resistance in the absence of the tip. SGM at this low density finds R1200(x, y) presented in Fig. 2(d), which is significantly different to both the re- sult in (b) and (c) at four times higher electron density. Instead of a maximum we find a R1200(x, y) minimum at x ≈ 2 µm, y ≈ 2.3 µm. In Fig. 3 we return to the high-density regime and com- pare all three vicinity resistances Rj measured at 7.9 K. Note that Fig. 3(b) reproduces Fig. 2(c) for convenience. The dashed lines form an equilateral triangle between the current-injecting orifice and the respective vicinity voltage probe. The tip of the triangle coincides with the maximum of Rj in all three images, suggesting a purely geometrical interpretation. It seems that the presence of the tip-induced potential in this symmetry point pre- vents the observation of viscous effects and reestablishes a positive vicinity voltage. In conjunction with Figs. 1(b) and 2(a) we have al- ready discussed the microscopic transport regimes which we now found to result in dramatic differences in the scanning gate images in Figs. 2(b)-(d). In the remaining parts of the paper, we discuss the imaging mechanism of the scanning gate technique in the viscous regime rep- resented by Figs. 2(c) and (d). Naively one could think that the scanning tip-induced potential introduces a new (a)x (mm)012345y (mm)1234051234512345(c)(b)R (W)1200010-10-20-30R (W)600020-40-20-60R (W)240005-10-5x (mm)x (mm)4 Figure 5. Classical trajectories: (a) Color plot showing the potential landscape in the 2DEG from tip and top-gates from finite element simulation. Red lines show classical trajectories starting at the green line in the source lead and ending in one of the voltage probes. (b) The number of trajectories ending in the voltage probes weighted by the trajectory length. the signal R600 if the tip is close to the respective voltage probe. As in the high-density case, we find a disagree- ment if the distance between the tip and the orifices is of the order of Dν. Since the hydrodynamic model does not describe ballistic effects, we consider this as a justi- fication for the hypothesis, that the presence of the tip leads to a revival of ballistic effects in the sample on the small length scale introduced by the tip. j j To test this hypothesis, we investigate ballistic contri- butions in a deliberately oversimplified classical model. We calculate electron trajectories emanating from the source orifice in the electrostatic potential of gates and tip exemplarily shown in Fig. 5(a). For tip positions along the dashed line we count the number of trajec- tories that end in one of the voltage probes as a quali- tative measure for the ballistic contribution Rbal to the corresponding vicinity resistance. We count each tra- jectory with a weight that decreases exponentially with trajectory length to account for electron-electron scatter- ing (details in supplemental material). Figure 5(b) shows the resulting maxima of Rbal for the tip positions in the middle between the source orifice and the corresponding voltage probe. This is in agreement with the experimen- tal observations at high density in Fig. 4(a), when the tip is close to the orifices. It supports our speculative inter- pretation that the resistance maxima in Fig. 3 result from an enhancement of ballistic contributions to the conduc- tance, which quench the visibility of the viscous effects. In summary, we have presented measurements of neg- ative vicinity resistances in Ga[Al]As heterostructures, which indicate viscous behavior. By increasing the tem- perature we observed the transition from the ballistic to the viscous regime when the electron-electron scattering length falls below the separation between current injector and voltage probes. These findings are qualitatively sim- ilar to observations on graphene samples, but both the charge carrier density and the characteristic temperature are an order of magnitude lower. The movable perturba- Figure 4. Comparison between experiment and hydrodynam- ical model: (a), (b) Rj along the dashed lines in Figs. 2(c) and (d), the x-coordinates xj of source orifice and voltage probes are marked by the vertical lines. (c), (d) Vicinity resistances calculated with the hydrodynamic model for the tip positions and length scales Dν in the experimental data of (a) and (b). The horizontal dotted lines denote the vicinity resistances in the absence of the tip. internal sample boundary, which leads to a reorganiza- tion of the viscous flow pattern and thereby to a change in the vicinity voltages. We will therefore discuss the agree- ment and differences between the hydrodynamic model in Fig. 1(c) and the scanning gate measurements first. The hydrodynamic model solves for the stationary flow of the classical incompressible viscous electron liquid at very low Reynolds numbers, where the non-linear con- vective acceleration term in the Navier-Stokes equation can be neglected. Thanks to the addition of a Drude- like momentum relaxation rate, the resulting equations are well suited to describe the transition from the vis- cous to the momentum-scattering dominated regime [9]. However, this model does not account for ballistic effects. We solve the model in the presence of a local Lorentzian- shaped decrease of the electron density caused by the tip potential [22] (details of the tip implementation in sup- plemental material). In Fig. 4 we compare the measured vicinity resistances along the dashed lines in Figs. 2(c) and (d) with the prediction of the model for the same tip positions and length scales Dν. For orientation, the vertical lines mark the x-coordinates of the source orifice and the voltage probes. In the high-density case in (a), (c) we find qual- itative agreement for tip positions x > 4 µm, but not at x < 3 µm where the distance between the tip and the orifices is of the order of Dν and no longer (cid:29) lee. We speculate that the disagreement originates from the close, tip-induced scattering site which revives ballistic effects. In the low-density case (b), (d) we find a rough agree- ment for all tip positions for R1200 and R2400, but not for (a)(b)0123450R (W)j2040600123450R (W)j-2080100-40R/rj(c)(d)01234500123450-0.10.10.150.05R600R600R1200R2400R2400R1200R/rjR1200R2400R600R1200R2400R600xsx600x1200x2400x600x1200x2400D = 1250 nmnD = 170 nmnexperimentexperimentx (mm)x (mm)x (mm)x (mm)xstheorytheory11-2n = 0.3310cm11-2n = 1.2310cm(a)(b)x (mm)0123450xsx600x1200x240011-2n = 1.2310cmtippotentialchannel gatee.V(meV)01EFermi234V2400V1200V600bal-3R (10)j55 arXiv:1806.09538 [cond-mat] (2018). [14] M. A. Eriksson, R. G. Beck, M. Topinka, J. A. Katine, R. M. Westervelt, K. L. Campman, and A. C. Gossard, Applied Physics Letters 69, 671 (1996). [15] M. A. Topinka, B. J. LeRoy, S. E. J. Shaw, E. J. Heller, R. M. Westervelt, K. D. Maranowski, and A. C. Gossard, Science 289, 2323 (2000). [16] M. Berl, L. Tiemann, W. Dietsche, H. Karl, and W. Wegscheider, Applied Physics Letters 108, 132102 (2016). [17] T. Jungwirth and A. H. MacDonald, Phys. Rev. B 53, 7403 (1996). [18] G. Giuliani and G. Vignale, Quantum Theory of the Elec- tron Liquid, 1st ed. (Cambridge University Press, Cam- bridge, 2008). [19] R. Crook, C. G. Smith, M. Y. Simmons, and D. A. Ritchie, Phys. Rev. B 62, 5174 (2000). [20] K. E. Aidala, R. E. Parrott, T. Kramer, E. J. Heller, R. M. Westervelt, M. P. Hanson, and A. C. Gossard, Nature Physics 3, 464 (2007). [21] S. Bhandari, G.-H. Lee, A. Klales, K. Watanabe, T. Taniguchi, E. Heller, P. Kim, and R. M. Westervelt, Nano Lett. 16, 1690 (2016). [22] M. A. Eriksson, R. G. Beck, M. A. Topinka, J. A. Katine, R. M. Westervelt, K. L. Campman, and A. C. Gossard, Superlattices and Microstructures 20, 435 (1996). [23] M. A. Topinka, B. J. LeRoy, R. M. Westervelt, S. E. J. Shaw, R. Fleischmann, E. J. Heller, K. D. Maranowski, and A. C. Gossard, Nature 410, 183 (2001). [24] C. W. J. Beenakker and H. van Houten, Solid State Physics 44, 1 (1991). [25] G. F. Giuliani and J. J. Quinn, Phys. Rev. B 26, 4421 (1982). tion by SGM introduces an additional, competing length scale. Scanning gate images in the ballistic and viscous regimes are markedly different. By forming a scattering site close to the source orifice and the voltage probes, ballistic effects can be restored even though the electron- electron scattering length is below the channel width. A hydrodynamic model explains some of the observed fea- tures including the negative vicinity resistances. From the difference between this model and the experiment we find that residual ballistic effects need to be consid- ered on small length scales even at a high temperature of 7.9 K. The theory developed in Ref. [12] based on the kinetic equation is well suited to describe the transition between the ballistic and the viscous regime of transport. It therefore remains an interesting open question, if this approach could be used for describing the scanning gate experiment, and if it yields agreement with the experi- ment over a larger range of parameters. We thank Leonid Levitov and Yigal Meir for valuable discussions. The authors acknowledge financial support from ETH Zurich and from the Swiss National Science Foundation (NCCR QSIT, SNF 2-77255). ∗ bbraem@phys.ethz.ch [1] L. D. Landau and E. M. Lifshitz, Fluid Mechan- ics, Second Edition: Volume 6, 2nd ed. (Butterworth- Heinemann, Amsterdam u.a, 1987). [2] L. W. Molenkamp and M. J. M. de Jong, Solid-State Electronics 37, 551 (1994). [3] M. J. M. de Jong and L. W. Molenkamp, Phys. Rev. B 51, 13389 (1995). [4] D. A. Bandurin, I. Torre, R. K. Kumar, M. B. Shalom, A. Tomadin, A. Principi, G. H. Auton, E. Khestanova, K. S. Novoselov, I. V. Grigorieva, L. A. Ponomarenko, A. K. Geim, and M. Polini, Science 351, 1055 (2016). [5] J. Crossno, J. K. Shi, K. Wang, X. Liu, A. Harzheim, A. Lucas, S. Sachdev, P. Kim, T. Taniguchi, K. Watan- abe, T. A. Ohki, and K. C. Fong, Science 351, 1058 (2016). [6] G. M. Gusev, A. D. Levin, E. V. Levinson, and A. K. Bakarov, AIP Advances 8, 025318 (2018). [7] P. J. W. Moll, P. Kushwaha, N. Nandi, B. Schmidt, and A. P. Mackenzie, Science 351, 1061 (2016). [8] J. Gooth, F. Menges, C. Shekhar, V. Suss, N. Ku- mar, Y. Sun, U. Drechsler, R. Zierold, C. Felser, and B. Gotsmann, arXiv:1706.05925 [cond-mat] (2017), arXiv: 1706.05925. [9] I. Torre, A. Tomadin, A. K. Geim, and M. Polini, Phys. Rev. B 92, 165433 (2015). [10] L. Levitov and G. Falkovich, Nature Physics 12, 672 (2016). [11] F. M. D. Pellegrino, I. Torre, A. K. Geim, and M. Polini, Phys. Rev. B 94, 155414 (2016). [12] D. A. Bandurin, A. V. Shytov, G. Falkovich, R. K. Ku- mar, M. B. Shalom, I. V. Grigorieva, A. K. Geim, and L. S. Levitov, arXiv:1806.03231 [cond-mat] (2018). [13] A. Shytov, J. F. Kong, G. Falkovich, and L. Levitov, Supplemental Materials: Scanning Gate Microscopy in a Viscous Electron Fluid This supplemental material contains information exceeding the scope of the main text. We provide details about the experimental methods and results of the electron transport in the bulk 2DEG. Furthermore, we show the absolute values of the electron-electron scattering length and the Drude scattering length as a function of the parameters T and n. The last two sections describe details of the hydrodynamic model and the classical trajectory calculations. S1 CONTENTS Acknowledgments References Experimental methods Controlling the sample temperature Electron density and mobility as a function of back-gate voltage at base temperature 5 5 S1 S3 S4 Drude mean free path and electron-electron scattering length as a function of temperature and electron density S5 Hydrodynamic model with locally reduced electron density Trajectory calculations S6 S7 EXPERIMENTAL METHODS We use a Ga[Al]As heterostructure with a 2DEG buried 130 nm below the surface and a grown back-gate 1.13 µm below the 2DEG [S16]. A voltage Vbg applied to the back-gate tunes the bulk electron density according to n = (1.21+ Vbg/1.67 V) × 1011 cm−2 (see the third section for details). At Vbg = 0 V the electron mobility is 6.2 × 106 cm2/Vs at 32 mK. On this heterostructure we define 35 nm high TiAu top-gates by electron-beam lithography. To deplete the 2DEG underneath, we apply a gate voltage of Vtop−gates = −0.18 × Vbg − 0.5 V with respect to the 2DEG potential (n = 1.2 × 1012 cm−2: Vbg = 0.0 V and Vtip = −8.0 V, n = 0.3 × 1012 cm−2: Vbg = −1.5 V and Vtip = −2.75 V) . To keep the tip-induced potential roughly proportional to the Fermi energy at all electron densities, we apply a tip voltage Vtip = −8 V − 3.5 × Vbg with respect to the 2DEG potential. Such a negative Vtip depletes the 2DEG below the tip, which is supported by the observation of the pattern of branched electron flow [S23] at base temperature. The corresponding maps of the two-terminal conductance as a function of tip position are shown in Fig. S1. We estimate the tip depletion diameter to be approximately 300 nm from choosing Vtip more negative than the depleting voltage. The finite element simulations used for the ballistic model (Fig. 5 in the main text) confirm this estimate: Figure S2 shows a vertical cut through the tip position and we find that the electrostatic potential (blue) exceeds the Fermi energy over a distance of approximately 300 nm. Outside the depletion disk, the tip induced potential approaches zero within a distance of 1 µm around the tip. Because the local electron density in the 2DEG is proportional to the difference of the Fermi energy and the electrostatic potential, it is reduced with respect to the bulk value within 1 µm around the tip. At larger distances, the tip induced density modulation is smaller than fluctuations expected from the random background potential present in GaAs 2DEGs. The red line in Fig. S2 indicates the Lorentzian potential describing the tip in the hydrodynamic model described in the last section. Each of the measurement cables to the source and drain contact (see Fig. 1(a) of the main text) has a resistance of 10 kΩ from the cold RC-filter. To determine the two-terminal resistance of the current injector orifice, we use two additional measurement leads which allow for the current-free measurement of the voltage between source and drain contact. To remain in the linear transport regime, we apply a small voltage of 100 µV to the room temperature ends of the cables. In the absence of the tip, the cable resistance (2 × 10 kΩ) dominates over the two-terminal resistance of the injector (depending on the electron density: 1.5 kΩ to 10 kΩ). Therefore the voltage between source and drain contact is much smaller than the applied 100 µV and the current I is limited by the filter resistors to Imax = S2 Figure S1. Two-terminal conductance G of the source orifice measured while measuring the vicinity resistances presented in Figs. 2(b) and (c) in the main text. (a),(b) G measured at base temperature with two different color scale settings to highlight the weak pattern of branched electron flow marked by the arrows. (c) G at T = 7.9 K showing similar behavior as in (a) and thus confirming the same invasiveness of the tip-induced potential as at base temperature. Due to the high temperature the branch pattern is reduced to a weak dip marked by the arrow. Figure S2. The blue line shows the Comsol-calculated potential from Fig. 5(a) of the main text (vertical cut through the tip position). Red shows the Lorentzian approximation of the potential as described by eq. (S3). The Fermi energy is marked by the black horizontal line. 100 µV/(2 × 10 kΩ) = 5 nA. When the tip approaches and depletes the current injector, we reach I = 0 and the voltage between source and drain contact is 100 µV. At base temperature, we observe at least 6 conductance plateaus conductance of each of the four orifices from the modes of the quantum point contacts. At T > 4 K the quantum point contact modes are obscured by thermal smearing. (a)0G2(2e/h)12341230(c)x (mm)012345y (mm)123405024611-2n = 1.2310cmT = 32 mK11-2n = 1.2310cmT = 7.9 Kx (mm)y (mm)0246810(b)0G2(2e/h)1234123011-2n = 1.2310cmT = 32 mKx (mm)y (mm)0246810G2(2e/h)81012y (mm)0123450potential (meV)5 lithographic channel widthLorentzianpotentialComsol calculatedpotentialE = 4.3 meVFermi6tipS3 Figure S3. Coulomb blockade measurement of a quantum dot in the SGM setup to determine the electron temperature at a cryostat temperature of T = 25 K. (a) Sample current I as a function of the plunger gate voltage Vgate and the bias voltage V between source and drain. (b) High-resolution measurement of I as a function of Vgate (red circles) at V = 2 µV. The blue line is the fit according to eq. (S1) to extract the electron temperature. (c) Te from fitting I(Vgate) as a function of V lie below 30 mK at small source-drain voltage. The red dot marks the example in panel (b). CONTROLLING THE SAMPLE TEMPERATURE In experiments at low temperatures, differences between the electronic temperature of the sample and the cryostat can arise. In the following we will describe in detail, how we determined the temperature, how the cryostat temperature was controlled, and why we can assume to have only a negligible difference between the electronic sample temperature and the temperature measured on the sample stage. We characterized the electronic temperature in our SGM setup by measuring Coulomb blockade resonances (using a different GaAs sample with a top-gate defined quantum dot) with exactly the same wiring and filtering of the electric signals. Figure S3 describes how we extract the electron temperature Te from fitting the current I across a Coulomb blockade resonance with I(Vgate) = I0 cosh2(αgate(Vgate − V0)/2kBTe) (S1) with I0 and V0 being the current and the gate voltage at the maximum of I(Vgate) following Ref. S24. The lever arm αgate describes the capacity of the gate to the quantum dot, its value is extracted from the slope of the blue dotted lines in Fig. S3(a). From fitting as shown in Fig. S3(b) we obtain the electron temperature as a function of source-drain voltage V as shown in Fig. S3(c). At a mixing chamber temperature of 25 mK (according to the Oxford Instruments thermometry), we extract an electronic temperature below 30 mK thanks to our improvements of the thermal anchoring of the cabling. The AFM cabling is thermally anchored similarly to the sample cabling and there is no indication of additional heating due to the presence of the tip. This supports our assumption to have only a negligible difference between electronic temperature and the cryostat temperature at millikelvin temperatures. At higher temperatures, the difference of electronic temperature and cryostat temperature typically decreases due to the better thermal conductance across material interfaces and through insulating materials (which are the main two problems for thermal anchoring at millikelvin temperatures). For the measurements at base temperature, we rely on the small difference between electron temperature and cryostat temperature shown in the Coulomb blockade measurements. In the following we describe our measures to reach higher temperatures, e.g. 7.9 K for the measurements in Figs. 2(c),(d) and Fig. 3 of the main text. First, we withdraw the mixture from the dilution unit. Second, we determine the sample temperature by measuring the resistance of a Lakeshore RX-202A RuOx thermometer mounted to the sample stage by standard lock-in technique. This thermometer is separated by less than 1 cm from the chip carrier and by approximately 28 cm from the mixing chamber plate. Third, we heat the mixing chamber plate using the heater installed by Oxford instruments, but a software-controllable voltage source. A PI-controller controls the heater voltage to achieve the desired sample temperature. We achieve a temperature stability of ±5 mK at 7.9 K, which is limited by the resolution of the lock- in amplifier measuring the RuOx resistor. To obtain the measurements at 7.9 K, we first heat the cryostat to this temperature for two days to ensure thermalization of the AFM components. Then we approach the tip to the sample and wait for additional 36 h before starting the measurement to avoid drifts and tip crashes. This slow procedure ensures a small temperature difference between the measured temperature and the electronic temperature of the sample. 0V (mV)0.51.0-0.5-1.0-699.2-699.0V (mV)gate15I (pA)1050datafit-702V (mV)gate-700-69820V (mV)60T (mK)e100-10-2050403020V =2mV(a)(b)(c)-50050I (pA)T =25 mK T S4 Figure S4. Hall density and electron mobility as a function of Vbg at T = 32 mK. The vertical arrow marks the onset of a second 2DEG forming at Vbg > 0.8 V. For the measurement as a function of temperature shown in Fig. 1(b), we do not control the temperature to each single point but heat the mixing chamber plate slowly and record the sample stage temperature together with the measurement data. The measurement shown in Fig. 1(b) has been obtained during a slow warm-up from 32 mK to 30 K over 24 h to ensure a slow heating and good thermalization of the sample and the sample stage. ELECTRON DENSITY AND MOBILITY AS A FUNCTION OF BACK-GATE VOLTAGE AT BASE TEMPERATURE We extract the electron density and mobility in Fig. S4 from standard longitudinal and Hall resistance measurements at base temperature. For back-gate voltages in the range Vbg = −1.8 V to 0.8 V we find the linear increase of n(Vbg) as expected from the parallel-plate capacitor model and a monotonic increase of the mobility. At Vbg < −1.8 V the electrons localize and the 2DEG is insulating. At high Vbg > 0.8 V a second 2DEG layer forms at a heterostructure interface between 2DEG and back-gate and we observe a decrease of the mobility. The measurements in the main text are obtained at Vbg = 0 V (n = 1.2 × 1011 cm−2) and at Vbg = −1.5 V (n = 0.3 × 1011 cm−2). T = 32 mKS5 Figure S5. Drude scattering length lD extracted from three measurements covering different temperature ranges, namely, at 30 mK, 30 mK - 4.2 K and 4.2 − 27 K together with the numerically calculated electron-electron interaction length [S17]. The presented SGM data in the main text is measured at electron densities of 1.2 × 1011 cm−2 and 0.3 × 1011 cm−2. For all densities, lD and lee are comparable at 1.5 K. At higher temperatures lD exceeds lee and viscous effects can arise. The black dots mark the parameters of the SGM images in Fig. 2 of the main text. DRUDE MEAN FREE PATH AND ELECTRON-ELECTRON SCATTERING LENGTH AS A FUNCTION OF TEMPERATURE AND ELECTRON DENSITY To observe viscous effects, the momentum relaxation length lD must exceed the electron-electron scattering length lee. This section presents the experimental values of lD and calculated values for lee. Figure S5 shows the measured Drude scattering length lD and the calculated electron-electron interaction length lee. lD is extracted from longitudinal and Hall resistance measurements, analogously to the electron density and the electron mobility in Fig. S4. We calculate the electron-electron scattering length lee = vFτee numerically according to the results of Jungwirth and MacDonald [S17], which contains corrections compared to the analytical expression by Giuliani and Quinn [S25]. The details are described in the following. We calculate the electron-electron scattering length from the imaginary part of the retarded quasiparticle self-energy Σ(k, ω) evaluated at the Fermi surface, i.e. ee = −2(cid:61)m[Σ(kF, 0)]/(¯hvF). Here kF = (cid:96)−1 2πn and vF = ¯hkF/mGaAs are, respectively, the Fermi momentum and velocity, whereas n is the electron density and mGaAs = 0.067me is the effective electron mass in the GaAs quantum well (me = 9.1× 10−31 Kg is the bare electron mass). The self-energy Σ(k, ω) is calculated within the G0W approximation, [S18] i.e. √ (cid:90) (cid:61)m[Σ(k, ω)] = d2 (2π)2(cid:61)m[W (q, ω − ξk−q)] [nB(ω − ξk−q) + nF(−ξk−q)] , (S2) where nF/B(ε) = (eβε ± 1)−1 stand for the equilibrium Fermi and Bose distributions, respectively, β = (kBT )−1 is the inverse temperature (kB is the Boltzmann constant), and ξk = ¯h2k2/(2mGaAs) − εF is the band energy from the Fermi energy εF = ¯h2k2 F/(2mGaAs). In Eq. (S2) W (q, ω) = V (q, ω)/
(q, ω) is the screened Coulomb interaction, while V (q, ω) = 2πe2/(
GaAsq) is the Fourier transform of the bare Coulomb interaction and
(q, ω) = 1− V (q, ω)χnn(q, ω). Here χnn(q, ω) is the density-density linear-response function, while
GaAs = 12 is the dielectric constant of GaAs. Note that (cid:61)m[Σ(k, ω)] as defined from Eq. (S2) only depends on the modulus of k. The dependence on the angle it forms with the x-axis can be removed by a change of variables in the integral on the right-hand side of Eq. (S2). Knowing both scattering lengths in the full parameter space spanned by n and T , we show a contour plot of lD and lee in Fig. S6(a). Even though transport regimes do not have abrupt limits, we indicate two areas where ballistic and viscous effects are expected according to the following rules: We mark the ballistic regime, where both lee and lD exceed 5 µm, by the green shade. Viscous effects are expected if lee is well below both lD and the sample size. Therefore we indicate the regime with lee < lD/10 and lee < 1 µm by the blue shade. The black dots and the dashed line mark the parameters of the measurements in the main text. Figure S6(b) reproduces Fig. 2(a) for convenient comparison. 01020T (K)51525leelDl, l (mm)Dee 010-11011011-2n = 1.5×10 cm11-2n = 1.2×10 cm11-2n = 0.9×10 cm11-2n = 0.6×10 cm11-2n = 0.3×10 cmT = 7.9 K(c)(c)(d)(d)(b)S6 Figure S6. (a) Absolute values of lee and lD as a function of T and n, from which we extract the ratio lD/lee. Thick lines with red numbers show calculated lee/1 µm. Thin lines with black numbers denote lD/1 µm. The ballistic regime with lee > 5 µm, lD > 5 µm is shaded green. The blue shade marks the regime of lee < lD/10 and lee < 1 µm. The labels (b), (c), (d) mark the parameters of the SGM measurements in Fig. 2 of the main text, and the dashed horizontal line marks the density of the measurement in Fig. 1(b). (b) From the absolute values in (a) we extract the ratio lD/lee as a function of T and n, which is illustrated by the red contours (repetition of Fig. 2(a) of the main text). Figure S7. Vicinity resistances as a function of temperature and charge carrier density. The parameters of the measurements in the main text are marked by the circles and the horizontal line. The inclined line marks the situation, when the separation between the current injecting orifice and the voltage probe equals Dν . These measurements are obtained in a second cool-down of the same sample and details might differ from the other presented data. At low density, the signals show noise of unknown origin, which was not present during the first cool-down. In a second cool-down of the same sample, we measured the vicinity resistance as a function of the full accessible parameter range of charge carrier density and temperature, but without scanning gate microscopy. The three measured vicinity resistances are shown in Fig. S7 as color plots with a black line highlighting the sign inversion. The circles and horizontal line marks the parameters of the measurements in Fig. 1(b) and Figs. 2(b)-(d) of the main text. The inclined dashed lines indicate the parameters where the characteristic length Dν of the hydrodynamic effect is equal to the distance between the current injecting opening and the corresponding voltage probe. These lines correspond to the grey dashed lines in Fig. S6 and in Fig. 2(a) of the main text. The results show an extended parameter range of negative vicinity resistances, around the temperature where Dν is similar as the separation of the voltage probe to the current injecting orifice. Towards high charge carrier density, the vicinity resistances become positive at a density coinciding with the formation of the second 2DEG as shown above in Fig. S4. HYDRODYNAMIC MODEL WITH LOCALLY REDUCED ELECTRON DENSITY As the SGM tip induces an approximately Lorentzian shaped potential in the 2DEG [S22], we approximate the electron density at a position x(cid:48), y(cid:48) in the channel by n(x(cid:48), y(cid:48)) = n0 − 1.2n0 l2 (x(cid:48) − x)2 + (y(cid:48) − y)2 + l2 (S3) ballistic(b)00.51.0103 3 1 0.30.151015T (K)11-2n (10 cm)(d)(c)(b)1(b)(a)00.51.051011-2n (10 cm)(c)(b)1(b)(d)l/lDee3010viscous5630T (K)10ballistic1200 nm600 nmD= 2400 nmn(a)11-2n (10 cm)1.6R600(W)0010T (K)20301.41.21.00.80.60.440-20-406020-60(b)11-2n (10 cm)(c)11-2n (10 cm)1.6R2400(W)0010T (K)20301.41.21.00.80.60.4155-520101.6R1200(W)0010T (K)20301.41.21.00.80.60.440-20-10302010(b)(c)(d)D= 2400 nmn1(b)(b)(c)(d)1(b)(b)(c)(d)1(b)D= 1200 nmnD= 600 nmnS7 Figure S8. Repetition of Fig. 5 in the main text with labels marking the 2DEG leads to the orifices in (a). with n0 the electron density in absence of the tip, a FWHM l = 300 nm, and a cut-off at zero (depleted 2DEG). The Comsol simulated charge distribution described in the next section supports this model of n(x(cid:48), y(cid:48)) in the vicinity of the tip. We simulate the SGM experiment by solving the hydrodynamic model from Bandurin et al. [S4] for every tip position on a line 0.5 µm from the upper channel boundary. These tip positions correspond to the dashed lines in Fig. 2(c) and (d) if we take a 2DEG depletion length of 150 nm around the top-gates into account. TRAJECTORY CALCULATIONS We model the potential in the 2DEG caused by SGM tip and QPC gates by calculating the charge distribution in Thomas-Fermi approximation with the finite element software COMSOL 5.0. The sample geometry includes the layer thickness of the Ga[Al]As heterostructure, the SGM tip size and the electron-beam lithography defined top-gates. The resulting electrostatic potential for one tip position is shown as a color plot in Fig. S8(a) as well as in Fig. 5(a) of the main text. Using this potential, we calculate the classical trajectories of electrons at the Fermi energy at an electron density n = 1.2 × 1011 cm−2. The trajectories start equidistantly and with a homogeneous angle distribution in the source lead, in Fig. S8(a) the starting line is indicated in green in the upper left corner. The red lines show post-selected trajectories that end in one of the three vicinity voltage probes V600, V1200, and V2400. So far, no random scattering is included in the calculation. We only consider electrons that have not scattered after leaving the source contact. We neglect the contributions of scattered electrons and their scattering partners for the sake of simplicity. The number of electrons that did not experience a scattering event decreases exponentially with trajectory length l. We therefore introduce a weight that exponentially decreases with l. As a qualitative measure Rbal of the ballistic contribution to the vicinity resistance Rj we count the weighted number of trajectories ending in the voltage probe Vj j (cid:88) Rbal j = e−lk/ls trajectory k ends in Vj with ls the typical length scale of scattering. At the parameters of Fig. 4(a) in the main text (T = 7.9 K, n = 1.2 × 1011 cm−2), electron-electron scattering is dominant (lee ≈ 370 nm (cid:28) lD ≈ 15 µm) so we use ls = 400 nm. Despite of its simplicity, this qualitative trajectory simulation illustrates the tip-position dependence of ballistic effects. We find a maximum of Rvic if the tip is in x-direction in the middle between the source orifice and the respective voltage probe, which agrees with the experimental results at high density described in the main text. This model neglects the contribution of the scattered electrons because it is beyond our capabilities to calculate. With the assumption, that the contribution of the scattered electrons is independent of tip position, we expect a vertical shift of the results in Fig. S8(b), which does not influence the x-position of the maxima. (a)(b)x (mm)0123450xsx600x1200x240011-2n=1.2310cmtippotentialchannel gatee.V(meV)01EFermi234V2400V1200V600bal-3R (10)j5V600V1200V2400s

1105.1390

1

1105

Giant microwave photoresistivity in a high-mobility quantum Hall system

We report the observation of a remarkably strong microwave photoresistivity effect in a high-mobility two-dimensional electron system subject to a weak magnetic field and low temperature. The effect manifests itself as a giant microwave-induced resistivity peak which, in contrast to microwave-induced resistance oscillations, appears only near the second harmonic of the cyclotron resonance and only at sufficiently high microwave frequencies. Appearing in the regime linear in microwave intensity, the peak can be more than an order of magnitude stronger than the microwave-induced resistance oscillations and cannot be explained by existing theories.

cond-mat.mes-hall

cond-mat

Giant microwave photoresistivity in high-mobility quantum Hall systems A. T. Hatke,1 M. A. Zudov,1, ∗ L. N. Pfeiffer,2 and K. W. West2 1School of Physics and Astronomy, University of Minnesota, Minneapolis, Minnesota 55455, USA 2Department of Electrical Engineering, Princeton University, Princeton, NJ 08544, USA (Received 28 October 2010; revised manuscript received 4 February 2011; published 7 March 2011) We report the observation of a remarkably strong microwave photoresistivity effect in a high-mobility two- dimensional electron system subject to a weak magnetic field and low temperature. The effect manifests itself as a giant microwave-induced resistivity peak which, in contrast to microwave-induced resistance oscillations, appears only near the second harmonic of the cyclotron resonance and only at sufficiently high microwave frequencies. Appearing in the regime linear in microwave intensity, the peak can be more than an order of magnitude stronger than the microwave-induced resistance oscillations and cannot be explained by existing theories. PACS numbers: 73.43.Qt, 73.21.-b, 73.40.-c, 73.63.Hs Transport properties of high-mobility two-dimensional electron systems (2DESs) subject to a weak magnetic field B and low temperature T can be modified dramatically by microwave radiation,1 thermally excited acoustic phonons,2 dc electric fields,3 or their combinations.4 In either case, the 2DES reveals a specific class of 1/B-periodic resistance os- cillations, which persist down to magnetic fields much lower than the onset of the Shubnikov-de Haas oscillations (SdHOs). In irradiated 2DESs, such oscillations, usually called microwave-induced resistance oscillations (MIROs), are con- trolled by a dimensionless parameter, ǫac = ω/ωc, where ω = 2π f is the microwave frequency and ωc = eB/m∗ is the cyclotron frequency of an electron with an effective mass m∗. The resistivity can be expressed as ρω = ρ + δρω, where ρ is the resistivity of nonirradiated 2DES and δρω(ǫac) is a sign-alternating photoresistivity. According to the "dis- placement" model, δρω originates from the radiation-induced impurity-assisted transitions between the Landau levels.5 In another mechanism, known as "inelastic", microwaves create a nonequilibrium distribution of electron states which, in turn, translates to the oscillatory δρω.6 In the regime of overlapping Landau levels both models give δρω ρ ∝ −Pωλ2ǫac sin(2πǫac). (1) Here, Pω ∝ ω−4 is the dimensionless parameter proportional to the microwave power, λ = exp(−π/ωcτq) is the Dingle fac- tor, and τq is the quantum lifetime. Even though Eq. (1) pre- dicts the MIRO maxima at ǫn+ ac ≃ n − 1/4 (n = 1, 2, 3, . . . ), experimentally the lower order peaks are often found at ǫn+ ac ≃ n − ϕ, with 0 < ϕ < 1/4. In a very clean 2DES, MIRO minima can evolve into zero- resistance states which are believed to originate from the ab- solute negative resistance and its instability with respect to formation of current domains.7 As a result, negative photore- sistivity never exceeds the dark resistivity by absolute value. In contrast, positive photoresistivity has no underlying insta- bilities and was routinely found to exceed the dark resistivity. In addition to MIROs, a remarkably strong and narrow pho- toresistivity peak was recently observed in close proximity to the cyclotron resonance.8 This peak showed thresholdlike de- pendence on microwave power and was explained by the bolo- metric effect due to resonant heating of electrons. In this paper, we report on another unusually strong mi- crowave photoconductivity effect in a high-mobility 2DES. This effect manifests itself as a giant photoresistance peak, which emerges only near the second harmonic of the cyclotron resonance. Similar to MIROs, the peak exhibits linear de- pendence on microwave intensity and quickly disappears with increasing temperature. However, in contrast to MIROs, the amplitude of which quickly decays with increasing frequency (as 1/ω4), the giant peak is observed only when the frequency is sufficiently high. While the peak roughly coincides with the second MIRO maximum, it can be more than an order of magnitude stronger than MIROs. Understanding the nature of such a dramatic effect remains a subject of future studies. While the effect was observed in several 2DESs, the data presented here were collected using a Hall bar (width w = 100 µm) etched from a symmetrically doped GaAs/AlGaAs quan- tum well. After a brief low-temperature illumination, the den- sity and the mobility at T = 0.5 K were ne ≃ 3.3 × 1011 cm−2 and µ ≃ 1.1 × 107 cm2/Vs, respectively. Microwave radia- tion was generated by Gunn and backward wave oscillators. Measurements were done using a quasi-dc lock-in technique at bath temperatures T from 0.5 K to 4.0 K. In Fig.1 (a) and 1 (b) we show the magnetoresistivity ρω(B) acquired at T ≃ 0.5 K under microwave irradiation (dark curves) of frequency f = 95 and 190 GHz. For comparison, each panel also includes the magnetoresistivity ρ(B) obtained without microwave irradiation (light curves). Dark resistivity ρ(B) shows a strong negative magnetoresistance effect at low magnetic fields; at B ≃ 1 kG the dark resistivity is reduced by nearly two orders of magnitude compared to its value at B = 0. At higher B, magnetoresistance becomes positive and SdHOs appear. Under microwave irradiation, the data show both MIROs and zero-resistance states. However, our main focus is the so-called X2 peak (cf.↓) near ǫac = 2, which, at least in the case of f = 190 GHz, is distinct from MIROs. We first notice that direct examination of Fig. 1 reveals that the X2 peak is considerably stronger than other oscillations. Second, the appearance of the peak is uniquely tied to the sec- ond harmonic of the cyclotron resonance as neither the third nor fourth harmonic shows similar features. Finally, com- ] (cid:58) (cid:62) (cid:3) (cid:90) (cid:85) (cid:3) (cid:15) (cid:85) ] (cid:58) (cid:62) (cid:3) (cid:90) (cid:85) (cid:3) (cid:15) (cid:85) a b 5 4 3 2 1 0 3 2 1 0 0 2 a b 95 GHz 190 GHz f = 190.0 GHz 152.5 GHz 130.0 GHz 1 10 0 10 ] (cid:58) [ c a (cid:72) / 10 (cid:3) (cid:90) (cid:85) (cid:71) 10 -1 -2 -3 10 + 3 (cid:90) (cid:85) (cid:71) (cid:18) (cid:90) (cid:85) (cid:71) 20 10 0 (cid:70) (cid:70) (cid:70) (cid:70) 95 GHz 2 1 MIRO ZRS (cid:70) 190 GHz 4 3 2 MIRO 1 ZRS 2 B [kG] 3 0 1 2 3 5 6 7 8 4 (cid:72) ac FIG. 1: (color online) (a), (b) Magnetoresistivities measured with (ρω(B), dark curves) and without (ρ(B), light curves) microwave ir- radiation of f = 95 and 190 GHz at T ≃ 0.5 K. Vertical lines are marked by integer ǫac. paring the 190 GHz to the 95 GHz data we find that MIROs are suppressed considerably, mostly due to ω−4 decay of Pω [cf. Eq. (1)]. On the other hand, the X2 peak becomes even more pronounced, suggesting a frequency dependence, which is clearly inconsistent with ω−4. This characteristic frequency dependence might explain why the X2 peak was not detected in earlier studies employing lower frequencies. Other neces- sary conditions for the observation of this unusual peak are sufficiently high mobility and low temperature. To quantitatively compare the X2 peak to MIROs, we present the normalized oscillation amplitude, δρω/ǫac, as a function of ǫac in Fig. 2 (a). For ǫac ≥ 3, both data sets exhibit anticipated exponential decay, δρω/ǫac ∝ λ2 = exp(−ǫac/ f τq), as illustrated by solid lines drawn with τq = 9.1 ps. It is clear, however, that the magnitude at the X2 peak significantly ex- ceeds these dependences; direct comparison shows that the photoresistance at the X2 peak is enhanced by roughly a fac- tor of 3 (18) for f = 95 (190) GHz. Here, we should notice that, while for ǫac & 3, Pω can be treated as ǫac independent, it is expected to be enhanced considerably near the cyclotron resonance.9 This enhancement can, in principle, increase the response near ǫac = 2 by a factor of about 2. If this correction is taken into account, the amplitude of the X2 peak measured at f = 95 GHz is in closer agreement with the MIRO ampli- tude but it is clearly not enough to explain the peak value at f = 190 GHz. We also notice that the X2 peak at f = 190 GHz is significantly sharper compared to a peak at f = 95 GHz which has a shape similar to conventional MIROs. We therefore concentrate on higher frequency data. We now switch to the evolution of the X2 peak with mi- crowave frequency. In Fig. 2 (b) we present the photoresistiv- FIG. 2: (color online) (a) Normalized oscillation amplitude δρω/ǫac versus ǫac at f = 95 and 190 GHz. Solid lines show exponential decay δρω/ǫac ∝ exp(−ǫac/ f τq) with τq = 9.1 ps. Extrapolation of these lines to ǫac = 0 gives the ratio of microwave intensities, Pω(95 GHz)/Pω(190 GHz) ≃ 30. (b) Photoresistivity normalized to the ǫ3+ ω versus ǫac for f = 130, 152.5, and 190 GHz (bottom to top) at T ≃ 0.5 K. The traces are vertically offset for clarity by 3. ac peak δρω/δρ3+ ity for three frequencies (as marked), normalized to the ǫ3+ ac MIRO peak, δρω/δρ3+ ω versus ǫac. Such normalization helps to account for the variation of the microwave intensity seen by our 2DES. We observe that the X2 peak (cf. ↓) grows with increasing ω confirming that its frequency dependence is in- consistent with that of MIROs. Furthermore, as prescribed by Eq. (1), all the peaks, including the X2 peak, are positioned near ǫn+ ac = n − φ, with φ > 0. This result is in contrast to Ref. 10, which, based on absorption measurements, concludes that the X2 peak occurs at ǫac = 2. We next examine the power dependence of MIROs and the X2 peak measured at f = 190 GHz. In Fig. 3 (a) we present magnetoresistivity ρω(B) acquired at selected attenuations, as marked, from 0 dB (top trace) to -11 dB (bottom trace). The attenuation steps were selected to roughly mimic a constant step in microwave intensity. The traces are vertically offset by 1 Ω for clarity. Figure 3 (a) clearly shows that, with decreasing radiation intensity, MIROs gradually diminish and so does the X2 peak. We also note that, once the zero-resistance state disappears, the data reveal another rather weak but sharp peak just below the cyclotron resonance (cf.↑). To understand the observed evolution with microwave in- tensity, we present in Fig. 3 (b) the resistance values at the peaks near ǫac = 2 (circles) and ǫac = 3 (squares, values multiplied by five) as a function of microwave intensity P (in units of intensity before attenuation). At not too high P, both X2 and ǫ3+ ac peaks show linear dependence on P. How- a 8 6 ] (cid:58) [ (cid:3) (cid:90) (cid:85) 4 2 0 0 23 1 b 8 a 23 1 3 2 1 ] (cid:58) [ (cid:3) (cid:90) (cid:85) 0 dB -1 dB -2.2 dB -4.2 dB -7 dB -11 dB 1 2 3 B [kG] (cid:67) 4 0 0.0 ] (cid:58) [ (cid:3) (cid:90) (cid:85) 6 4 2 1.5 K (cid:70) (cid:67) n = 2 n = 3 (5x) 1.0 0 0 0.5 P 4.0 K 1 2 3 4 B [kG] 3 n = 2 b 6 4 2 1 8 6 ] (cid:58) [ (cid:3) (cid:90) (cid:85) (cid:71) 4 2 0.1 8 6 4 2 0 n = 3 4 8 [K 2 ] 2 T 12 FIG. 3: (color online) (a) Magnetoresistivity ρω(B) measured at f = 190 GHz at different attenuations, from 0 dB to -11 dB. The traces are marked by attenuation factors and are vertically offset by 1 Ω. (b) Resistivity at the X2 peak (circles) and ǫ3+ ac peak (squares, values multiplied by five) versus P. ever, at highest P, both dependencies show signs of saturation. Such saturation may originate from the nonresonant heating of the 2DES by microwaves, which is manifested in the pro- gressively stronger damping of the SdHOs with increasing P. Another possible origin for the sublinear dependence is the intrinsic nonlinearity of the photoresponse due to multiphoton processes.9 We now turn to our results of a temperature-dependence study. In Fig. 4 (a) we show the magnetoresistivity ρω(B) mea- sured at f = 190 GHz and at different temperatures from T = 1.5 K (top) to T = 4.0 K (bottom), in steps of 0.5 K. For clarity, the traces are vertically offset by 1 Ω. With in- creasing T , both MIROs and the X2 peak gradually weaken and eventually decay away. Another interesting observation is a sharp photoresistivity minimum (cf.↓), which emerges at the lower B edge of the zero-resistance state developed between 3 and 4 kG at T = 1.5 K. Finally, similar to the low P data in Fig. 3 (a) one observes a sharp peak at the higher B edge of the zero-resistance state near ǫac ≃ 1 (cf.↑). In Fig. 4 (b) we present the photoresistivity δρω at the X2 ac peak (squares) versus T 2 and ob- peak (circles) and ǫ3+ serve that both data sets are well described by δρω(T ) ∝ exp(−T 2/T 2 0 ), with T0 ≃ 2.2 K (cf. lines). We note that similar behavior, recently observed for all classes of in- duced resistance oscillations, was attributed to electron- electron interactions.11 In the regime of separated Landau levels theory6 predicts, up to a factor of the order of unity, kBT0 ≃ √ΓεF /2π, where kB is the Boltzmann constant, εF is the Fermi energy, and 2Γ is the Landau level width. Using τq ≃ 10−11 s and 2Γ = /τq yields T0 ≃ 3 K, in agreement with experiment. We thus conclude that the T -dependence of the X2 peak could also be explained by electron-electron in- FIG. 4: (color online) (a) Magnetoresistivity ρω(B) measured at f = 190 GHz at P = 1 and different temperatures, from 1.5 K (top) to 4.0 K (bottom), in step of 0.5 K. The traces are vertically offset by 1 Ω. (b) Photoresistivity δρω at the X2 peak (circles) and ǫ3+ ac peak (squares) versus T 2. teractions. In summary, we reported on a novel microwave-induced resistivity peak emerging in a high-mobility 2DES at low temperatures. Similar to MIROs, this peak grows linearly with power and decays exponentially with temperature, but is clearly of a different origin. First, it appears only near the second harmonic of the cyclotron resonance and is not ob- served at other harmonics, regardless of the magnetic field. Second, the peak appears only at sufficiently high microwave frequencies and its frequency dependence differs dramatically from MIROs, the amplitude of which decays as ω−4. Fi- nally, it can be more than an order of magnitude stronger than the microwave-induced resistance oscillations and more than two orders of magnitude larger than the dark resistivity. This phenomenon cannot be explained by any of the existing theories and prompts for further developments in the field of nonequlibrium transport of quantum Hall systems. We thank I. Dmitriev, R. Du, M. Dyakonov, M. Khodas, and B. Shklovskii for discussions and S. Hannas, G. Jones, J. Krzystek, T. Murphy, E. Palm, J. Park, D. Smirnov, and A. Ozarowski for technical assistance. A portion of this work was performed at the National High Magnetic Field Labo- ratory, which is supported by NSF Cooperative Agreement No. DMR-0654118, by the State of Florida, and by the DOE. The work at Minnesota was supported by the DOE Grant No. de-sc0002567 (measurements at NHMFL) and by the NSF Grant No DMR-0548014 (low frequency measurements at Minnesota). The work at Princeton was partially funded by the Gordon and Betty Moore Foundation as well as the NSF MRSEC Program through the Princeton Center for Complex Materials (DMR-0819860). A.T.H. acknowledges support by the University of Minnesota. ∗ Corresponding author: zudov@physics.umn.edu 1 M. A. Zudov, R. R. Du, J. A. Simmons, and J. L. Reno, Phys. Rev. B 64, 201311(R) (2001); M. A. Zudov, ibid. 69, 041304(R) (2004); R. G. Mani, V. Narayanamurti, K. von Klitzing, J. H. Smet, W. B. Johnson, and V. Umansky, ibid. 69 161306 (2004); ibid. 70, 155310 (2004); S. I. Dorozhkin, J. H. Smet, V. Uman- sky, and K. von Klitzing, ibid. 71, 201306(R) (2005); S. A. Stu- denikin, M. Potemski, A. Sachrajda, M. Hilke, L. N. Pfeiffer, and K. W. West, ibid. 71, 245313 (2005); S. A. Studenikin et al., ibid. 76, 165321 (2007); M. A. Zudov, R. R. Du, L. N. Pfeiffer, and K. W. West, ibid. 73, 041303(R) (2006); Z. Q. Yuan, C. L. Yang, R. R. Du, L. N. Pfeiffer, and K. W. West, ibid. 74, 075313 (2006); A. Wirthmann, B. D. McCombe, D. Heitmann, S. Holland, K.-J., Friedland, and C.-M. Hu, ibid. 76, 195315 (2007); S. Wiedmann, G. M. Gusev, O. E. Raichev, A. K. Bakarov, and J. C. Portal, ibid. 78, 121301(R) (2008); ibid. 80, 035317 (2009); O. M. Fedorych, M. Potemski, S. A. Studenikin, J. A. Gupta, Z. R. Wasilewski, and I. A. Dmitriev, ibid. 81, 201302 (2010); P. D. Ye, L. W. Engel, D. C. Tsui, J. A. Simmons, J. R. Wendt, G. A. Vawter, and J. L. Reno, Appl. Phys. Lett. 79, 2193 (2001); S. I. Dorozhkin, JETP Lett. 77, 577 (2003); S. I. Dorozhkin, A. A. Bykov, I. V. Pechenezhskii, and A. K. Bakarov, ibid. 85, 576 (2007); S. I. Dorozhkin, J. H. Smet, K. von Klitzing, L. N. Pfeiffer, and K. W. West, ibid. 86, 543 (2007); A. A. Bykov, A. K. Bakarov, A. K. Kalagin, and A. I. Toropov, ibid. 81, 284 (2005); A. A. Bykov, D. R. Islamov, D. V. Nomokonov, and A. K. Bakarov, ibid. 86, 779 (2007); A. A. Bykov ibid. 87, 233 (2008); 87, 551 (2008); 89, 575 (2009); 91, 361 (2010); A. A. Bykov and I. V. Marchishin, ibid. 92, 71 (2010); I. V. Andreev, V. M. Murav'ev, I. V. Kukushkin, J. H. Smet, K. von Klitzing, and V. Umanskii, ibid. 88, 616 (2009); L.-C. Tung, C. L. Yang, D. Smirnov, L. N. Pfeiffer, K. W. West, R. R. Du, and Y.-J. Wang, Solid State Commun. 149, 1531 (2009). 2 M. A. Zudov, I. V. Ponomarev, A. L. Efros, R. R. Du, J. A. Sim- mons, and J. L. Reno, Phys. Rev. Lett. 86, 3614 (2001); I. V. Pono- marev and A. L. Efros, Phys. Rev. B 63, 165305 (2001); X. L. Lei, ibid. 77, 205309 (2008); O. E. Raichev, ibid. 80, 075318 (2009); M. A. Zudov, J. Zhang, R. R. Du, J. A. Simmons, and J. L. Reno, Physica E (Amsterdam) 12, 443 (2002); A. A. Bykov, A. K. Kala- gin, and A. K. Bakarov, JETP Lett. 81, 523 (2005). 3 C. L. Yang, J. Zhang, R. R. Du, J. A. Simmons, and J. L. Reno, Phys. Rev. Lett. 89, 076801 (2002); A. A. Bykov, J. Zhang, S. Vitkalov, A. K. Kalagin, and A. K. Bakarov, Phys. Rev. B 72, 245307 (2005); W. Zhang, H.-S. Chiang, M. A. Zudov, L. N. Pfeiffer, and K. W. West, ibid. 75, 041304(R) (2007); M. G. Vav- ilov, I. L. Aleiner, and L. I. Glazman, Phys. Rev. B 76, 115331 (2007); A. Auerbach and G. V. Pai, ibid. 76, 205318 (2007); A. T. Hatke, H.-S. Chiang, M. A. Zudov, L. N. Pfeiffer, and K. W. West, ibid. 82, 041304(R) (2010); A. T. Hatke, M. A. Zudov, L. N. Pfeiffer, and K. W. West, ibid. 83, 081301(R) (2011); X. L. Lei, Appl. Phys. Lett. 90, 132119 (2007); Physica E (Amsterdam) 42 63 (2009). 4 W. Zhang, M. A. Zudov, L. N. Pfeiffer, and K. W. West, Phys- 4 ica E (Amsterdam) 40, 982 (2008); Phys. Rev. Lett. 98, 106804 (2007); 100, 036805 (2008); A. T. Hatke, H.-S. Chiang, M. A. Zudov, L. N. Pfeiffer, and K. W. West, ibid. 101, 246811 (2008); Phys. Rev. B 77, 201304(R) (2008); M. Khodas and M. G. Vav- ilov, ibid. 78 245319 (2008); I. A. Dmitriev, R. Gellmann, and M. G. Vavilov, ibid. 82, 201311 (2010); X. L. Lei, ibid. 79, 115308 (2009); Appl. Phys. Lett. 91 112104 (2007); X. L. Lei and S. Y. Liu, ibid. 93, 082101 (2008); M. A. Zudov, H.-S. Chiang, A. T. Hatke, W. Zhang, L. N. Pfeiffer, and K. W. West, Int. J. of Mod. Phys. B 23, 2684 (2009). 5 V. I. Ryzhii, Sov. Phys. Solid State 11, 2078 (1970); V. I. Ryzhii, R. A. Suris, and B. S. Shchamkhalova, Sov. Phys. Semicond. 20, 1299 (1986); A. C. Durst, S. Sachdev, N. Read, and S. M. Girvin, Phys. Rev. Lett. 91, 086803 (2003); J. Shi and X. C. Xie, ibid. 91, 086801 (2003); X. L. Lei and S. Y. Liu, ibid. 91, 226805 (2003); M. G. Vavilov and I. L. Aleiner, Phys. Rev. B 69, 035303 (2004); I. A. Dmitriev, M. Khodas, A. D. Mirlin, D. G. Polyakov, and M. G. Vavilov, ibid. 80, 165327 (2009). 6 I. A. Dmitriev, A. D. Mirlin, and D. G. Polyakov, Phys. Rev. Lett. 91, 226802 (2003); 99, 206805 (2007); Phys. Rev. B 75, 245320 (2007); I. A. Dmitriev, M. G. Vavilov, I. L. Aleiner, A. D. Mirlin, and D. G. Polyakov, ibid. 71, 115316 (2005). 7 R. G. Mani, J. H. Smet, K. von Klitzing, V. Narayanamurti, W. B. Johnson, and V. Umansky, Nature (London) 420, 646 (2002); M. A. Zudov, R. R. Du, L. N. Pfeiffer, and K. W. West, Phys. Rev. Lett. 90, 046807 (2003); 96, 236804 (2006); C. L. Yang, M. A. Zudov, T. A. Knuuttila, R. R. Du, L. N. Pfeiffer, and K. W. West, ibid. 91, 096803 (2003); A. V. Andreev, I. L. Aleiner, and A. J. Millis, ibid. 91, 056803 (2003); R. G. Mani, J. H. Smet, K. von Klitzing, V. Narayanamurti, W. B. Johnson, and V. Umansky, ibid. 92 146801 (2004); R. L. Willett, L. N. Pfeiffer, and K. W. West, ibid. 93, 026804 (2004); A. Auerbach, I. Finkler, B. I. Halperin, and A. Yacoby, ibid. 94, 196801 (2005); P. W. Anderson and W. F. Brinkman, arXiv:cond-mat/0302129; R. R. Du, M. A. Zudov, C. L. Yang, L. N. Pfeiffer, and K. W. West, Physica E (Amsterdam) 22, 7 (2004); Int. J. of Mod. Phys. B 18 3465 (2004); C. L. Yang, R. R. Du, L. N. Pfeiffer, and K. W. West, ibid. 74, 045315 (2006); I. G. Finkler and B. I. Halperin, ibid. 79, 085315 (2009). 8 J. H. Smet, B. Gorshunov, C. Jiang, L. Pfeiffer, K. West, V. Uman- sky, M. Dressel, R. Meisels, F. Kuchar, and K. von Klitzing, Phys. Rev. Lett. 95, 116804 (2005). 9 M. Khodas, H.-S. Chiang, A. T. Hatke, M. A. Zudov, M. G. Vav- ilov, L. N. Pfeiffer, and K. W. West, Phys. Rev. Lett. 104, 206801 (2010). 10 Y. Dai, R. R. Du, L. N. Pfeiffer, and K. W. West, Phys. Rev. Lett. 105, 246802 (2010). 11 A. T. Hatke, M. A. Zudov, L. N. Pfeiffer, and K. W. West, Phys. Rev. Lett. 102, 066804 (2009); 102, 086808 (2009); Phys. Rev. B 79, 161308(R) (2009); Physica E (Amsterdam) 42, 1081 (2010); A. A. Bykov and A. V. Goran, JETP Lett. 90, 578 (2009). 

1501.02551

1

1501

Crossover between Silicene and Ultra-Thin Si Atomic Layers on Ag(111) Surfaces

We report on total-energy electronic structure calculations in the density-functional theory performed for the ultra-thin atomic layers of Si on Ag(111) surfaces. We find several distinct stable silicene structures: $\sqrt{3}\times\sqrt{3}$, $3\times3$, $\sqrt{7}\times\sqrt{7}$ with the thickness of Si increasing from monolayer to quad-layer. The structural bistability and tristability of the multilayer silicene structures on Ag surfaces are obtained, where the calculated transition barriers infer the occurrence of the flip-flop motion at low temperature. The calculated STM images agree well with the experimental observations. We also find the stable existence of $2\times1$ $\pi$-bonded chain and $7\times7$ dimer-adatom-stacking fault Si(111)-surface structures on Ag(111), which clearly shows the crossover of silicene-silicon structures for the multilayer Si on Ag surfaces. We further find the absence of the Dirac states for multilayer silicene on Ag(111) due to the covalent interactions of silicene-Ag interface and Si-Si interlayer. Instead, we find a new state near Fermi level composed of $\pi$ orbitals locating on the surface layer of $\sqrt{3}\times\sqrt{3}$ multilayer silicene, which satisfies the hexagonal symmetry and exhibits the linear energy dispersion. By examining the electronic properties of $2\times1$ $\pi$-bonded chain structures, we find that the surface-related $\pi$ states of multilayer Si structures are robust on Ag surfaces.

cond-mat.mes-hall

cond-mat

5 Crossover between Silicene and Ultra-Thin Si 1 0 2 Atomic Layers on Ag(111) Surfaces Zhi-Xin Guo ‡ and Atsushi Oshiyama Department of Applied Physics, The University of Tokyo, Tokyo 113-8656, Japan Abstract. We report on total-energy electronic structure calculations in the density- functional theory performed for the ultra-thin atomic layers of Si on Ag(111) surfaces. We find several distinct stable silicene structures: √3 × √3, 3 × 3, √7 × √7 with the thickness of Si increasing from monolayer to quad-layer. The structural bistability and tristability of the multilayer silicene structures on Ag surfaces are obtained, where the calculated transition barriers infer the occurrence of the flip-flop motion at low temperature. The calculated STM images agree well with the experimental observations. We also find the stable existence of 2×1 π-bonded chain and 7×7 dimer- adatom-stacking fault Si(111)-surface structures on Ag(111), which clearly shows the crossover of silicene-silicon structures for the multilayer Si on Ag surfaces. We further find the absence of the Dirac states for multilayer silicene on Ag(111) due to the covalent interactions of silicene-Ag interface and Si-Si interlayer. Instead, we find a new state near Fermi level composed of π orbitals locating on the surface layer of √3 × √3 multilayer silicene, which satisfies the hexagonal symmetry and exhibits the linear energy dispersion. By examining the electronic properties of 2 × 1 π-bonded chain structures, we find that the surface-related π states of multilayer Si structures are robust on Ag surfaces. n a J 2 1 ] l l a h - s e m . t a m - d n o c [ 1 v 1 5 5 2 0 . 1 0 5 1 : v i X r a ‡ Present address: Xiangtan University, Xiangtan, Hunan 411105, China Crossover between Silicene and Ultra-Thin Si Atomic Layers on Ag(111) Surfaces 2 1. Introduction A honeycomb-structured two-dimensional (2D) atomic layer consisting of group IV atoms exhibits peculiar electronic properties due to a fact that electrons near the Fermi level (EF) follow effectively the massless Dirac equation (Weyl equation) [1]. A well- known and only unequivocally measured example is graphene where intriguing properties such as an anomalous quantum Hall effect is observed [2, 3]. Another element in group IV, Si, which has sustained our modern life, should exhibit such fascinating properties [4, 5] thus opening a new door to the next-generation technology with its pronounced relativistic effects related to the spin degrees of freedom [6, 7, 8]. The 2D structure of Si, known as silicene, is first synthesized on Ag(111) surfaces. The monolayer silicene on Ag(111) exhibits variety of structures. The superperiodicities of 4 × 4 [9, 10, 11], √13 × √13 [10, 11], and 2√3 × 2√3 [11, 12] with respect to 1 × 1 Ag(111) surface have been observed, and the simulated scanning tunneling microscopy (STM) images of theoretically determined structures reproduce the observed images √3 silicene on Ag(111) with respect to 1 × 1 excellently [13, 14]. Recently, the √3 × silicene has been reported by Chen et al.[15, 16]. They also claimed the flip-flop motion between the two stable hexagonal √3 structures to explain their honeycomb symmetry STM images at higher temperature and its freeze at lower temperature [16]. However, our density-functional theory (DFT) calculations have clarified that Chen's observation should correspond to the bilayer silicene [17]. On the other hand, the multilayer (including bilayer) silicene structures have been also synthesized on Ag(111) surfaces in other experiments [18, 19, 20, 21, 22], which interestingly exhibit the similar honeycomb √3 × √3 STM images as that reported by Chen et al., irrespective of the thickness of silicene. Thus it is very important to clarify the structural details for the multilayer silicene on Ag(111). On the other hand, Si surfaces are premier stages on which the fabrication of most electron devices is achieved. Structural characteristics of such Si surfaces are well identified experimentally and theoretically [23]. In particular, the atomic layers of the (111) surface has the hexagonal network in the lateral plane, and thus the similarity and dissimilarity between the silicene and the surface atomic layers of Si(111) surface are intriguing. Actually the cleaved Si(111) surface shows 2 × 1 superperiodicity and then its annealing converts it to the 7× 7 periodicity. The 2× 1 structure has been identified as the π-bonded chain structure [24] in which the top two Si layers are drastically reconstructed to form chains of π orbitals associated with the five- and seven-membered rings of the first and the second surface layers. The 7 × 7 structure is more complicated and a dimer-adatom-stacking-fault (DAS) model [25, 26] is now established [27, 28]. It is important and interesting to clarify how a crossover between the silicene structure and the reconstructed surface structure takes place in the multilayer silicene. the situation is controversial for the monolayer silicene on Ag(111): Angular-resolved photoelectron spectroscopy (ARPES) measurements [9, 29] show the existence of the electron state with the linear energy As for the electron states near EF, Crossover between Silicene and Ultra-Thin Si Atomic Layers on Ag(111) Surfaces 3 dispersion (Dirac state) for the 3 × 3 silicene on 4 × 4 Ag(111), whereas the DFT calculations clarify the absence of Dirac electrons due to the strong silicene-substrate interactions [13, 14, 30, 31, 32]; it is of note that no Landau-level sequences peculiar to Dirac electrons have been observed [33]. From the interference patterns obtained by scanning tunneling spectroscopy (STS) measurements, Chen et al. claimed the presence √3 silicene on Ag(111), whereas Arafune et al. deduced the contrary conclusion, the absence of Dirac electrons, from essentially identical STS experiments [34]. Our DFT calculations show that the of Dirac states with the linear dispersion [15, 16] for the √3 × √3 × √3 silicene structure observed by Chen et al. actually correspond to the bilayer silicene and the linear energy dispersion is attributed to the sp band of Ag [17]. Recently, √3 the ARPES measurements also observed the linear energy dispersions for the √3 × multilayer silicene on Ag(111), where the electron velocity is much smaller than that observed by Chen et al. The linear energy bands are recognized to be the Dirac π and anti-bonding π (labeled as π∗) bands of multilayer silicene [19, 35]. Details of the electronic structure of the multilayer silicene are still mysterious and further theoretical work is in great need for the clarification. In this work, we report on total-energy electronic structure calculations for the ultra-thin atomic layers of Si on Ag(111) surfaces. We find several distinct stable silicene structures: √3×√3, 3× 3, √7×√7 with the thickness of Si increasing from monolayer to quad-layer. We also find the existence of 2 × 1 and 7 × 7 Si(111)-surface structures on Ag(111). The crossover of silicene-silicon structures on Ag surface appears when the thickness of Si layer becomes larger than 2. We also find the structural bistability and tristability of the multilayer silicene structures, where the calculated transition barriers infer the existence of flip-flop motion at low temperature. We further find the absence of the Dirac states for multilayer silicene on Ag(111) due to the covalent interactions of silicene-Ag interface and Si-Si interlayer. Interestingly, we find a new state near EF composed of π orbitals locating the top layer of √3 × √3 multilayer silicene, which satisfies the hexagonal symmetry and exhibits the linear energy dispersion. Finally, we explore the electronic properties of Si(111)-surface structures on Ag surfaces. 2. Calculations The total-energy electronic-structure calculations have been performed in the density functional theory using VASP code [36, 37]. As for the exchange-correlation functional, the vdW-DF [38, 39], being capable of treating the dispersion force is adopted [40]. Calculations for monolayer silicene and some of bilayer silicene have been also performed by the local density approximation (LDA) and the generalized gradient approximation (GGA) [13, 14]. The results obtained are essentially same among the three exchange- correlation functionals, although quantitative values for the total energies are different. The electron-ion interaction is described by the projector augmented wave method [41], and the cutoff energy of 250 eV in the plane-wave basis set is used. The Ag surface is simulated by a repeating slab model in which a five-atomic-layer slab is cleaved from Crossover between Silicene and Ultra-Thin Si Atomic Layers on Ag(111) Surfaces 4 the face-centered-cubic (fcc) Ag with the experimental lattice constant (4.09 A). The Ag slabs are separated from their images by the vacuum regions with the thickness of 14 A, 17 A, 21 A, and 24 A for the monolayer, bilayer, tri-layer and quad-layer Si deposited on Ag(111), respectively. The geometry optimization is performed until the remaining forces become less than 0.02 eV/A. We have carefully examined the validity of the k-point mesh in Brillouin zone (BZ) integration, and found that the spacing between adjacent k-point being less than 0.016 A−1 is sufficient to obtain converged results. It has been recognized that the 3 × 3 and √7 × √7 silicene are commensurate to the 4× 4 and √13×√13 Ag(111), respectively [13]. To explore the π-bonded chain and DAS structures on Ag(111) surface, we prepare initial structures by depositing the Si layers that are cleaved from the Si(111) with the periodicity of the 2 × 2 (two times of √7 and 2√21 × 2√21 Ag(111), respectively, the 2 × 1 unit cell) and 7 × 7 on the √7 × where the lattice mismatches are 0.4 % and 1.4 %. Following the convention of the silicene layer, we define a single layer, i.e., a single bilayer, as consisting of 4 Si atoms at the upper positions and the remaining 4 Si atoms at the lower positions in the 2 × 2 lateral cell of the Si(111) surface. To assess relative stability among various stable and metastable structures, we introduce the cohesive energy Ec and the binding energy Eb. They are defined as Ec = (EAg(111) + NSiµSi − Etot)/NSi and Eb = (EAg(111) + Esi−layers − Etot)/NLsi, respectively. Here Etot, EAg(111) and Esi−layers are the total energies of the Si layers on Ag(111), the clean Ag(111) surface and the freestanding Si layers, respectively. NSi and NLsi are the total number of Si atoms and the number of Si atoms per Si layer, respectively. µSi is the chemical potential of Si which is adopted as the total energy of an isolated Si atom in this paper. The cohesive energy defined above is the energy gain to make Si ultra-thin layers on the Ag(111) surface from the clean Ag surface plus the constituent Si atoms. This may be a measure of the catalytic ability of the Ag surface in forming the Si ultra-thin layers. The binding energy is the measure of the energy gain to put the freestanding Si structures on the Ag(111) surface. In any case, Ec and Eb represent relative stability of various Si structures on the Ag(111) surface. 3. Stabilities and structures of Si layers on Ag(111) In this section we present structural characteristics and energetics of ultra-thin Si atomic layers on Ag(111) surfaces. We examine Si monolayer (ML), bilayer (BL), tri-layer (TL) and quad-layer (QL) structures. An interesting finding is the wealth of the structural multi-stability being composed of a variety of distinct structures with the cohesive- energy difference in the range of less than 100 meV per Si atom (meV/Si). The π-bonded chain and DAS structures observed on the Si(111) surface are found to exist also in the ultra-thin Si layers on the Ag(111) surface. Crossover between Silicene and Ultra-Thin Si Atomic Layers on Ag(111) Surfaces 5 (a) (b) Figure 1. (color online) Top (upper panels) and side (under panels) views of geometry optimized structures for the 3 × 3 monolayer (ML) silicene on 4 × 4 Ag(111). After optimization, the silicene shows the √3 × √3 (a) and the 3 × 3 (b) periodicities, respectively. The Si atoms in the top and bottom vertical positions are depicted by the large red and small blue balls, respectively. The large gray balls depict the positions of the substrate Ag atoms. The simulated lateral unit cell is indicated by the dashed (pink) lines in the top view, and its boundary is indicated by the solid (black) lines in the side views. The obtained rhombic √3 × √3 periodicity is indicated by the solid (orange) lines in (a). 3.1. Monolayer Si on Ag(111) Structures of ML silicene have been explored by our previous DFT calculations [13, 14]. In this subsection, we present those results and then add more details. We have found two distinct stable structures with 3 × 3 silicene periodicity, and four distinct stable structures with √7 × √7 periodicity. The calculated cohesive energies of all the six structures are in the range of 5.906 - 5.975 eV in LDA, inferring the possible observation of several distinct structures. Actually the calculated STM images of the three most stable structures agree with the experimental images [9, 10, 12] excellently and two of the three metastable structures have been recently observed in experiments [42]. The most stable 3 × 3 silicene obtained by the present vdW-DF calculation is shown in Fig. 1(b), where 6 of 18 Si atoms in a unit cell protrudes by 0.86 A, whereas the remaining 12 Si keep nearly the same height in the lower positions. The structural difference between the LDA and the vdW-DF is tiny. The cohesive energy obtained by vdW-DF is 5.410 eV (Table 1), which is between the LDA value (5.972 eV) and the GGA value (5.215 eV). We have also found a √3× √3 ML silicene structure [Fig. 1(a)] for the 3×3 silicene/4×4 Ag(111) commensuration, in which one of six Si atoms in a unit cell protrudes about 1 A, whereas the remaining five Si keep nearly the same height in the bottom layer. Chen et al. [16] have also found two distinct √3 × √3 structures both of which are different from the structure in Fig. 1(a). We have clarified that the structures obtained by Chen et al. emerge when we perform the computation with insufficient k-point sampling [17]. The structural parameters for the stable √3 × √3 [Fig. 1(a)] and 3 × 3 [Fig. 1(b)] silicene on 4 × 4 Ag(111) are shown in Table 1 along with the cohesive energies Ec and Crossover between Silicene and Ultra-Thin Si Atomic Layers on Ag(111) Surfaces 6 Table 1. Calculated cohesive energy Ec (eV/Si), binding energy Eb (eV/Si) and structural parameters for the silicene layers and the surface layers of Si(111) on Ag(111). dsi−Ag (A) is the spacing between the bottom Si layer and the topmost Ag layer. Note that the Si layer is buckled. ∆z (A) represents the averaged amount of the buckling of the top Si layer and dsi−si (A) is the interlayer distance defined as the minimum spacing between top Si layer and its underlying Si layer. ML BL TL QL √3 × √3 3 × 3 √3 √3 × 3 × 3-AA √7 × √7-AA′ 2 × 1 7 × 7 √3 √3 × 3 × 3-AAA 3 × 3-ABC′ √7 × √7-AAA′ Ec 5.387 5.410 5.339 5.342 5.331 5.350 5.316 5.384 5.364 5.391 5.370 5.403 2 × 1 √3 √3 × 3 × 3-AAAA 3 × 3-ABCA′ √7 × √7-ABCA 5.404 5.418 5.384 5.423 5.434 2 × 1 Eb 0.63 0.65 0.78 0.79 0.80 0.58 0.63 0.83 0.77 0.85 0.80 0.81 0.86 0.73 0.88 0.89 0.88 dsi−Ag dsi−si ∆z 2.37 -- 1.05 2.17 -- 0.86 2.25 2.18 2.27 2.26 2.23 2.26 2.18 2.26 2.26 2.28 2.28 2.18 2.26 2.26 2.27 0.96 2.58 0.81 2.47 0.99 2.39 2.25 0.98 1.40 -- 2.53 2.48 2.41 2.40 2.24 2.52 2.49 2.44 2.49 2.24 1.08 0.89 1.06 1.10 1.05 1.09 0.88 1.17 0.95 1.06 √3 structure is 23 meV/Si less stable the binding energies Eb. From Table 1, the √3 × than the 3× 3 silicene on Ag(111) which has been widely observed in experiments. This shows that the √3× √3 ML silicene is hard to be synthesized, being consistent with the previous experimental results [19, 20]. As for the π-bonded chain structure, commonly observed on Si(111) surface, the topmost two surface layers show drastic reconstruction. Hence the ML silicene cannot evolve to the π-bonded chain structure. 3.2. Bilayer Si on Ag(111) In this subsection, we present the results for the BL silicene. We focus on the 3 × 3 BL silicene on 4 × 4 Ag(111) surface and on the √7 × √7 BL silicene on √13 × √13 Ag(111) , where the lattice mismatches are 0.4 % and 2.6 % (with respect to the (111) plane of diamond-structured Si), respectively. Another degree of freedom is the way of Crossover between Silicene and Ultra-Thin Si Atomic Layers on Ag(111) Surfaces 7 (a) (b) (c) Figure 2. (color online) Top (upper panels) and side (middle panels) views of the calculated stable structures and corresponding STM images (bottom panels) of the three rhombic √3 × √3 bilayer (BL) silicene on Ag(111). The (a), (b) and (c) are labeled as str1, str2 and str3 of BL silicene, respectively. The large gray balls and the small yellow balls depict the positions of Ag atoms of substrate and the Si atoms of the bottom Si layer, respectively. The large red balls and small blue balls depict the positions of protruded and unprotruded Si atoms of the top Si layer. The lateral unit cells are indicated by the dashed (pink) lines in the top views, and their boundaries are depicted by the solid (black) lines in the side views. The obtained rhombic √3 × √3 periodicity is indicated by the solid (orange) lines. stacking of the two Si layers. When the two planar hexagonal networks are stacked in the same way, we call it the AA stacking, whereas when the two networks are rotated by 60 degrees we call it the AB stacking, following the convention of graphite. From these two different stacking ways, we start the geometry optimization. We also examine stability of the π-bonded chain structure and the DAS structure for this BL Si atomic layers on the Ag(111) surface. One of the interesting features is the multi-stability of various structures of which the cohesive energies are in the range of 5.316-5.350 eV. Starting from the 3 × 3 BL silicene on the 4 × 4 Ag(111) with AB stacking, we have reached three stable √3 × √3 rhombic silicene structures shown in Fig.2. The by about 1 A from the remaining five Si atoms in the lateral √3 × similar to the √3 × three structures, labeled as str1, str2 and str3, are equivalent to each other when we neglect the existence of the Ag substrate: In the top layer, a single Si atom is protruded √3 periodicity, √3 structure of ML silicene in Fig. 1(a). In the bottom silicene layer, Si atoms are buckled with the amount of about 0.8 A, as they are on the (111) plane of diamond-structured Si. These features are common for the three structures. Moreover, the total-energy difference among the three structures are within 0.36 meV/Si, inferring coexistence of all three structures in usual experiments. These structures can √3 unit cells along their be transformed to each other by moving the rhombic √3 × diagonal direction. Crossover between Silicene and Ultra-Thin Si Atomic Layers on Ag(111) Surfaces 8 (a) (b) (c) (d) Figure 3. (color online) Top (upper panels) and side (middle panels) views of the calculated stable structures and corresponding STM images (bottom panels) of BL silicene and the π-bonded chain and the DAS structures on Ag(111). (a), (b) The 3 × 3 and the √7 × √7 BL silicene structures, respectively. (c), (d) The 2 × 1 and the 7 × 7 BL Si(111)-surface structures, respectively. The color codes are the same as in Fig. 2. The obtained 2 × 1 periodicity in (c) is indicated by the solid (orange) lines. The calculated STM images [43] of the three √3 × √3 structures show the perfect rhombic geometry, which agrees well with that observed by Chen et al. in experiment below 40K [16]. As will be discussed in section 4, the flip-flop motion would happen between two of the three structures at higher temperatures. As a result, the honeycomb will be obtained. √3 × √3 STM image [Fig. 8(a)] that is widely reported in the experiments [19, 21, 22] The structure of the 3× 3 BL silicene on the 4× 4 Ag(111) with AA stacking keeps its periodicity even after the geometry optimization [Fig.3(a)]. The top layer of the structure resembles the 3×3 ML silicene on Ag(111) [Fig. 1(b)]. The corresponding STM image is further calculated , where each protruded Si atom in the top layer corresponds to a bright spot [bottom panel of Fig. 3(a)]. The STM image is nearly the same as that of the 3 × 3 ML silicene on Ag(111) which has been extensively observed in experiments [9]. It is known that the usual experimental approaches (STM and LEED) can only reflect the geometry feature of the top silicene layer. This means that people are easily mislead by the STM/LEED geometries in identifying the thickness of silicene in experiments. Although, the height along particular direction of the silicene layers can be measured and thus used to distinguish silicene thickness, it does not work when the Ag surface is not well verified [44] or the Ag surface is not flat[20]. This indicates that not only the STM measurement but also the determination of the Si coverage is important for the structural identification. As for the √7×√7 BL silicene on √13×√13 Ag(111), we have also considered both AA and AB stacking cases. The most stable structure we have found is the √7 × √7 BL silicene with AA′ stacking [Fig. 3(b)], where A′ means the in-plane dislodgment of the top Si layer from the proper A stacking. The calculated STM image [bottom panel Crossover between Silicene and Ultra-Thin Si Atomic Layers on Ag(111) Surfaces 9 of Fig. 3(b)] further shows the perfect √7×√7 periodicity with respect to that of 1× 1 present the 1 × 1 and √7 × √7 periodicities, which are 1.5 meV/Si and 29 meV/Si silicene. The most stable AB-stacking and AA-stacking structures we have obtained higher in energy than the AA′-stacking structure, respectively. The cohesive energies for the BL silicene on Ag(111) are shown in Table 1, all of which are larger than that of freestanding silicene (4.75 eV/Si in our calculation), showing the catalytic function of Ag in forming silicene. On the other hand, the cohesive energies of BL silicene on Ag(111) are smaller than those of the ML silicene on Ag(111) by 37-94 meV/Si. This explains why the ML silicene structures are usually synthesized before the BL silicene in experiments [18, 19]. The 3 × 3 BL silicene with AA stacking on Ag(111) [Ec=5.342 eV/Si, Fig. 3(a)] is the most stable among all the √3 BL silicene structures, which is lower in the total energy by 3 meV/Si than the √3× silicene structures (AB stacking, Fig. 2) widely observed in experiments [16, 18, 19, 20]. Considering that the 3 × 3 BL silicene exhibits nearly the same STM image as the 3 × 3 ML silicene on Ag(111) [9], this strongly indicates the existence of the 3 × 3 BL silicene structure on the Ag surface, which may be misrecognized to be the ML silicene √7 BL silicene on Ag(111) is higher in in previous experiments. In addition, the √7 × the total energy by about 10 meV/Si than the 3 × 3 and √3 × √3 structures, showing that it is rarely synthesized in experiment. To explore the possibility of Si(111) reconstructed surface emerging on the Ag(111) substrate, we deposit the topmost two surface layers of the π-bonded chain structure [24] and the DAS structure [25, 26] on Ag(111), though the two top-layer atoms in the DAS structure is incomplete for the full reconstruction in the DAS model. After geometry optimization, it is found that both structures are stable [Figs. 3(c) and 3(d)]. The obtained stable 2× 2 Si surface shows the periodicity of 2× 1 in which the bond network in the Si double layer contains five-membered and seven-membered rings and thus two top surface atoms in a unit cell form a π-bonded chain, being essentially identical to the π-bonded structure on the Si(111) surface. We have also calculated the corresponding STM image. As shown in the bottom panel of Fig. 3(c), the STM image agrees well with that of the pure 2×1 Si(111) surface [45]. These results clearly show the emergence of the 2 × 1 Si(111)-surface structure on Ag(111). The 7 × 7 Si(111)-surface structure is also preserved when we deposit the top two Si layers on Ag(111) [Fig. 3(d)]: the top layer composed of the 12 Si atoms in a unit cell keeps nearly the same geometry as that of the pure 7 × 7 Si(111) surface, and the geometry of its underlying layer composed of the remaining 90 Si atoms is also basically preserved on Ag surface. The calculated STM image [bottom panel of Fig. 3(d)] is similar to that of the pure 7 × 7 Si(111) surface [46]. The above results clearly show that the bottom Si layer acts as a buffer layer on the Ag surfaces, which preserves the structure of the Si layers above it. We have also calculated the cohesive energies for the Si(111)-BL system deposited on Ag(111). It is found that the π-bonded chain structure on the Ag(111) is the most stable among all the BL Si on Ag(111). The cohesive energy of the π-bonded chain Crossover between Silicene and Ultra-Thin Si Atomic Layers on Ag(111) Surfaces 10 structure (Ec=5.35 eV/Si) is larger by about 10 meV/Si than those of the 3×3 (Ec=5.342 eV/Si) and the √3 × √3 (Ec=5.339 eV/Si) BL silicene structures (Table 1). As we will explain below, this difference in the cohesive-energy mainly come from the difference in the stability of the top Si layer. A common feature for these BL Si structures (except for the 7 × 7 DAS structure) on Ag(111) is that half amount of Si atoms in the bottom layer located in the lower positions are chemically bonded with the Ag atoms, whereas the remaining half amount of Si atoms located at the higher positions are chemically bonded with the Si atoms in top layer (side views of Figs. 2 and 3). This makes the bottom layer of Si being sp3 bonded with the Ag surface atoms and the top-layer Si atoms. If we consider the Ag substrate and the bottom Si layer as a united system, it can be expected that the cohesive energies of this united systems are nearly the same for various BL Si structures on Ag(111). Therefore, the cohesive-energy difference mainly comes from the surface- energy difference of the top layer for all the BL Si structures on Ag(111). To confirm this, we have explored the stable structures of pure Si(111) surface with the √3 × √3 periodicity and found that its cohesive energy is Ec=5.18 eV/Si, being smaller than the pure 2 × 1 π-bonded Si(111) surface by about 10 meV/Si. This amount coincides √3 and 2 × 1 BL Si structures on with the cohesive-energy difference between the √3 × It should be mentioned that the 7 × 7 BL Si(111)-surface structure does not follow the rule stated above. The 7 × 7 DAS structure is most stable on Si(111) surface, whereas it is least stable as the BL system on the Ag(111) (Table 1). This is due to the complex reconstruction in the DAS model in which subsurface atoms are relaxed substantially. The BL system on the Ag(111) is not thick enough to incorporate the complex reconstruction. However, we expect that the 7× 7 Si(111)-surface structure on Ag(111) would be the most stable when the number of Si layers increases. Yet this issue will not be discussed in this paper mainly due to the limited resource of the computation. Ag(111). 3.3. Tri-layer Si on Ag(111) In this subsection, we present the results for the TL silicene. Let us start with 3× 3 TL silicene on 4 × 4 Ag(111). Interestingly, we have found three stable √3 × √3 rhombic silicene structures for the TL silicene [Fig. 4], as in the 3 × 3 BL silicene. The three structures are transformed to each other and the total energies of the three structures are nearly the same, showing the tristability of the √3 × √3 TL silicene. The three Si layers are stacked as ABC, being identical to the stacking of the Si(111). The structure √3 of the top silicene layer of the TL silicene is very similar to that in the BL √3 × silicene. The calculated STM images are nearly the same with those of the √3 × √3 Since we have found the 3 × 3 AA-stacking BL silicene [Fig. 3(a)], we explore the similar 3×3 TL silicene structure with AAA stacking where the top layer resembles that of the AA-stacking BL silicene. As a result, the 3×3 structure is found to be unstable and BL silicene on Ag(111) [the bottom panels of Fig. 4]. Crossover between Silicene and Ultra-Thin Si Atomic Layers on Ag(111) Surfaces 11 (a) (b) (c) Figure 4. (color online) Top (upper panels) and side (middle panels) views of the calculated stable structures and corresponding STM images (bottom panels) of the three rhombic √3 × √3 tri-layer (TL) silicene on Ag(111). The (a), (b) and (c) are labeled as str1, str2 and str3 of TL silicene, respectively. The large gray balls and the small yellow balls depict the positions of Ag atoms of substrate and the Si atoms under the top Si layer, respectively. The large red balls and small blue balls depict the positions of protruded and unprotruded Si atoms of the top Si layer. The simulated lateral unit cell is indicated by the dashed (pink) lines in the top view, and its boundary is indicated by the solid (black) lines in the side views. The obtained rhombic √3×√3 periodicity is indicated by the solid (orange) lines. (a) (b) (c) (d) (e) Figure 5. (color online) Top (upper panels) and side (middle panels) views of the calculated stable structures and corresponding STM images (bottom panels) of the TL silicene and Si(111)-surface structures on Ag(111). (a), (b) The two symmetry-related 3 × 3 silicene structures with the AAA stacking, which are labeled as str1 and str2, respectively. (c) The 3 × 3 silicene structure with ABC′ stacking. (d) The √7 × √7 silicene structure. (e) The 2× 1 Si(111) π-bonded chain structure. The color codes are the same as in Fig. 4. The obtained 2 × 1 periodicity in (e) is indicated by the solid (orange) lines. Crossover between Silicene and Ultra-Thin Si Atomic Layers on Ag(111) Surfaces 12 relaxes into either of other two 3 × 3 structures with AAA stacking [str1, str2 shown in Figs. 5(a) and 5(b)], where 5, not 6 in the case of the BL silicene, of 18 Si toms in the top layer protrude by about 0.9 A. The resultant two structures are symmetrically equivalent if we neglect the existence of the Ag substrate. The total energy difference between the two symmetry-related structures is therefore nearly zero, showing the bistability of the 3× 3 structures in TL silicene. The calculated STM images of the two structures reflect the characteristics of the top-layer atom arrangements [bottom panels of Figs. 5(a) and 5(b)]. We have found a new 3 × 3 TL silicene structure where the stacking of the three Si layers are dislodged [Fig. 5(c)]: The stacking is ABC′ where the top layer C is dislodged in the lateral plane from the proper C stacking as in diamond-structured Si. In the top layer, 4 of 18 Si atoms are protruded by about 1 A, whereas the remaining 14 Si atoms locate in the lower positions. This causes very different STM profile from the other Si structures on Ag(111) [bottom panel of Fig. 5(c)]. Since the cohesive energy of this structure is large enough (see below), this unusual STM pattern should be observed. As for the √7 × √7 TL silicene on √13 × √13 Ag(111), we have found that the √7 periodicity with the AAA′ stacking [Fig. 5 (d)]. The structure of top silicene layer and the STM pattern are nearly the same as most stable structure keeps the √7 × those of the √7 × √7 BL silicene [Fig. 3 (b)]. The most prominent characteristic of the stable ML, BL, and TL silicene obtained in the present calculations is the protrusion of some of top-layer Si atoms causing the nearby buckling of the top layer. This structural relaxation was originally proposed to explain 2 × 1 Si(111) surface by Haneman [47] before the π-bonded chain model was proposed by Pandey [24]. Hence it is recognized that the restriction in the lateral plane caused by the Ag substrate realizes the buckling relaxation which is irrelevant in the free Si(111 ) surface. The buckling structure caused by the external environment may induce unexpected phenomena such as spin polarization [48]. To verify the stability of the π-bonded structure in the TL silicene on the Ag(111), we deposit the topmost three surface layers of the π-bonded chain structure on Ag(111). After geometry optimization, it is found that the π-bonded chain structure is stable [Figs. 5(e)]. The calculated STM image [bottom panel of Fig. 5(e)] agrees well with that of the 2 × 1 π-bonded chain structure on the Si(111) surface [45]. The calculated cohesive energies for the TL Si structures on Ag(111) are shown in Table 1. A common feature is that the cohesive energies in the TL structures are obviously larger than those of BL structures by tens of meV/Si, indicating that it is easier to obtain the TL Si structures on Ag(111). The most stable silicene structure is the 3× 3 silicene with ABC′ stacking [Ec=5.391 eV/Si, Fig. 5(c)]. This is different from that in the BL silicene, where the 3 × 3 silicene with AA stacking is the most stable. The cohesive energy of the √3×√3 TL silicene structure (Ec=5.384 eV/Si) is very close to that of the most stable TL silicene structure, indicative of their coexistence. On the √7 silicene structures with AAA and AAA′ stacking other hand, the 3 × 3 and √7 × are smaller in Ec by 20-26 meV/Si than the most stable one, indicating that it may be Crossover between Silicene and Ultra-Thin Si Atomic Layers on Ag(111) Surfaces 13 (a) (b) (c) Figure 6. (color online) Top (upper panels) and side (middle panels) views of the calculated stable structures and corresponding STM images (bottom panels) of the three rhombic √3 × √3 quad-layer (QL) silicene on Ag(111) labeled as str1 (a), str2 (b) and str3 (c), respectively. The large gray balls and the small yellow balls depict the positions of Ag atoms of the substrate and the Si atoms under the top Si layer, respectively. The large red balls and small blue balls depict the positions of the protruded and unprotruded Si atoms of the top Si layer. The simulated lateral unit cell is indicated by the dashed (pink) lines in the top view, and its boundary is indicated by the solid (black) lines in the side views. The obtained rhombic √3×√3 periodicity is indicated by the solid (orange) lines. difficult to observe them in experiments. It is noteworthy that the 2 × 1 π-bonded chain structure is stable even on the Ag(111) substrate and further its cohesive energy (Ec=5.403 eV/Si) is larger than that of the most stable silicene structure (Ec=5.391 eV/Si) by about 10 meV/Si. This confirms our argument stated above: The relaxed or reconstructed Si(111)-surface structures on Ag(111) are more stable than the silicene structures owning to their lower surface energies. 3.4. Quad-layer Si on Ag(111) In this subsection, we present the results for the QL silicene. Starting from the 3 × 3 √3 TL silicene on the 4 × 4 Ag(111) surface, we have reached the three stable √3 × rhombic silicene structures, as in the BL and TL cases. The stacking is ABCA [Fig. 6] which seems to be a natural extension of the TL silicene with the ABC stacking. The structure of the top silicene layer and the STM pattern are nearly the same with those of the BL [Fig. 2] and TL [Fig. 4] √3 × √3 silicene on Ag(111). The 3 × 3 silicene structures of AAAA [Fig. 7(a)] and ABCA′ [Fig. 7(b)] stackings Crossover between Silicene and Ultra-Thin Si Atomic Layers on Ag(111) Surfaces 14 (a) (b) (c) (d) Figure 7. (color online) Top (upper panels) and side (middle panels) views of the calculated stable structures and corresponding STM images (bottom panels) of the QL silicene and the π-bonded chain structures on Ag(111). (a), (b) The 3 × 3 silicene structures with AAAA and ABCA′ stacking, respectively. (c) The √7 × √7 silicene structure with ABCA stacking. (d) The 2 × 1 π-bonded chain structure. The color codes are the same as in Fig. 6. The obtained 2 × 1 periodicity in (d) is indicated by the solid (orange) lines. are also stable. The AAAA structure seems to be an extension of the structures of TL silicene with the AAA stacking [Figs. 5(a) and 5(b)]. While the top-layer structure and the STM pattern present a different rotational angle with respect to the Ag surface. Other structure mapped by the symmetry operations is not found for the QL case. In the ABCA′ structure, the bottom 3 Si layers present the perfect ABC stacking. The top Si layer (A′ layer) presents a in-plane dislodgment from the proper A stacking, the geometry of which resembles the C′ layer of ABC′-stacking structure in the TL case, but with a different rotational angle with respect to the Ag surface. Thus the STM pattern [bottom pattern of Figs. 7(b)] presents some similarity with that of the ABC′ TL silicene on Ag(111) [bottom pattern of Figs. 5(c)]. The most stable structure of the √7 × √7 silicene on √13 × √13 Ag(111) also presents the √7 × √7 silicene periodicity [ABCA stacking, Fig. 7(c)] as that in the TL case [Fig. 5(d)], where 3 of 14 Si atoms in the top layer protruded by about 1 A constitute the triangle geometry. On the other hand, the STM images of the two √7 structures are different from each other [bottom panels of Fig. 7(c) and √7 × Fig. 5(d)] due to the different positions of the protruded Si atoms. This shows the Si thickness dependence of the most stable silicene structures for the √7 × √7 silicene on √13 × √13 Ag(111). To verify the stability of the 2 × 1 π-bonded structure in the QL silicene on the Ag(111), we deposit the topmost four surface layers of the π-bonded chain structure on Ag(111) and perform the geometry optimization. As a result, the π-bonded chain Crossover between Silicene and Ultra-Thin Si Atomic Layers on Ag(111) Surfaces 15 structure is also stable [Fig. 7(d)] on Ag(111) and the calculated STM image [bottom panel of Fig. 7(d)] agrees well with that of the 2 × 1 π-bonded chain structure on the Si(111) surface [45]. silicene structure [Ec=5.418 eV/Si] is very close to it, showing their coexistence. The The calculated cohesive energies for the QL Si structures on Ag(111) are shown in Table 1. The cohesive energies in the QL structures are obviously larger than those of the BL and TL structures, showing the increment of cohesive energies with the number of layers for multilayer silicene. The most stable silicene structure is the 3 × 3 structure with ABCA′ stacking [Ec=5.423 eV/Si, Fig. 7(b)]. The cohesive energy of the √3×√3 √7×√7 and 3×3 silicene structures with ABCA and AAAA stackings are smaller in Ec by 19-39 meV/Si than the most stable one, implying that it is hard to synthesize them in experiments. The cohesive energy (Ec=5.434 eV/Si) of 2×1 π-bonded chain structure is larger than that of the most stable silicene structure also by about 10 meV/Si, confirming that the relaxed or reconstructed Si(111)-surface structures are more stable than the silicene structures on Ag(111). The reason why the 2 × 1 π-bonded chain and/or 7 × 7-DAS structures have not been observed on Ag(111) in experiments is attributed to the synthesizing method, where the structures are always produced via depositing the Si atoms on Ag(111). This makes the Si structures on Ag(111) grown layer by layer: the BL Si structures are grown based on the structures of ML silicene and the TL Si structures are on the BL Si structures etc. Since the structure of ML silicene is very different from that of the Si(111) layer, it is not expected to obtain the 2 × 1 BL Si(111)-surface structure on Ag(111) by using the synthesizing method widely adopted to make silicene. It is similar for synthesizing thicker (TL, QL and so forth) Si structures on Ag(111). On the other hand, it is well known that the 2×1 Si(111) structure is obtained by cleaving the Si(111) surface at room temperature, and it further converts to the more stable 7 × 7 structure when heated above 400 ◦C. This shows that different experimental approaches produce the different Si structures. Therefore, although the 2 × 1 and 7 × 7 Si structures are more stable than the silicene structures on Ag(111), they have never been observed in experiment because no one has performed such experiments as heating the multilayer silicene structures at high temperatures to convert the silicene structures to the 2 × 1 or 7 × 7 Si(111) structures. Table 1 also shows the structure details and the binding energies for all the Si structures from ML to QL on Ag(111). The binding energies for all the structures (the energy gain in the deposition of Si layers on the Ag surface) are in the range of 0.6-0.9 eV/Si, larger than the typical vdW interaction energy manifested in the cases of graphene on metal surfaces by an order of magnitude [49], showing the covalent nature of the Si-Ag interface interactions. Moreover, the spacings between the bottom Si layer and Ag surface are in the range of 2.17-2.37 A. Considering that the atomic radii of Si and Ag are 1.18 and 1.65 A, respectively, the small spacing values confirm the formation of the Si-Ag covalent bonds. This result agrees with the recent DFT results which studied various monolayer silicene structures on Ag(111) [31, 50]. While Crossover between Silicene and Ultra-Thin Si Atomic Layers on Ag(111) Surfaces 16 Figure 8. (color online) Calculated reaction energy profile among the three rhombic √3 × √3 structures of BL (a), TL (b) and QL (c) silicene on Ag(111), shown in Figs. 2, 4 and 6, respectively. (d) The calculated reaction energy profile between the two rotation-symmetric 3×3 structures of TL silicene with AAA stacking on Ag(111) [Figs. 5(a) and 5(b)]. The energy in the vertical axis is defined as (E − Estr1)/n, where E and Estr1 are the total energy of the structure at reaction coordinate and the total energy of str1, respectively. n is the number of atoms in the top silicene layer. The transition barriers between str1 and str2, str2 and str3, and str1 and str3 are depicted by the black squares, blue dots and red triangles, respectively. The insets show the STM images obtained by superimposing the images of str1 and str2 structures. The simulated 3 × 3 lateral unit cells in the top views are indicated by the dashed (pink) lines. it disagrees with the argument in Ref. [51] that the Si-Ag interaction is weak. On the other hand, the interlayer distances between the top Si layer and the underlying layer for all the structures are smaller than 2.6 A. This shows the covalent nature of Si-Si interlayer interactions. As discussed below, the covalent interactions of Si-Ag interface and Si-Si interlayer would significantly influence the electronic properties of silicene. 4. Flip-flop motion in multilayer silicene on Ag(111) In this section, we present the flip-flop motion of multilayer silicene on Ag surface. We focus on the √3× √3 tristable BL, TL and QL silicene structures and the 3× 3 bistable Crossover between Silicene and Ultra-Thin Si Atomic Layers on Ag(111) Surfaces 17 TL silicene structures. The calculated transition barriers are indicative of the flip-flop motion among/between the tristable/bistable silicene structures at low temperature. The results agree well with the STM images observed in recent experiments. transition barriers among the three √3 × In use of the nudged elastic band (NEB) method [52, 53], we have calculated the √3 structures, str1, str2, str3 of each BL, TL and QL silicene on Ag(111) [Figs. 8(a), 8(b) and 8(c)]. The calculated transition barriers for the BL silicene are in the range of 7-9 meV/Si, whereas they are in the range of 14-28 meV/Si for the TL and QL silicene on Ag(111). Anyhow, these barriers are low enough to allow the structural transition to be thermoactivated at temperatures of dozens of Kelvin. The insets of Figs. 8(a), 8(b) and 8(c) show superpositions of two of the three STM images for the BL, TL, and QL silicene structures, respectively, √3 structures. These results agree well which present the perfect honeycomb √3 × It is noted that two other models had been recently proposed to explain the √3 × √3 bilayer silicene with the recent experimental observations [16, 19, 20, 21, 22, 35]. structure observed in experiments [54, 55]. However, both of them can not explain all the experimental observations at lower (below 40K, rhombic structure)[16] and higher temperatures (above 77K, honeycomb structure) [16, 19, 20, 21, 22, 35]: The model in Ref.[54] cannot explain the honeycomb STM structure and the model in Ref.[55] cannot explain its conversion from the honeycomb STM structure to the rhombic STM structure at lower temperature [16]. From this sense, our model agrees better with the experimental observations so far. From the previous section, it is recognized that the bottom Si layer acts as a buffer layer on the Ag surfaces, above which the structure √3 silicene structures and the low transition barriers among them are also expected to exist in the thicker multilayer silicene on Ag(111). This explains the experimental observations where the of Si layers can be well preserved. Thus the tristable √3 × same honeycomb √3 × √3 STM images are observed independent of the thickness of silicene [16, 18, 19, 20, 21, 22, 35]. The flip-flop motion also happens in other silicene structures. Fig. 8 (d) shows the calculated transition barriers between the two bistable structures of the 3×3 TL silicene with AAA stacking [Figs. 5(a) and 5(b)]. The very small energy barrier (3.5 meV/Si) shows that the flip-flop motion can happen at very low temperature. The inset shows the superposition of the STM images of the two structures, which exhibits the similar honeycomb 3 × 3 STM image as that of the 3 × 3 ML and BL silicene [bottom panel of Fig. 3(a)] on Ag surfaces. The results confirm the characteristic of flip-flop motion in multilayer silicene on Ag(111) at low temperature. 5. Dirac states and surface-localized π states for the multilayer Si on Ag(111) Previously, we have studied the electronic properties of 3 × 3 and √7 × √7 ML silicene on Ag(111), which are very different from that of the freestanding silicene [13, 14]. A common feature for all the silicene structures on Ag(111) is that the Dirac π and π∗ states Crossover between Silicene and Ultra-Thin Si Atomic Layers on Ag(111) Surfaces 18 2 1 0 -1 -2 2 1 0 -1 ) V e ( y g r e n E ) V e ( y g r e n E !"#$ !%#$ (cid:46) (cid:42)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:48)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:46)(cid:3) ) V e ( y g r e n E 2 1 0 -1 -2 !+#$ (cid:46)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:42) (cid:3)(cid:48)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:46) !'#$ !(#$ 2 1 0 -1 ) V e ( y g r e n E !)#$ 2 1 0 -1 ) V e ( y g r e n E !&#$ (cid:46)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:42)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:48)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:46) !*#$ 2 1 0 -1 -2 2 1 0 -1 ) V e ( y g r e n E ) V e ( y g r e n E -2 -2 (cid:46)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:42) (cid:3)(cid:48)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:46)(cid:3) (cid:46)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:42)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:48)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:46) -2 -2 (cid:46)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:42) (cid:3)(cid:48)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:46)(cid:3) (cid:46)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:42)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:48)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:46) Figure 9. (color online) Calculated energy bands of the 3×3 silicene on 4×4 Ag(111), where the silicene exhibits the √3 × √3 structure. The energy bands of the ML, BL, TL, and QL silicene on Ag(111) are shown in (a), (b), (c) and (d). The energy bands of the freestanding ML, BL, TL, and QL silicene that are peeled from the Ag surface are shown in (e), (f), (g), (h), respectively. The origin of the energy is set to be EF. The states that have characters of the π (π∗) for the ML and BL silicene on Ag(111) are indicated by the blue circles on Γ point. The surface-π states in TL and QL silicene are indicated by the blue squares on Γ point. The states of the TL silicene on Ag(111) that have characters of the surface-π along the Γ − K direction are marked by the sequence of black squares in (c), which present linear energy dispersion. marked by the squares in Figs. 9(c) and (d) of the √3×√3 TL (a) and QL (b) silicene surface of the KS orbitals marked in Figs. 9(g) and 9(h) of the √3 × √3 TL and QL on Ag(111). Figures (c) and (d) show the corresponding contour plots and isovalue Figure 10. (color online) Contour plots (left panels) and isovalue surface at its value of 10% of the maximum value (right panels) of the squared Kohn-Sham (KS) orbitals silicene peeled from the Ag surfaces. The gray and blue balls depict Ag and Si atoms, respectively. Crossover between Silicene and Ultra-Thin Si Atomic Layers on Ag(111) Surfaces 19 near EF in the freestanding silicene disappear due to the strong silicene-Ag interactions. In this section, we present the results for the √3×√3 silicene on Ag(111) because they are the only well observed multilayer silicene structures. Since the √3 × √3 structure has stable phases from ML to QL, such study can help us to understand the evolution of electronic properties with the increasing Si thickness on Ag surfaces. Additionally, we explore the effect of Ag substrate on the electronic properties of Si(111)-surface structures, i.e., the 2 × 1 π-bonded chain structure. The energy bands are calculated in the periodicity of 3×3 silicene on 4×4 Ag(111). Figs. 9(a) and 9(b) show the calculated energy bands for the ML and BL silicene on Ag(111), where the K point in the BZ of the 1 × 1 silicene is folded on the Γ point. For both ML and BL cases, we have found no apparent energy band near EF that shows the linear energy dispersion peculiar to Dirac π (π∗) state. Instead, the states with characters of π (π∗) appear deep below EF [indicated by blue circles in Figs. 9(a) and 9(b)]. The energy bands of ML and BL silicene which we obtain by peeling from the Ag surface are further calculated. It is found that the states with characters of π (π∗) shift upward near EF from the deep positions in the valence bands due to the annihilation of the interaction of Ag substrate [indicated by blue circles in Figs. 9(e) and 9(f)]. This shows that the covalent interactions between silicene and Ag surface makes the π (π∗) states being deep in the valence bands. We also find no linear energy band peculiar to the Dirac π (π∗) state near EF for the TL and QL silicene on Ag(111) [Figs. 9(c) and 9(d)]. We have additionally calculated the energy bands of the TL and QL silicene peeled from the Ag surfaces [Figs. 9(g) and 9(h)]. As a result, no Dirac characteristic π (π∗) linear energy band near EF is found. This implies that the covalent Si-Si interlayer interactions also cause the absence of Dirac states. These results strongly indicate the absence of Dirac states for multilayer silicene on Ag(111) due to the covalent interactions at the Si-Ag interface and in the Si layers (Table.1). On the other hand, we have found a new state near EF for the TL and QL silicene on Ag surfaces, as indicated by the blue squares on Γ point in Figs. 9(c) and 9(d), which are mainly composed of the π orbitals located at the top silicene layer [Figs. 10(a) and 10(b)]. Here we call the new state the surface-π state. The distributions of the Kohn-Sham (KS) orbitals for the surface-π state [right panels of Figs. 10(a) and 10(b)] are similar with those of the Dirac π (π∗) orbitals in the freestanding ML silicene, both of which present the hexagonal symmetry. Thus we can also expect that the surface-π state exhibits linear energy dispersion near EF [1]. Through the detailed analysis of the KS orbitals along the Γ-K direction in BZ for all the energy bands, we have indeed observed such linear energy dispersion nearby EF for the surface-π state, as indicated by the black squares in Figs. 9(c) for the TL silicene on Ag(111). The Fermi velocity of the surface-π state is much smaller than that of the Dirac states in the freestanding ML silicene because of the (√3 times) larger distances between the two nearest π orbitals which cause the weaker transfer energy. Moreover, the KS orbitals of the surface-π states are similar for the TL and the QL silicene on Ag(111) [Figs. 10(a) √3 and 10(b]. This implies that the surface-π state always exists near EF for the √3 × Crossover between Silicene and Ultra-Thin Si Atomic Layers on Ag(111) Surfaces 20 (a) (c) (b) (d) (color online) Calculated energy bands (a) and isovalue surface at its Figure 11. value of 25% of the maximum value of squared KS orbitals (side view) of the tri-layer 2 × 2 Si(111) on √7 × √7 Ag(111), where the Si surface presents 2 × 1 π-bonded chain structure [Fig. 5(e)]. The energy bands and isovalue surface of squared KS orbitals of the TL Si structure peeled from Ag surface are further shown in (b) and (d), respectively. The states that have characters of the isolated π (π∗) are indicated the blue squares on Γ point in (a) and (b). The gray and blue balls depict Ag and Si atoms, respectively. energy bands of the √3 × multilayer silicene on Ag(111), irrespective of the thickness of silicene. We expect that the surface-π state has great potential applications in the field of spin electronics [48]. To understand the origin of the surface-π state, we have further calculated the √3 TL and QL silicene that are peeled from the Ag surface. As a result, the surface-π states near EF are also obtained [indicated by the blue squares in Figs. 9(g) and 9(h)] , although it contains π orbitals on the subsurface Si layer for the TL silicene [Fig. 10(c)]. It is noteworthy that the geometries of the surface-π orbitals of the TL and QL silicene peeled from Ag(111) [Figs. 10(c) and 10(d)] are similar with those of the TL and QL silicene on the Ag(111) [Figs. 10(a) and 10(b)]. This shows √3 Si surface structure. Therefore, one can always expect the appearance of surface-π state near EF in the thick that the surface-π state is originated from the unique √3 × √3 × √3 multilayer silicene structures. Since the Si(111)-surface structures are also stable on Ag surfaces, it is important to explore the effects of Ag substrate on the electronic properties of Si(111)-surface structures. It is well known that the 2 × 1 π-bonded chain Si surface presents a semiconductor property, where both the valence band maximum (VBM) and the Crossover between Silicene and Ultra-Thin Si Atomic Layers on Ag(111) Surfaces 21 conduction band minimum (CBM) come from the isolated π (π∗) states of the π-bonded atomic chain [56, 57]. Here we have calculated the energy bands of the TL 2 × 1 π- bonded chain structure on Ag(111), where the unit cell adopted in the calculation has the periodicity of 2×2 Si(111) on √7×√7 Ag(111) [Fig. 5(e)]. The calculated energy bands are shown in Fig. 11(a), which present the metallic property. Through the analysis of calculated KS orbitals, we find the metallic bands are mainly from the sates containing both the Ag orbitals and Si orbitals, as well as the sp band of Ag substrate. The states with characters of isolated π and π∗ appear 0.2 eV below and 0.5 eV above EF on Γ point, respectively, as indicated by the blue squares of Fig. 11(a). The corresponding KS orbitals are shown in Fig. 11(c), which clearly shows the isolated π and π∗ states located on the π-bonded atomic chain. We have also calculated the energy bands of the π-bonded chain structure that are peeled from the Ag surface [Fig. 11(b)]. The isolated π and π∗ states are found about 0.5 eV below and above EF on Γ point [indicated by the blue squares in Fig. 11(b)], respectively. The corresponding KS orbitals are further shown in Fig. 11(d), which resemble those of the 2 × 1 structure on Ag(111), although there are some additional π orbitals on the subsurface Si layer, as shown in left panel of Fig. 11(d). The energy levels of the isolated π and π∗ states of the peeled 2 × 1 structure are close to those on the Ag surface, showing the minor effect of the Ag-Si interface interactions on the isolated π and π∗ states of top Si layer. We have also obtained the similar results for the QL π-bonded chain structure on Ag(111). Combining with the results of the √3 × √3 silicene on Ag(111), it can be concluded that the strong Ag-Si interface interactions do not destroy the surface-related π states of the multilayer Si structures on Ag surfaces. 6. Conclusion We have performed the density-functional calculations for the ultra-thin atomic layers of Si on Ag(111) surfaces, with the thickness of Si layers varying from monolayer to quad-layer. We have found several stable structures for the silicene on Ag(111) with √3, 3× 3, and √7× √7 with respect to the 1× 1 silicene. We have periodicities of √3× also found that the 2 × 1 π-bonded chain and 7 × 7 DAS Si(111)-surface structures are stable on Ag surfaces, showing the crossover of silicene and Si(111)-surface structures on Ag(111). From the calculated cohesive energies, several common features can be deduced: (1) ML silicene is more stable than the BL and TL silicene on Ag(111) but becomes less stable than QL silicene; (2) Cohesive energies of multilayer silicene and Si(111)-surface structures on Ag(111) monotonically increase with their thickness increasing; (3) The Si(111)-surface structures are more stable than the silicene structures on Ag(111) due to the lower surface energies. Moreover, we have found the structural tristability and bistability for the √3 × √3 and 3 × 3 multilayer silicene on Ag(111), respectively. The calculated transition barriers are indicative of the flip-flop motion among/between the tristable/bistable structures at low temperature. The flip-flop √3 structures produces the honeycomb STM motion between two of the three √3 × Crossover between Silicene and Ultra-Thin Si Atomic Layers on Ag(111) Surfaces 22 structures observed in experiments. A common feature for the multilayer Si structures on Ag surfaces is the covalent interactions at the Si-Ag interfaces and in the Si layers, which cause the absence of Dirac states in silicene. Instead, we have found a new state with the characters of π orbitals located on the top silicene layer (surface-π state), which appears near Fermi level and presents the linear energy dispersion. We expect the surface-π states have great potential applications in the spin-related nano-devices. Finally, we have explored the electronic properties of 2 × 1 π-bonded chain Si structure on Ag(111). Our results show the robust characteristic of the surface-related π states for the multilayer Si structures on Ag surfaces. Acknowledgments This work was supported by the research project "Materials Design through Computics" (http://computics-material.jp/index-e.html) by MEXT and also by "Computational Materials Science Initiative" by MEXT, Japan. Computations were performed mainly at Supercomputer Center in ISSP, University of Tokyo. ZX acknowledges the support of National Natural Science Foundation of China (Grant No. 11204259, 11374252, 11074212, 11275163 and 11304264), the Program for New Century Excellent Talents in University (Grant No. NCET-12-0722), the Program for Changjiang Scholars and Innovative Research Team in University (IRT13093), and the Scientific Research Fund of Education Department of Hunan Province (Grant No. 13B117). References [1] Slonczewski J C, Weiss P R 1958 Phys. Rev. 109 272. [2] Novoselov K S, Geim A K, Morozov S V, Jiang D, Katsnelson M L, Grigorieva I V, Dubonos S V and Firsov A A 2005 Nature 438 197. [3] Zhang Y, Tan Y W, Stormer H L and Kim P 2005 Nature 438 201. [4] Takeda K and Shiraishi K 1994 Phys. Rev. B 50 14916. [5] Cahangirov S, Topsakal M, Akturk E, S¸ahin H and Ciraci S 2009 Phys. Rev. Lett. 102 236804. [6] Liu C C, Feng W and Yao Y 2011 Phys. Rev. Lett. 107 076802. [7] Ezawa E 2012 Phys. Rev. Lett. 109 055502. [8] Ezawa E 2013 Phys. Rev. Lett. 110 026603. [9] Vogt P, De Padova P, Quaresima C, Avila J, Frantzeskakis E, Asensio M C, Resta A, Ealet B and Le Lay G 2012 Phys. Rev. Lett. 108 155501. [10] Lin C L, Arafune R, Kawahara K, Tsukahara N, Minamitani E, Kim Y, Takagi N and Kawai M. 2012 Appl. Phys. Exp. 5 045802. [11] Jamgotchian H, Colington Y, Hamzaouri N, Ealet B, Hoarau J, Aufray B and Bib´erian J P 2012 J. Phys. Cond Mat. 24 172001. [12] Feng B, Ding Z, Meng S, Yao Y, He X, Cheng P, Chen L and Wu K 2012 Nano Lett. 12 3507. [13] Guo Z X, Furuya S, Iwata J I and Oshiyama A 2013 J. Phys. Soc. Jpn. 82 063714. [14] Guo Z X, Furuya S, Iwata J I and Oshiyama A 2013 Phys. Rev. B 87 235435. [15] Chen L, Liu C C, Feng B, He X, Cheng P, Ding Z, Meng S, Yao Y and Wu K 2012 Phys. Rev. Lett. 109 056804. [16] Chen L, Li H, Feng B, Ding Z, Qiu J, Cheng P, Wu K and Meng S 2013 Phys. Rev. Lett. 110 085504. Crossover between Silicene and Ultra-Thin Si Atomic Layers on Ag(111) Surfaces 23 [17] Guo Z X and Oshiyama A 2014 Phys. Rev. B 89 155418. [18] Arafune R, Lin C L, Kawahara K, Tsukahara N, Minamitani E, Kim Y, Takagi N and Kawai M 2013 Surf. Sci. 608 297. [19] Padova P De, Vogt P, Resta A, Avila J, Razado-Colambo I, Quaresima C, Ottaviani C, Olivieri B, Bruhn T, Hirahara T, Shirai T, Hasegawa S, Asensio M C and Le Lay G 2013 Appl. Phys. Lett. 102 163106. [20] Resta A, Leoni T, Barth C, Ranguis A, Becker C, Bruhn T, Vogt P and Le Lay G 2013 Sci. Rep. 3 2399. [21] Vogt P, Capiod P, Berthe M, Resta A, De Padova P, Bruhn T, Le Lay G and Grandidier B 2014 Appl. Phys. Lett. 104 021602. [22] De Padova P, Ottaviani C, Quaresima C, Olivieri B, Imperatori P, Salomon E, Angot T, Quagliano L, Romano C, Vona A, Muniz-Miranda M, Generosi A, Paci B and Le Lay G 2014 2D Mater. 1 021003. [23] For a review, Dabrowski J and Mussig H J 2000 Silicon Surfaces and Formation of Interfaces (World Scientific, Singapole) [24] Pandey K C 1981 Phys. Rev. Lett. 47 1913. [25] Takayanagi K, Tanishiro Y, Takahashi M and Takahashi A (1985) J. Vac. Sci. Technol. B 4 1079. [26] Takayanagi K, Tanishiro Y, Takahashi M and Takahashi A (1985) Surf. Sci. 164 367. [27] Stich I, Payne M C, King-Smith R D and Lin J-S 1992 Phys. Rev. Lett. 68 1351. [28] Brommer K D, Needels M, Larson B E and Joannopoulos J D 1992 Phys. Rev. Lett. 68 1355. [29] Avila J, De Padova P, Cho S, Colambo I, Lorcy S, Quaresima C, Vogt P, Resta A, Le Lay G and Asensio M C 2013 J. Phys. Cond Matt 25 262001. [30] Wang Y P and Cheng H P 2013 Phys. Rev. B 87 245430. [31] Pflugradt P, Matthes L and Bechstedt F 2014 Phys. Rev. B 89 035403. [32] Chen M X and Weinert M 2014 Nano Lett. 14 5189 [33] Lin C L, Arafune R, Kawahara K, Kanno M, Tsukahara N, Minamitani E, Kim Y, Kawai M and Takagi N 2013 Phys. Rev. Lett. 110 076801. [34] Arafune R, Lin C L, Nagano R, Kawai M and Takagi N 2013 Phys. Rev. Lett. Comment 110 229701; Chen L, Liu C C, Feng B, He X, Cheng P, Ding Z, Meng S, Yao Y and Wu K 2013 Phys. Rev. Lett. Reply 110 229702. [35] Padova P De, Avila J, Resta A, Razado-Colambo I, Quaresima C, Ottaviani C, Olivieri B, Bruhn T, Hirahara T, Vogt P, Asensio M C and Le Lay G 2013 J. Phys.: Condens. Matter 25 382202. [36] Kresse G and Hafner J 1994 Phys. Rev. B 49 14251. [37] Kresse G and Furthmuller J 1996 Phys. Rev. B 54 11169. [38] Dion M, Rydberg H, Schroder E, Langreth D C and Lundqvist B I 2004 Phys. Rev. Lett. 92 246401. [39] Klimes J, Bowler D R and Michaelides A 2011 Phys. Rev. B 83 195131. [40] The optB86b-vdW functional in Ref.[39] is adopted in vdW-DF calculations. The calculated lattice constant for the face-centered-cubic Ag is 4.10 A, which agrees well with experimental value (4.09 A). [41] Blochl P E 1994 Phys. Rev. B 50 17953. [42] Liu ZL, Wang MX, Xu JP, Ge JF, Le Lay G, Vogt P, Qian D, Gao CL, Liu C and Jia JF 2014 New J. Phys. 16 075006. [43] STM images are simulated by calculating the local electron density of states following the Tersoff- Hamann approach: Tersoff J and Hamann D R 1985 Phys. Rev. B 31 805. [44] Le Lay G, De Padova P, Resta A, Bruhn T and Vogt P 2012 J. Phys. D: Appl. Phys. 45 392001. [45] Andrienko I and Haneman D 1999 J. Phys.: Condens. Matter 11 8437. [46] Haneman D 1987 Rep. Prog. Phys. 50 1045. [47] Haneman D 1961 Phys. Rev. 121 1093. [48] Okada S, Shiraishi K and Oshiyama A 2003 Phys. Rev. Lett. 90 026803. [49] Vanin M, Mortensen J J, Kelkkanen A K, Garcia-Lastra J M, Thygesen K S and Jacobsen K W Crossover between Silicene and Ultra-Thin Si Atomic Layers on Ag(111) Surfaces 24 2010 Phys. Rev. B 81 081408. [50] Kaltsas D, Tsetseris L and Dimoulas A 2012 J. Phys.: Condens. Matter 24 442001. [51] Scalise A, Cinquanta E, Houssa M, van den Broek B, Chiappe D, Grazianetti C, Pourtois G, Ealet B, Molle A, Fanciulli M, Afanas'ev V V and Stesmansa A 2014 Appl. Surf. Sci. 291 113. [52] Henkelman G and J´onsson H 2000 J. Chem. Phys. 113 9978. [53] Henkelman G, Uberuaga B P and J´onsson H 2000 J. Chem. Phys. 113 9901. [54] Pflugradt P, Matthes L and Bechstedt F 2014 Phys. Rev. B 89 205428. [55] Cahangirov S, Oz¸celik VO, Xian L, Avila J, Cho S, Asensio M C, Ciraci S and Rubio A 2014 Phys. Rev. B 90 035448. [56] Zitzlsperger A, Honke R, Pavone P and Schrıder U 1997 Surf. Sci. 377 108. [57] Violante C, Mosca Conte A and Pulci O 2012 J. Phys. Conf. 383 012015 

1906.02159

1

1906

Large Anomalous Hall Effect in Topological Insulators with Proximitized Ferromagnetic Insulators

We report a proximity-driven large anomalous Hall effect in all-telluride heterostructures consisting of ferromagnetic insulator Cr2Ge2Te6 and topological insulator (Bi,Sb)2Te3. Despite small magnetization in the (Bi,Sb)2Te3 layer, the anomalous Hall conductivity reaches a large value of 0.2e2/h in accord with a ferromagnetic response of the Cr2Ge2Te6. The results show that the exchange coupling between the surface state of the topological insulator and the proximitized Cr2Ge2Te6 layer is effective and strong enough to open the sizable exchange gap in the surface state.

cond-mat.mes-hall

cond-mat

Large Anomalous Hall Effect in Topological Insulators with Proximitized Ferromagnetic Insulators Masataka Mogi1,*, Taro Nakajima2, Victor Ukleev2,3, Atsushi Tsukazaki4, Ryutaro Yoshimi2, Minoru Kawamura2, Kei S. Takahashi2,5, Takayasu Hanashima6, Kazuhisa Kakurai2,6, Taka-hisa Arima2,7, Masashi Kawasaki1,2 and Yoshinori Tokura1,2,8,† 1Department of Applied Physics and Quantum Phase Electronics Center (QPEC), University of Tokyo, Bunkyo-ku, Tokyo 113-8656, Japan. 2RIKEN Center for Emergent Matter Science (CEMS), Wako, Saitama 351-0198, Japan. 3Laboratory for Neutron Scattering and Imaging (LNS), Paul Scherrer Institute (PSI), CH-5232, Villigen, Switzerland. 4Institute for Materials Research, Tohoku University, Sendai, Miyagi 980-8577, Japan. 5PRESTO, Japan Science and Technology Agency (JST), Chiyoda-ku, Tokyo 102-0075, Japan. 6Comprehensive Research Organization for Science and Society (CROSS), Tokai, Ibaraki, 319-1106, Japan. 7Department of Advanced Materials Science, University of Tokyo, Kashiwa, Chiba 277-8561, Japan. 8Tokyo College, University of Tokyo, Bunkyo-ku, Tokyo 113-8656, Japan. Abstract We report a proximity-driven large anomalous Hall effect in all-telluride heterostructures consisting of ferromagnetic insulator Cr2Ge2Te6 and topological insulator (Bi,Sb)2Te3. Despite small magnetization in the (Bi,Sb)2Te3 layer, the anomalous Hall conductivity reaches a large value of 0.2e2/h in accord with a ferromagnetic response of the Cr2Ge2Te6. The results show that the exchange coupling between the surface state of the topological insulator and the proximitized Cr2Ge2Te6 layer is effective and strong enough to open the sizable exchange gap in the surface state. 1 Main text In magnetically doped three-dimensional (3D) topological insulator (TI) films, exotic magnetic quantum phases such as a quantum anomalous Hall (QAH) insulator and an axion insulator have been achieved [1-11]. The formation of an exchange gap at the Dirac surface states of TI films and the Fermi-level tuning into the gap are two requisites for the emergence of the topological phenomena, e.g. the chiral edge conduction in the QAH state. In the magnetically doped TI, the size of the energy gap (~50 meV [12]) is produced by the interaction between magnetic impurities and the surface states, whereas it suffers from disorders due to spatial inhomogeneity of magnetic dopants [12,13] and electronic potentials [14]. In fact, the observable temperature of the QAH effect reported so far is lower than about 100 mK in the uniformly Cr- or V-doped (Bi,Sb)2Te3 films [5,6]. The modulation doping or co-doping technique of the magnetic ions has been developed to reduce the disorder, yet the observable QAH temperature still remains at most around 2 K [7,8]. The ferromagnetic proximity effect is anticipated to be an alternative ideal approach to introduce the uniform magnetic interaction to the surface states [1-3]. The choice of materials for ferromagnetic insulators (FMIs) is a key issue to induce the effective coupling with less disorder; candidates for the FMIs facing the TI is of great variety. Indeed, several FMI/TI heterostructures have been proposed theoretically and synthesized to date [15-29]. Although these studies have demonstrated several potential magnetoelectronic responses, such as magnetoresistance, anomalous Hall effect [17-24], and unconventional surface magnetization [25,26] even at room temperature, the magnitude of the response or the coupling strength to the surface state of TI remains far smaller than expected. In this Letter, we report a large anomalous Hall effect, being reminiscent of an incipient QAH state, in a FMI/TI heterostructure consisting of Cr2Ge2Te6 (CGT) [21,30-33] and (Bi,Sb)2Te3 (BST) [5-13,22,23,26]. The observation indicates the formation of a sizable exchange gap in the surface state of the TI. Through combined characterizations of the interfacial magnetic property by spin-polarized 2 neutron reflectometry and by magneto-transport measurements, we demonstrate that the exchange coupling is induced by the magnetic proximity effect. We fabricated the CGT/BST heterostructures on InP(111) substrates by molecular-beam epitaxy (MBE) (see the Supplemental Material [34] for the detailed methods). Ferromagnetic CGT thin layers have recently been achieved not only by mechanical exfoliation of bulk crystals [32] but also by thin film growth with MBE [33]. CGT has a rhombohedral crystal structure of a van der Waals (vdW) type [30], which is presumably matched to the interface formation with a similar triangular lattice constant of BST [Fig. 1(b)] [21,31]. Furthermore, it has been reported that the MBE-grown CGT films possess a perpendicular anisotropic remanent magnetization [33,34], which is favorable to produce the exchange gap in the surface state of TI. Experimentally, the structural characterization of the interface was carried out by a cross-sectional scanning transmission electron microscopy (STEM). Figure 1(c) displays the STEM image of a MBE-grown CGT(12 nm)/BST(9 nm)/CGT(12 nm) heterostructure, exhibiting abrupt interfaces with the ordered stacking orientation in favor of the hexagonal Te arrangements of CGT and BST. By performing Fourier transformation in the lateral direction of the image, the lattice distance of each layer is achieved as depicted in the right panel of Fig. 1(c). Sharp changes of the lateral lattice distance at the interfaces reflect weak epitaxial strain at the interfaces which are a notable feature of the vdW heterointerface. Furthermore, energy dispersive x-ray spectroscopy (EDX) ensures almost no inter-diffusion of atoms [34]. On the basis of the MBE-grown clean heterostructures, we examine the interfacial magnetism of the CGT/BST/CGT sandwiched heterostructure by depth-sensitive polarized neutron reflectometry (PNR) [34]. The PNR measurements, being quantitatively responsive to in-plane magnetization, were conducted at 3 K with an in-plane magnetic field 0H = 1 T [Fig. 2(a)] which is strong enough to fully align the magnetic moments to the field direction as confirmed by the magnetization hysteresis loops measured at 2 K [Fig. 2(b)]. Figure 2(c) shows the x-ray and non-spin-flip specular PNR reflectivity curves R+ and R−, where + (−) denotes the incident neutron spins parallel (antiparallel) to the direction 3 of H as a function of the momentum transfer vector Qz. The in-plane saturated magnetization of the sample is directly reflected in the spin-asymmetry ratio defined as (R+−R−)/(R++R−) [Fig. 2(d)]. In addition, we combined an x-ray reflectivity (XRR) measurement at room temperature [Fig. 2(c)] to conduct a model analysis for the structural parameters including thickness and roughness of each layer. The depth profile of the x-ray scattering length density (SLD) shown in Fig. 2(e), corresponding to the electron density distribution in the heterostructure, reflects the structural interface roughness. Notably, the root mean square roughness of all interfaces in the SLD profiles is less than 1 nm, which is consistent with the STEM image shown in Fig. 1(c). The structural parameters derived from the XRR fitted model were used to refine the PNR curves. Figure 2e displays the magnetic SLD depth-profiles based on the fitting results on the R+, R− [Fig. 2(c)] and the spin-asymmetry ratio [Fig. 2(d)], taking into account the structural depth profile obtained from the XRR data. The fitting analysis yields magnetizations of 152±8 emu/cm3 and 0±20 emu/cm3 for the CGT and BST layers [34], respectively. Although it is difficult to precisely determine the induced magnetization in the BST layer due to the spatial broadening, it will be reasonable to conclude from the present fitting analysis that the induced magnetization in the BST layer is far smaller than the magnetization of the CGT layer. The ferromagnetic proximity effect on the surface states of the TI can be assessed by magneto- transport measurements. The measurements were conducted with the sandwiched CGT/BST/CGT trilayers and the CGT/BST bilayers. Because of the high electric resistance of the CGT layer [33], its contribution to electrical transport is negligibly small [34]. For the TI layer, instead of simple single- layered (Bi1−xSbx)2Te3, we engineered a multi-layer structure of (Bi1−xSbx)2Te3(2 nm)/Bi2Te3(2 nm)/(Bi1−xSbx)2Te3(2 nm) [Fig. 3(a)] which works as a conduction channel. The reason for adopting the multi-layer structure is as follows. In the CGT/BST/CGT heterostructures, the charge neutrality point takes place at a relatively small value of x (0.3 < x < 0.4) due to possible hole transfer from CGT to BST [34]. According to an ARPES study on BST [41], small x causes the Dirac point to submerge below the bulk valence band. To approach the Dirac point with the charge neutrality condition, we 4 need to dope electrons while keeping x > 0.5. To fulfill this requirement, we inserted the electron-rich Bi2Te3 layer between the BST layers to assist electron doping. The value of x in the BST layer is kept larger than 0.5 assuming that the surface band structure is mainly affected by the environment near the interface [36]. Consequently, we could prepare the samples with low carrier density at reasonably large Sb compositions, x = 0.6 and 0.64, which show semiconducting temperature (T) dependence of the longitudinal sheet resistivity (xx) as depicted in Fig 3(b). In these samples, large anomalous Hall resistance (> 1 k) appear with perpendicular anisotropic hysteresis loops as shown in Fig. 3(c). We show in Fig. 3(d) the x dependence of the sheet carrier density (n2D), the longitudinal sheet conductivity (xx) and the Hall conductivity (xy) at zero fields as converted from xx and yx. The notable feature is that the xy exceeds 0.2e2/h in the most insulating sample (x = 0.6) where xx ~ 2e2/h and n2D ~ 1012 cm-2 [46]. The sheet carrier densities are estimated from the slope of Hall resistance above the saturation field. The carrier types are electrons for x = 0.3 and holes for x = 0.64, demonstrating that the Fermi level is systematically shifted from n- type to p-type with increasing x [Fig. 3(c)]. These observations in Fig. 3 can be understood by the opening of an exchange gap in the dispersion relation of the TI surface state. When an exchange gap opens on the surface of TI, the Berry curvature is strongly enhanced near the band edge resulting in the large xy. At the same time, when the Fermi energy is tuned within or close to the exchange gap, xy takes a maximum while xx takes a minimum. In the present study, we observe that xy takes a maximum accompanied by a nearly minimum value of xx in the sample with the low carrier density (x = 0.6) as expected. Also, the decrease in xy and increase in xx are observed as the carrier density is detuned from the optimum value. These carrier density dependencies are consistent with the picture described above, indicating the opening of the exchange gap by the magnetic proximity effect. The increase in xy accompanied by the decrease in xx leads to an enhancement in the Hall angle H = tan-1(xy/xx), discriminating the anomalous Hall effect of extrinsic origin [45]. The obtained values of the xy and the H are 5 dramatically increased in our samples compared to those reported in other FMI/TI systems [Fig. 3(e)] [34]. This trend suggests that the Fermi level is close to the exchange gap and/or that the exchange gap is large in our samples compared among those FMI/TI systems, although the quantitative estimation of the size of the exchange gap is difficult due to residual disorder/inhomogeneity in the samples [43,44] (see [34] for the detailed discussion). The CGT-layer thickness dependence provides additional evidence that the observed anomalous Hall effect is induced by the magnetic proximity effect, excluding other origins arising from Cr diffusion into the BST layer. Figure 4(a) shows the temperature dependent magnetization (M- T) curves of four CGT/BST bilayers [the inset of Fig. 4(c)] films with representative CGT-layer thicknesses under 0H⟂ = 0.05 T. Both magnetization M and Curie temperature TC decrease systematically with decreasing the CGT film thickness t. As shown in Fig. 4(b), the low-temperature values of xy also decrease with decreasing t. In Fig. 4(c), the t dependencies of xy at T = 2 K and the saturated magnetization of CGT Ms are plotted together. The agreement in t-dependencies of xy and Ms indicates that xy is almost proportional to Ms. This observation shows that the exchange gap on the surface state of the TI can be tuned by the magnetization of the CGT layer, directly pointing to the proximity-coupling origin of the anomalous Hall effect. The decrease in Ms in the range of t < 2 nm is attributed perhaps to the finite size effect of the 2D ferromagnetic CGT layer [31]. In contrast to xy, xx is almost constant with variation of t across t ~ 2 nm [Fig. 4 (c)]. The constancy of xx suggests that the Fermi energy and the scattering time are not largely affected by the thickness of the CGT layer. On the basis of the above experimental results, we discuss possible mechanisms of the exchange gap formation at the CGT/BST interface. One conceivable scenario would be the induction of magnetization in the TI layer by the adjacent FMI layer as discussed in the earlier works [15- 17,25,26,28]. However, this scenario is unlikely applicable to the present case. At the interfaces of EuS/Bi2Se3 and EuS/BST, large magnetizations of about 270 and 160 emu/cm3, respectively, in the TI layer have been reported [25,26]. In contrast, in the present study, the magnetization in the CGT layer 6 is already smaller than these values. Therefore, the induced magnetization, even if it existed, in the BST layer of the present CGT/BST heterostructure would be much smaller than that reported for the EuS-based heterostructures [25,26]. Despite the small induced magnetization, our transport measurements have revealed that the xy and H are much enhanced in the CGT/BST system. One other possible scenario to understand these observations is the formation of the exchange gap by penetration of the TI surface state wave function into the FMI layer. In this scenario, even if the interfacial magnetization in the BST layer is small, the penetrated part of the surface state wave function can interact with the magnetic moment in the CGT to produce a sizable exchange gap. A recent first-principle calculation work [29] indicates the formation of a large exchange gap in Te-based heterostructure MnBi2Te4/Bi2Te3 based on the wave function penetration mechanism. In summary, we have synthesized CGT/BST/CGT heterostructures and have studied the ferromagnetic proximity effect at the interface of the heterostructures. We have observed the depth profile of the magnetization by the PNR measurement which suggests small induced magnetization in the BST layer. We have also observed a large anomalous Hall angle in magneto-transport measurements, which indicates a sizable exchange gap. To explain both observations, we have proposed the exchange gap formation due to the penetration of the TI surface state wave function into the FMI layer. We acknowledge helpful discussions with K. Yasuda, R. Watanabe, R. Fujimura, Y. Fujishiro, Y. Okamura, Y. Kaneko, S. Maekawa, J. G. Checkelsky and N. Nagaosa. This research was partly supported by JSPS/MEXT Grant-in-Aid for Scientific Research (No. 15H05853, 15H05867, 17J03179, 18H04229, 18H01155), and JST CREST (No. JPMJCR16F1). This neutron experiment at MLF, J-PARC was performed under a user program (Proposal No.2017B0195). *mogi@cmr.t.u-tokyo.ac.jp †tokura@riken.jp [1] M. Z. Hasan, and C. L. Kane, Rev. Mod. Phys. 82, 3045 (2010). 7 [2] X. L. Qi, and S.-C. Zhang, Rev. Mod. Phys. 83, 1057 (2011). [3] X.-L. Qi, T. L. Hughes, and S.-C. Zhang, Phys. Rev. B 78, 195424 (2008). [4] R. Yu, W. Zhang, H. J. Zhang, S.-C. Zhang, X. Dai, and Z. Fang, Science 329, 61 (2010). [5] C.-Z. Chang, J. Zhang, X. Feng, J. Shen, Z. Zhang, M. Guo, K. Li, Y. Ou, P. Wei, L.-L. Wang, Z.- Q. Ji, Y. Feng, S. Ji, X. Chen, J. Jia, X. Dai, Z. Fang, S.-C. Zhang, K. He, Y. Wang, L. Lu, X.-C. Ma, and Q.-K. Xue, Science 340, 167 (2013). [6] C.-Z. Chang, W. W. Zhao, D. Y. Kim, H. J. Zhang, B. A. Assaf, D. Heiman, S.-C. Zhang, C. X. Liu, M. H. W. Chan, and J. S. Moodera, Nat. Mater. 14, 473 (2015). [7] M. Mogi, R. Yoshimi, A. Tsukazaki, K. Yasuda, Y. Kozuka, K. S. Takahashi, M. Kawasaki, and Y. Tokura, Appl. Phys. Lett. 107, 182401 (2015). [8] Y. Ou, C. Liu, G. Jiang, Y. Feng, D. Zhao, W. Wu, X.-X. Wang, W. Li, C. Song, L.-L. Wang, W. Wang, W. Wu, Y. Wang, K. He, X.-C. Ma, and Q.-K. Xue, Adv. Mater. 30, 1703062 (2018). [9] M. Mogi, M. Kawamura, R. Yoshimi, A. Tsukazaki, Y. Kozuka, N. Shirakawa, K. S. Takahashi, M. Kawasaki, and Y. Tokura, Nat. Mater. 16, 516 (2017). [10] M. Mogi, M. Kawamura, A. Tsukazaki, R. Yoshimi, K. S. Takahashi, M. Kawasaki, and Y. Tokura, Sci. Adv. 3, eaao1669 (2017). [11] D. Xiao, J. Jiang, J.-H. Shin, W. Wang, F. Wang, Y.-F. Zhao, C. Liu, W. Wu, M. H. W. Chan, N. Samarth, and C.-Z. Chang, Phys. Rev. Lett. 120, 056801 (2018). [12] I. Lee, C. K. Kim, J. Lee, S. J. L. Billinge, R. Zhong, J. A. Schneeloch, T. Liu, T. Valla, J. M. Tranquada, G. Gu, and J. C. Séamus Davis, Proc. Natl. Acad. Sci. USA 112, 1316 (2015). [13] E. O. Lachman, A. F. Young, A. Richardella, J. Cuppens, H. R. Naren, Y. Anahory, A. Y. Meltzer, A. Kandala, S. Kempinger, Y. Myasoedov, M. E. Huber, N. Samarth, and E. Zeldov, Sci. Adv. 1, e1500740 (2015). [14] H. Beidenkopf, P. Roushan, J. Seo, L. Gorman, I. Drozdov, Y. S. Hor, R. J. Cava, and A. Yazdani, Nat. Phys. 7, 939 (2011). [15] W. Luo, and X.-L. Qi, Phys. Rev. B 87, 085431 (2013). [16] S. V. Eremeev, V. N. Men'shov, V. V. Tugushev, P. M. Echenique, and E. V. Chulkov, Phys. Rev. B 88, 144430 (2013). [17] P. Wei, F, Katmis, B. A. Assaf, H. Steinberg, P. Jarillo-Herrero, D. Heiman, and J. S. Moodera, Phys. Rev. Lett. 110, 186807 (2013). [18] Q. I. Yang, M. Dolev, L. Zhang, J. Zhao, A. D. Fried, E. Schemm, M. Liu, A. Palevski, A. F. Marshall, S. H. Risbud, and A. Kapitulnik, Phys. Rev. B 88, 081407(R) (2013). [19] A. Kandala, A. Richardella, D. W. Rench, D. M. Zhang, T. C. Flanagan, and N. Samarth, Appl. Phys. Lett. 103, 202409 (2013). [20] M. Lang, M. Montazeri, M. C. Onbasli, X. Kou, Y. Fan, P. Upadhyaya, K. Yao, F. Liu, Y. Jiang, W. Jiang, K. L. Wong, G. Yu, J. Tang, T. Nie, L. He, R. N. Schwartz, Y. Wang, C. A. Ross, and K. L. Wang, Nano Lett. 14, 3459 (2014). 8 [21] L. D. Alegria, H. Ji, N. Yao, J. J. Clarke, R. J. Cava, and J. R. Petta, Appl. Phys. Lett. 105, 053512 (2014). [22] Z. Jiang, C.-Z. Chang, C. Tang, P. Wei, J. S. Moodera, and J. Shi, Nano Lett. 15, 5835 (2015). [23] C. Tang, C.-Z. Chang, G. Zhao, Y. Liu, Z. Jiang, C.-X. Liu, M. R. McCartney, D. J. Smith, T. Chen, J. S. Moodera, and J. Shi, Sci. Adv. 3, e1700307 (2017). [24] S.Zhu, D. Meng, G. Liang, G. Shi, P. Zhao, P. Cheng, Y. Li, X. Zhai, Y. Lu, L. Chen, and K. Wu, Nanoscale 10, 10041 (2018). [25] F. Katmis, V. Lauter, F. S. Nogueira, B. A. Assaf, M. E. Jamer, P. Wei, B. Satpati, J. W. Freeland, I. Eremin, D. Heiman, P. Jarillo-Herrero, and J. S. Moodera, Nature 533, 513 (2016). [26] M. Li, Q. Song, W. Zhao, J. A. Garlow, T.-H. Liu, L. Wu, Y. Zhu, J. S. Moodera, M. H. W. Chan, G. Chen, and C.-Z. Chang, Phys. Rev. B 96, 201301(R) (2017). [27] T. Hirahara et al., Nano Lett. 17, 3493 (2017). [28] V. N. Men'shov, V. V. Tugushev, S. V. Eremeev, P. M. Echenique, and E. V. Chulkov, Phys. Rev. B 88, 224401 (2013). [29] M. M. Otrokov, T. V. Menshchikova, M. G. Vergniory, I. P. Rusinov, A. Y. Vyazovskaya, Y. M. Koroteev, G. Bihlmayer, A. Ernst, P. M. Echenique, A. Arnau, E. V. Chulkov, 2D Mater. 4, 025082 (2017). [30] V. Carteaux, D. Brunet, G. Ouvrard, and G. André, J. Phys.: Condens. Matter 7, 69 (1995). [31] H. Ji, R. A. Stokes, L. D. Alegria, E. C. Blomberg, M. A. Tanatar, A. Reijnders, L. M. Schoop, T. Liang, R. Prozorov, K. S. Burch, N. P. Ong, J. R. Petta, and R. J. Cava, J. Appl. Phys. 114, 114907 (2013). [32] C. Gong, L. Li, Z. Li, H. Ji, A. Stern, Y. Xia, T. Cao, W. Bao, C. Wang, Y. Wang, Z. Q. Qiu, R. J. Cava, S. G. Louie, J. Xia, and X. Zhang, Nature 546, 265 (2017). [33] M. Mogi, A. Tsukazaki, Y. Kaneko, R. Yoshimi, K. S. Takahashi, M. Kawasaki, and Y. Tokura, APL Mater. 6, 091104 (2018). [34] See Supplemental Material, which includes Refs. [12, 14, 17, 21-25, 33, 35-45]. [35] R. Yoshimi, A. Tsukazaki, Y. Kozuka, J. Falson, K. S. Takahashi, J. G. Checkelsky, N. Nagaosa, M. Kawasaki, and Y. Tokura, Nat. Commun. 6, 6627 (2015). [36] C.-Z. Chang, P. Tang, X. Feng, K. Li, X.-C. Ma, W. Duan, K. He, and Q.-K. Xue, Phys. Rev. Lett. 115, 136801 (2015). [37] M. Takeda et al., Chin. J. Phys. 50, 161 (2012). [38] K. Nakajima et al. Quantum Beam Sci. 1, 9 (2017). [39] Y. Inamura, T. Nakatani, J. Suzuki, and T. Otomo, J. Phys. Soc. Jpn. 82, SA031(2013). [40] M. Björck, and G. Andersson, J. Appl. Crystallogr. 40, 1174 (2007). [41] J. Zhang, C.-Z. Chang, Z. Zhang, J. Wen, X. Feng, K. Li, M. Liu, K. He, L. Wang, X. Chen, Q.- K. Xue, X. Ma, and Y. Wang, Nat. Commun. 2, 574 (2011). [42] R. Dingle, H. L. Störmer, A. C. Gossard, and W. Wiegmann, Appl. Phys. Lett. 33, 665 (1978). [43] B. Skinner, T. Chen, and B. I. Shklovskii, Phys. Rev. Lett. 109, 176801 (2012). 9 [44] N. A. Sinitsyn, J. E. Hill, H. Min, J. Sinova, and A. H. MacDonald, Phys. Rev. Lett. 97, 106804 (2006). [45] N. Nagaosa, J. Sinova, S. Onoda, A. H. MacDonald, and N. P. Ong, Rev. Mod. Phys. 82, 1539 (2010). [46] The n2D at x = 0.6 is slightly higher than that at x = 0.5. We speculate that this can be related to the coexistence of electrons and holes near the charge neutrality condition. 10  Figures FIG. 1. (a) Schematic drawing of the interfacial exchange coupling in a 3D topological insulator (TI) sandwiched by ferromagnetic insulators (FMIs). (b) Schematic of crystal structures of TI (Bi,Sb)2Te3 (BST) and FMI Cr2Ge2Te6 (CGT) with the relationship of stacking orientation expected from their Te arrangements in the respective layer planes. (c) Cross-sectional high-angle annular dark-field STEM image of the CGT(12 nm)/BST(9 nm)/CGT(12 nm) heterostructure grown on an InP substrate with a BST buffer layer (left panel). The lateral atomic distance of each layer obtained by the Fourier transformation of the left image plotted along the growth direction (right panel). 11  FIG. 2. (a) Schematic of the PNR experimental set-up for the CGT/BST/CGT structure. (b) Magnetization hysteresis loops under out-of-plane (H⟂) and in-plane (H) magnetic fields for the identical CGT(12 nm)/BST(9 nm)/CGT(12 nm) sample used in the PNR experiments. The black arrow represents the saturation field for in-plane direction. (c) Measured (dots) and fitted (solid lines) reflectivity curves for the x-ray (black), neutron of spin-up (R+) (red) and spin-down (R−) (blue) as a 12 function of momentum transfer (Qz) on a logarithmic scale. The error bars represent one standard deviation. (d) PNR spin-asymmetry ratio (R+ − R−)/(R+ + R−) obtained from experimental and fitted reflectivity curves in (c). The error bars represent one standard deviation. (e) X-ray scattering length density (SLD) (black), and neutron SLD divided into the nuclear (blue) and the magnetic (red) SLDs as a function of the distance from the InP substrate surface (z). The re in the unit of the X-ray SLD denotes the classical electron radius of 2.8179...×10-15 m. For the magnetic SLD, the value of M corresponding to the neutron SLD is shown in the right ordinate. 13  FIG. 3. (a) Schematic layout of CGT(12 nm)/(Bi1-xSbx)2Te3(2 nm)/Bi2Te3(2 nm)/(Bi1-xSbx)2Te3(2 nm)/CGT(12 nm) heterostructure. (b, c), Temperature (T) (out-of-plane magnetic field (0H⟂)) dependence of the longitudinal sheet resistivity (xx) in zero-field (b) (the Hall resistivity (yx) at 2 K (c)) of CGT/(Bi1-xSbx)2Te3/Bi2Te3/(Bi1-xSbx)2Te3/CGT (x = 0.3, 0.6, 0.64) heterostructures. (d) Sb fraction (x) dependence of the sheet carrier density (n2D) (top panel), the longitudinal sheet conductivity (xx) (middle panel), and the Hall conductivity (xy) (bottom panel) at 2 K. The insets are the simplified schematics of band structures representing the different Fermi energies; the blue lines represent the dispersion of the surface state. (e) The anomalous Hall conductivity xy A is plotted against 14 xx plot with the use of the data for the present heterostructures shown in (d) in comparison with other various FMI/TI heterostructures [17, 21-25]. The values of xy A and xx are taken from the data obtained at the lowest temperature in the measurements (2-5 K) for the magnetized state where the out-of-plane magnetization saturates. 15  FIG. 4. (a, b) M-T (a) and xy-T (b) curves measured under field cooling with 0H⟂ = 0.05 T in BST/CGT (t = 1.4, 2.2, 2.9 and 8.1 nm) bilayer structures. Black arrows indicate the Curie temperature TC for highlighting the changes against t. (c) CGT thickness t dependence of xy (red closed circles, left red ordinate) and the spontaneous magnetization Ms (black closed squares, right black ordinate) at 2 K under zero magnetic field (top panel), and xx at 2 K (blued closed circles, left blue ordinate) and TC (black closed squares, right black ordinate) (bottom panel). The inset shows the schematic layout of (Bi0.5Sb0.5)2Te3 (6 nm)/CGT (t nm) bilayer structure. 16 Supplementary Material for Large Anomalous Hall Effect in Topological Insulators with Proximitized Ferromagnetic Insulators M. Mogi, T. Nakajima, V. Ukleev, A. Tsukazaki, R. Yoshimi, M. Kawamura, K. S. Takahashi, T. Hanashima, K. Kakurai, T. Arima, M. Kawasaki and Y. Tokura I. Methods. Sample fabrication. The CGT/BST heterostructures were grown on semi-insulating (> 107  cm) InP(111)A substrates using standard Knudsen cells in a MBE chamber under a vacuum condition (~ 1 × 10-7 Pa). The respective layers were grown in the same procedures as described in Refs. [S1,S2]. Before the growth of the CGT, 1-2 quintuple layers of BST were grown as an insulating buffer layer to improve the crystallinity of the CGT. For the Fermi-level tuning in the trilayers depicted in Fig. 3, we modulated Bi/Sb ratio by open and close of the shutter for the Sb cell and succeeded in the tuning as investigated by Chang et al. [S3]. Taking the films out of the MBE chamber, the AlOx capping layer (~5 nm) was immediately deposited by an atomic layer deposition (ALD) system at room temperature to prevent deterioration of the films. Thicknesses of respective layers were determined by XRR measurements. For transport measurements, the films were patterned into Hall bars with 200-300 m in width and 700-1,000 m in length by using photolithography and chemical wet etching by H2O2/H3PO4/H2O (1:1:8) and HCl-H2O (1:4) mixtures. Electrical contact was made of Ti(3 nm)/Au(27 nm) by electron beam evaporation. Polarized neutron and x-ray reflectometry. PNR experiments were performed at BL-17 SHARAKU of the Materials and Life Science Experimental Facility (MLF) in the Japan Proton Accelerator Research Complex (J-PARC), Tokai, 17 Japan [S4,S5]. The sample was loaded into a closed cycle refrigerator and was cooled to 3 K. An electromagnet was used to apply a magnetic field of 1 T along the in-plane direction of the film, which was perpendicular to the incident neutron beam and neutron momentum transfer, Qz. The PNR spectra were measured by means of the time-of-flight technique with a pulsed polychromatic incident neutron beam; the wavelength range was from 2.4 to 8.8 Å. We selected three different incident angles (0.3, 0.9 and 2.7 degrees) to provide access to the momentum transfer range from 0.08 to 2 nm-1. A supermirror polarizer, guide-field coils, and a spin flipper were employed to obtain the polarized incident neutron beam, whose polarization direction was set to be parallel or antiparallel to the external magnetic field at the sample position. The beam polarization was approximately 98 %. The intensities of reflected neutrons were measured without analyzing the spin state of the neutrons and converted to the PNR spectra by the UTSUSEMI software [S6], in which effects of the imperfect beam polarization were corrected. The PNR spectra were fitted using GenX software [S7], assuming that the magnetic moments of Cr were parallel to the magnetic field. Complementary XRR spectrum was measured by Bruker D8 x-ray diffractometer at room temperature to determine layer thicknesses of the sample. The sample used in the XRR measurement is identical to that used in the PNR measurements. An incident x-ray beam with a wavelength of 1.5406 Å was obtained by Cu Kα radiation. The intensity of specular reflection was measured with varying the incident angle from 0.4 to 4.3 deg to cover Qz range from 0.3 to 3 nm-1. Transport and magnetization measurements. Electrical transport and magnetization measurements were carried out using a Quantum Design physical property measurement system (PPMS) and a magnetic property measurement system (MPMS) superconducting quantum interference device (SQUID) magnetometer, respectively, in the temperature range from 2 to 300 K. xy and xx in the four terminal measurements were converted from yx and xx following the tensor relations xy = yx/(xx 2+yx 2) and xx = xx/(xx 2+yx 2). 18 II. Characterization and fundamental properties of CGT/BST/CGT heterostructures. FIG. S1. Characterization of the CGT/BST/CGT heterostructure prepared for STEM and PNR. (a) Cross-sectional schematic of the CGT(12 nm)/BST(9 nm)/CGT(12 nm) on a 1-nm-thick BST buffered InP substrate, where the Sb fraction of BST is x = 0.3, used for STEM/EDX and PNR studies in the main text. (b) XRD pattern on a logarithmic scale for the CGT/BST/CGT structure shown in (a). (c-e) Temperature (T) dependence of the longitudinal sheet resistivity (xx) (c), the Hall resistivity (yx) (d) and the magnetization (M) (e) of the CGT/BST/CGT structure shown in (a). In (c), xx of a CGT single-layer (t = 12 nm) grown on a 1-nm-thick BST buffered InP substrate and a BST single-layer (t = 9 nm) directly grown on an InP substrate are also shown. Black arrows indicate the rising temperature of yx and M. (f) Perpendicular magnetic field (µ0H ⊥) dependence of the magnetization (M) at various temperatures. (g) Arrott plot to determine the Curie temperature (TC ~ 80 K). Figure S1 shows the x-ray diffraction (XRD) characterization and fundamental physical properties of the CGT/BST/CGT sandwiched heterostructure [Fig. S1(a)], which was characterized by STEM and PNR in the main text. Figure S1(b) shows the XRD pattern for the heterostructure. We observe diffraction peaks from both BST and CGT layers as expected. Figure S1(c) shows the electrical resistivity (xx) of the heterostructure. The xx of a CGT single-layer (t = 12 nm) is more than two 19 orders of magnitude larger than that of a BST single-layer (t = 9 nm), ensuring the least contribution of parasitic conduction in the CGT layer. In fact, xx of CGT/BST/CGT structure is comparable to that of a 9-nm-thick BST single-layer. Figure S1(d) shows the temperature (T) dependence of the Hall resistivity (yx). With decreasing T, yx and M rise at around T = 80 K as highlighted by black arrows in Fig. S1(e). In Figs. S1(f) and (g), we show the magnetic hysteresis loops at various temperatures and the Arrott plot, respectively. We find the Curie temperature TC ~ 80 K in accord with the M-T curve [Fig. S1 (e)]. It should be noted that the TC is slightly higher than that of the bulk crystals. In fact, TC of the MBE-grown CGT films alone also shows a similar enhanced TC possibly due to some defects in the films [S1]. FIG. S2. Transport properties of CGT/BST/CGT sandwiched heterostructures. (a) Cross- sectional schematic of the CGT(12 nm)/BST(12 nm)/CGT(12 nm) on a 1-nm-thick BST buffered InP substrate (b) Out-of-plane magnetic field (B) dependence of the Hall resistivity (yx) at T = 2 K for the CGT/(Bi1-xSbx)2Te3/CGT (x = 0.3, 0.32, 0.36) heterostructures. (c) Sb fraction (x) dependence of the sheet carrier density (n2D) (top panel), the longitudinal sheet conductivity (xx) (middle panel), and the Hall conductivity (xy) (bottom panel) in CGT/BST/CGT and BST single-layer films. The data for the BST single-layer films is an excerpt from [S8]. (d) Schematics of the relationship between the Fermi-level (EF) and the TI band structure for the respective films which are shown in (b). 20  For transport measurements described in the main text (Fig. 3), we used the (Bi1- xSbx)2Te3/Bi2Te3/(Bi1-xSbx)2Te3 heterostructures sandwiched by CGT layers to tune the Fermi-level close to the Dirac point. To explain the effectiveness of this structure, we show measurements for (Bi1- xSbx)2Te3 single-layers sandwiched by CGT [Fig. S2(a)]. The Hall responses are shown in Fig. S2(b) for representative three samples (x = 0.3, 0.32, 0.36). One important result here is the discrepancy between the charge neutrality point (between x = 0.32 to 0.36) and the maximal anomalous Hall resistivity composition (x < 0.32) as summarized by the Sb fraction (x) dependence in Fig. S2(c). We assign the discrepancy to the hole conduction in the bulk region as illustrated in Fig. S2(d), where the Dirac point buries in the bulk valence band as reported in the angle-resolved photoemission study for Bi-rich BST (x ~ 0.3) [S8]. In such a situation, even if the Fermi-level is tuned near the Dirac point, parasitic conduction in the bulk valence band appears. Furthermore, the value of x ~ 0.32, corresponding to the Fermi-level roughly tuned at the charge neutrality point in the CGT/BST/CGT structures, is much lower than that for pristine BST (x ~ 0.8) films [S8]. We preliminarily attribute the shift of the charge neutrality point to a charge transfer between p-type CGT and n-type BST layers. When holes are transferred to the BST layer, it is necessary to compensate the carriers by introducing electrons with lowering x. In fact, the MBE-grown CGT films show the weak p-type conduction possibly because of Ge deficiency [S1], which can generate holes. To overcome this situation, we applied the (Bi1-xSbx)2Te3/Bi2Te3/(Bi1-xSbx)2Te3 heterostructures to suppress the bulk hole conduction with the electron-rich Bi2Te3 thin layer at the inner region. Such a modulation doping technique is adopted for conventional semiconductor heterostructures [S9]. In addition, the proximitized region should be carefully designed to maximize the Berry curvature at the Dirac point. We insert the (Bi1-xSbx)2Te3 layer for the Fermi-level tuning of the isolated Dirac point in the bulk band gap because the surface band structure is strongly affected by the environment near the surface [S3]. 21 III. Cross-sectional elemental distribution in the CGT/BST/CGT heterostructure. FIG. S3. STEM-EDX for the CGT/BST/CGT heterostructure. (a) STEM image of CGT/BST/BST corresponding to the EDX scan area. (b-e) Distribution maps of each element, Cr (b), Ge (c), Bi (d), and Te (e). (f) Line profiles of Cr, Ge, Bi, and Te. Figure S3 shows energy dispersive x-ray spectroscopy (EDX) mappings of Cr (b), Ge (c), Bi (d), and Te (e) for the CGT(12 nm)/BST(9 nm)/CGT(12 nm) grown on a 1-2 nm BST buffered InP substrate, which is an identical sample measured by STEM in the main text [Fig. 1(c) and Fig. S3(a)]. Uniform chemical composition is detected as expected for the constituted layer structure. The line profiles in Fig. S3(f) indicate that no discernible inter-diffusion of magnetic Cr atoms is detected. 22 IV. Scheme of fitting and alternative models for the PNR reflectivity data. Firstly, we explain the scheme of fitting the data shown in Fig. 2 of the main text, in which we show the best-fitted data, indicating that the magnetization resides almost only in the CGT layers. The refinement of the PNR and the XRR fittings was performed using GenX software [S7] for a multilayered model consisting of InP substrate/BST buffer layer/CGT/2 quintuple layer (QL) interfacial BST/BST/2-QL interfacial BST/CGT/AlOx capping layer. Fitting parameters were layer thicknesses, densities, and interfacial roughness. Furthermore, we introduced magnetization in the CGT and the 2-QL interfacial BST layers as fitting parameters. We note that we use models containing different thicknesses of the AlOx capping layer for the convergence of XRR and PNR simulations possibly due to a little deterioration of the capping layer in the interval of their experiments. The fitting curves were evaluated by the logarithmic figure of merit taking errors into account (FOMlogbar) as below, FOMlogbar = 1 𝑁 − 𝑝 ∑ 𝑖 log10𝑌𝑖 − log10𝑆𝑖 𝐸𝑖log10𝑌𝑖 , where N is the total number of data consisting of XRR and PNR reflectivity data, p is the number of fitting parameters, Yi and Si represent the experimental and simulated dataset, respectively, Ei is the uncertainty in the experimental data, and i indicates individual elements of the dataset. By minimizing FOMlogbar (= 0.99), we obtain the magnetization of 152 emu/cm3 for the CGT layer and of -- 0.4 emu/cm3 for the interfacial BST layer. 23 FIG. S4. Fitting curves for additional models. (a) Magnetization depth profiles, in which the interfacial roughness is not included for the clarity, for simulation of CGT/BST/BST structural models with interfacial magnetization in 2 quintuple layers (QL) in BST (0, 1, 2 B per formula unit of BST). (b) Simulations of PNR spin-asymmetry ratio (solid lines) with experimental data (open dots). The double-headed arrow represents the most statistically reliable region of Qz. We further investigate the magnetization at the interface (Fig. S4). We prepare models in which we intentionally introduce the magnetization (1 and 2 B/f.u. for BST) at the 2-QL interfacial BST region as shown in Fig. S4(a), following the PNR study on EuS/Bi2Se3 films [S10]. Figure S4(b) displays the simulation data corresponding to the models in Fig. S4(a). The data are well fitted by the model in the range of 0.1 < Qz < 0.6 although a clear difference is difficult to see among the results with the fitting parameters M = 0, 1, and 2 B/f.u. assumed for the interfacial 2-QL BST. 24 V. Additional comparison of anomalous Hall effect with various FMI/TI structures. FIG. S5. Tangent anomalous Hall angle (xy A /xx) versus temperature (T) of CGT/BST and various FMI/TI heterostructures. In addition to the xx-xy A plane plot shown in Fig. 3(e) of the main text, we plot the tangent Hall angle (xy A/xx) for the CGT/BST heterostructures compared with previously reported various FMI/TI heterostructures (Fig. S5). The tangent Hall angle is two or three orders of magnitude larger than those of other heterostructures including CGT/chemical vapor deposition (CVD)-grown Bi2Te3 (2.5 K) [S11], EuS/Bi2Se3 (< 5 K) [S10,S12] LaCoO3/Bi2Se3 (1.7-100 K) [S13], Y3Fe5O12/BST (1.9-200 K) [S14] and Tm3Fe5O12/BST (2-400 K) [S15]. We note that the present MBE-grown CGT/BST heterostructure was greatly improved as compared with the previously studied CGT/Bi2Te3 heterostructure grown by chemical vapor deposition (CVD) [S11], possibly because a thick (30 nm) Bi2Te3 layer with a high bulk carrier density was used in the previous study, on which the ferromagnetic proximity effect would be masked. 25 VI. Disorder/inhomogeneity in samples and estimation of the size of the exchange gap. We discuss an attempt to estimate the size of the exchange gap from the transport results on the basis of a simple Dirac fermions model, although the estimation is impeded by the presence of disorder/inhomogeneity in samples as below. We firstly describe the influence of disorder/inhomogeneity in our samples from the temperature-dependent transport data. (i) Influence of disorder/inhomogeneity in the samples FIG. S6. Temperature-dependent transport properties for CGT(12 nm)/(Bi1-xSbx)2Te3(2 nm)/Bi2Te3(2 nm)/(Bi1-xSbx)2Te3(2 nm)/CGT(12 nm) heterostructures. (a) Temperature (T) dependence of the longitudinal resistivity (xx) on a logarithmic scale from 300 K to 2 K with 26 variation of x. (b) xx as functions of T-1. The black broken line is the Arrhenius fitting result [xx  exp(T0/T), where kBT0 = 11 meV] for the x = 0.5 sample. (c) Magnetic field (0H) dependence of the Hall resistivity (yx) for the x = 0.5 sample at various T. (d) T dependence of carrier density [electron: n2D (closed circles), hole: p2D (open circles)]. Figures S6(a) and 2(b) show the temperature (T) dependence of the longitudinal resistivity (xx) for the CGT(12 nm)/(Bi1-xSbx)2Te3(2 nm)/Bi2Te3(2 nm)/(Bi1-xSbx)2Te3(2 nm)/CGT(12 nm) samples shown in Fig. 3 of the main text; logxx vs. (a) T and (b) 1/T. The samples with x = 0.3, 0.5 and 0.6, which exhibit large anomalous Hall angles, show double-step temperature dependence: increase in xx with decreasing T to 100 K, saturation at around 100 K, and increase again below 10 K. The T dependence in the higher temperature region (T > 100 K) can be attributed to freezing of the bulk carriers in the BST/BT/BST layers. The curves at the higher temperature region in Fig. S6(b) are rounded and the fitting to the Arrhenius-type temperature dependence [xx  exp(T0/T), where T0 is the thermal activation energy] does not work well. The deviation from the Arrhenius-type T dependence may suggest the presence of variable-range hopping transport among the localized states in the bulk part of the BST/BT/BST layer [S16-S18]. As T is lowered below 100 K, xx turns to be saturated. The saturation behavior ensures that the transport below 100 K is dominated by the surface states which are expected to show metallic T dependence. However, below 10 K, xx turns to increase again. This trend can be seen in the insulating samples. The xx increase at low T can be attributed to the reduction of surface carriers due to the exchange gap formation and/or their localization due to the disorder in the magnetized surface state. Such reduction of the surface carriers is also observed in Hall responses as shown in Fig. S6(c) for the x = 0.5 sample and in Fig. S6(d) for all the samples. (ii) The attempt to estimate the size of the exchange gap In the clean limit, the Hamiltonian of the TI surface state can be written as 27 𝐻(𝑘) = ℏ𝑣𝐹(𝑘𝑦𝜎𝑥 − 𝑘𝑥𝜎𝑦) + ∆ 2 𝜎𝑧, (1) where 𝑣𝐹 is the Fermi velocity, 𝜎𝑖 (i = x, y, z) are Pauli matrices acting on spins, and  is the exchange gap. The Hall conductivity (xy) is related to the Berry curvature, which increases as the Fermi level EF approaches E = ±. When the Fermi level lies in the gap (𝐸𝐹 < ∆/2), xy from two Dirac states (top and bottom surfaces) is quantized to 2 × 𝑒2/2ℎ = 𝑒2/ℎ. When the Fermi level lies above the gap (𝐸𝐹 > ∆/2), xy can be obtained as [S19] 𝜎𝑥𝑦 = 2 × 𝑒2 2ℎ ∆/2 √(ℏ𝑣𝐹)2𝑘𝐹 2 + (∆/2)2 = 𝑒2 ℎ ∆ 2𝐸𝐹 . (2) We consider only the intrinsic contribution from the band structure and ignore other contributions from skew scatterings and side jumps; the latter two extrinsic mechanisms are unlikely to work in the present case judging from the features of low carrier density and high scattering rate contrary to the case of clean itinerant ferromagnets [S20]. In addition, the carrier density of the surface state (n2D) is written as 𝑛2𝐷 = 2 𝑘𝐹 2𝜋 = 𝐸𝐹 2 − (∆/2)2 2𝜋(ℏ𝑣𝐹)2 , (3) therefore it is possible to estimate EF and  from the measured xy and n2D. Although we are already aware that our samples are influenced by disorder/inhomogeneity, it is inviting to estimate rough energy scales of EF and  by putting the experimentally measured values in Eqs. (2) and (3). We focus on the data of the bulk-insulating CGT(12 nm)/(Bi1-xSbx)2Te3(2 nm)/Bi2Te3(2 nm)/(Bi1-xSbx)2Te3(2 nm)/CGT(12 nm) sample (x = 0.5) at 2 K. From the normal Hall coefficient RH at high magnetic field, we estimate 𝑛2𝐷 = 1.9 × 1012 cm-2 assuming that all the contributions to RH are from the surface state. The value of anomalous Hall conductivity is 𝜎𝑥𝑦 = 0.16𝑒2/ℎ as shown in Fig. 3 of the main text. Putting these values together with 𝑣𝐹 = 3.6 × 105 m s-1 from the ARPES result [S8], EF and  are estimated to be 83 meV and 27 meV, respectively. 28 The estimated values may be valid if the amplitude of the potential fluctuation due to inhomogeneity is much smaller than EF and . However, as is the case in pristine and magnetically-doped TIs, the potential fluctuation can amount to be several tens meV [S16, S17]. Therefore, we should be cautious in applying the clean limit model to analyze the experimental result. In the presence of inhomogeneity, EF and  can have spatial variation. Then the sample becomes a patched network of different EF and . In such a case, the measured xy becomes a complicated combination of local xy. Nevertheless, the estimated EF value is well below the bulk bandgap value (~ 300 meV) and the estimated  value is large as ~ 1/3 of the EF. This may point to the necessary existence of the exchange gap, albeit spatially varying, so as to give rise to such a large anomalous Hall conductivity (or Hall angle) as observed. For more quantitative discussion, we need to estimate the amplitude of the potential fluctuation in the CGT/BST system and to construct an elaborate model taking the potential fluctuation into account, which will be an issue of future work. 29 VII. Characterization and magneto-transport data of BST/CGT bilayer structures. FIG. S7. X-ray reflectivity (XRR) for BST(6 nm)/CGT(t nm)/AlOx(5 nm) structures. (a) Schematic layout of the sample structure. (b) Measured (dots) and fitted (solid lines) XRR curves on a logarithmic scale for the BST/CGT bilayer structures with the various CGT thicknesses (t) as a function of momentum transfer (Qz). (c) X-ray SLDs as a function of the distance from the InP substrate surface, z. Triangles indicate the bottom and the top surface of CGT layers. By measuring x-ray reflectivity (XRR), we investigated CGT thickness (t) of the BST/CGT bilayer structures [Fig. S7(a)] used in Fig. 4 of the main text. Figure S7(b) and S7(c) show the XRR reflectivity curves and the depth profiles of the x-ray SLD, respectively. The SLD profiles, representing the thicknesses of layers and sharp interfaces, were simulated by the same procedure as the analysis of the data in Fig. 2(c) of the main text. We note that the root-mean-square roughness of CGT is as small as 0.74(5) nm. 30 FIG. S8. Additional magneto-transport data of BST(6 nm)/CGT(t nm)/AlOx(5 nm) structures. (a, b) Out-of-plane magnetic field dependence (0H ⊥ ) of the magnetization (M) (a) and the anomalous Hall conductivity (xy A) (b) at 2 K in the BST/CGT (t = 1.4, 2.2, 2.9, 4.6, 8.1, 12 nm) structures. (c-e) Temperature (T) dependence of the longitudinal sheet resistivity (xx) (c), the Hall resistivity (yx) (d), and the longitudinal sheet conductivity (xx) (e) measured under field cooling at 0H⊥ = 0.05 T. Figures S8(a) and S8(b) show the out-of-plane field (H⊥) dependence of the magnetization (M) and the anomalous Hall conductivity (xy A) derived by subtracting the H⊥-linear component at T = 2 K, 31 respectively. Both magnetic hysteresis loops in M and xy A are consistent with each other. We note that the negative slopes of the magnetization for t = 1.4 and 2.2 nm originate from the subtraction procedure of diamagnetism of InP substrates, which is conducted by a reference measurement of an InP substrate without films. Therefore, in Fig. 4(d) of the main text, we used the zero-field value of xy A and M to eliminate the residual substrate contribution. In Fig. S8(c)-S8(e), the temperature (T) dependence of the longitudinal sheet resistivity (xx), the Hall resistivity (yx) and the longitudinal sheet conductivity (xx) are shown for supporting the data of Fig. 4(d) in the main text. 32 References 1. M. Mogi, A. Tsukazaki, Y. Kaneko, R. Yoshimi, K. S. Takahashi, M. Kawasaki, and Y. Tokura, APL Mater. 6, 091104 (2018). 2. 3. R. Yoshimi, A. Tsukazaki, Y. Kozuka, J. Falson, K. S. Takahashi, J. G. Checkelsky, N. Nagaosa, M. Kawasaki, and Y. Tokura, Nat. Commun. 6, 6627 (2015). C.-Z. Chang, P. Tang, X. Feng, K. Li, X.-C. Ma, W. Duan, K. He, and Q.-K. Xue, Phys. Rev. Lett. 115, 136801 (2015). 4. M. Takeda et al., Chin. J. Phys. 50, 161 (2012). 5. 6. K. Nakajima et al. Quantum Beam Sci. 1, 9 (2017). Y. Inamura, T. Nakatani, J. Suzuki, and T. Otomo, J. Phys. Soc. Jpn. 82, SA031 (2013). 7. M. Björck, and G. Andersson, J. Appl. Crystallogr. 40, 1174 (2007). 8. 9. J. Zhang, C.-Z. Chang, Z. Zhang, J. Wen, X. Feng, K. Li, M. Liu, K. He, L. Wang, X. Chen, Q.- K. Xue, X. Ma, and Y. Wang, Nat. Commun. 2, 574 (2011). R. Dingle, H. L. Störmer, A. C. Gossard, and W. Wiegmann, Appl. Phys. Lett. 33, 665 (1978). 10. F. Katmis, V. Lauter, F. S. Nogueira, B. A. Assaf, M. E. Jamer, P. Wei, B. Satpati, J. W. Freeland, I. Eremin, D. Heiman, P. Jarillo-Herrero, and J. S. Moodera, Nature 533, 513 (2016). 11. L. D. Alegria, H. Ji, N. Yao, J. J. Clarke, R. J. Cava, and J. R. Petta, Appl. Phys. Lett. 105, 053512 (2014). 12. P. Wei, F, Katmis, B. A. Assaf, H. Steinberg, P. Jarillo-Herrero, D. Heiman, and J. S. Moodera, Phys. Rev. Lett. 110, 186807 (2013). 13. S. Zhu, D. Meng, G. Liang, G. Shi, P. Zhao, P. Cheng, Y. Li, X. Zhai, Y. Lu, L. Chen, and K. Wu, Nanoscale 10, 10041 (2018). 14. Z. Jiang, C.-Z. Chang, C. Tang, P. Wei, J. S. Moodera, and J. Shi, Nano Lett. 15, 5835 (2015). 15. C. Tang, C.-Z. Chang, G. Zhao, Y. Liu, Z. Jiang, C.-X. Liu, M. R. McCartney, D. J. Smith, T. Chen, J. S. Moodera, and J. Shi, Sci. Adv. 3, e1700307 (2017). 16. H. Beidenkopf, P. Roushan, J. Seo, L. Gorman, I. Drozdov, Y. S. Hor, R. J. Cava, and A. Yazdani, Nat. Phys. 7, 939 (2011). 17. I. Lee, C. K. Kim, J. Lee, S. J. L. Billinge, R. Zhong, J. A. Schneeloch, T. Liu, T. Valla, J. M. Tranquada, G. Gu, and J. C. Séamus Davis, Proc. Natl. Acad. Sci. USA 112, 1316 (2015). 18. B. Skinner, T. Chen, and B. I. Shklovskii, Phys. Rev. Lett. 109, 176801 (2012). 19. N. A. Sinitsyn, J. E. Hill, H. Min, J. Sinova, and A. H. MacDonald, Phys. Rev. Lett. 97, 106804 (2006). 20. N. Nagaosa, J. Sinova, S. Onoda, A. H. MacDonald, and N. P. Ong, Mod. Rev. Phys. 82, 1539 (2010). 33 

1504.06081

2

1504

Subharmonic transitions and Bloch-Siegert shift in electrically driven spin resonance

We theoretically study coherent subharmonic (multi-photon) transitions of a harmonically driven spin. We consider two cases: magnetic resonance (MR) with a misaligned, i.e., non-transversal driving field, and electrically driven spin resonance (EDSR) of an electron confined in a one-dimensional, parabolic quantum dot, subject to Rashba spin-orbit interaction. In the EDSR case, we focus on the limit where the orbital level spacing of the quantum dot is the greatest energy scale. Then, we apply time-dependent Schrieffer-Wolff erturbation theory to derive a time-dependent effective two-level Hamiltonian, allowing to describe both MR and EDSR using the Floquet theory of periodically driven two-level systems. In particular, we characterise the fundamental (single-photon) and the half-harmonic (two-photon) spin transitions. We demonstrate the appearance of two-photon Rabi oscillations, and analytically calculate the fundamental and half-harmonic resonance frequencies and the corresponding Rabi frequencies. For EDSR, we find that both the fundamental and the half-harmonic resonance frequency changes upon increasing the strength of the driving electric field, which is an effect analogous to the Bloch-Siegert shift known from MR. Remarkably, the drive-strength dependent correction to the fundamental EDSR resonance frequency has an anomalous, negative sign, in contrast to the corresponding Bloch-Siegert shift in MR which is always positive. Our analytical results are supported by numerical simulations, as well as by qualitative interpretations for simple limiting cases.

cond-mat.mes-hall

cond-mat

a Subharmonic transitions and Bloch-Siegert shift in electrically driven spin resonance Judit Romh´anyi,1, 2 Guido Burkard,3 and Andr´as P´alyi2, 4 1Leibniz-Institute for Solid State and Materials Research, IFW-Dresden, D-01171 Dresden, Germany 2Institute of Physics, Eotvos University, Budapest, Hungary 3Department of Physics, University of Konstanz, D-78457 Konstanz, Germany 4MTA-BME Condensed Matter Research Group, Budapest University of Technology and Economics, Budapest, Hungary (Dated: August 23, 2018) We theoretically study coherent subharmonic (multi-photon) transitions of a harmonically driven spin. We consider two cases: magnetic resonance (MR) with a misaligned, i.e., non-transversal driv- ing field, and electrically driven spin resonance (EDSR) of an electron confined in a one-dimensional, parabolic quantum dot, subject to Rashba spin-orbit interaction. In the EDSR case, we focus on the limit where the orbital level spacing of the quantum dot is the greatest energy scale. Then, we apply time-dependent Schrieffer-Wolff perturbation theory to derive a time-dependent effective two-level Hamiltonian, allowing to describe both MR and EDSR using the Floquet theory of period- ically driven two-level systems. In particular, we characterise the fundamental (single-photon) and the half-harmonic (two-photon) spin transitions. We demonstrate the appearance of two-photon Rabi oscillations, and analytically calculate the fundamental and half-harmonic resonance frequen- cies and the corresponding Rabi frequencies. For EDSR, we find that both the fundamental and the half-harmonic resonance frequency changes upon increasing the strength of the driving elec- tric field, which is an effect analogous to the Bloch-Siegert shift known from MR. Remarkably, the drive-strength dependent correction to the fundamental EDSR resonance frequency has an anoma- lous, negative sign, in contrast to the corresponding Bloch-Siegert shift in MR which is always positive. Our analytical results are supported by numerical simulations, as well as by qualitative interpretations for simple limiting cases. PACS numbers: 71.70.Ej 73.21.La 76.20.+q I. INTRODUCTION Magnetic resonance (MR) is an established method to coherently control the quantum state of spins. A sim- ple example is a spin-1/2 electron subject to a time- dependent magnetic field1 -- 4, described by the Hamilto- nian H = − 1 2 gµB B(t) · σ, (1) where the magnetic field B(t) = (0,−Bac cos ωt, B) con- sists of a 'longitudinal' static component B and a 'trans- verse' ac component characterised by the drive strength Bac and the drive frequency ω, and couples to the elec- tron spin represented by the vector σ = (σx, σy, σz) of Pauli matrices. A typical initial-value problem considered in MR is when the initial state of the spin ψ(t = 0) = ↑i is the ground state of the static Hamiltonian, − 1 2 gµBBσz, In the and driving is switched on abruptly at t = 0. case of weak driving Bac ≪ B, the rotating wave ap- proximation (RWA) often provides a satisfactory descrip- tion of the dynamics. Using this approximation, one finds the following simple phenomenology. If the reso- nance condition ω = gµBB is fulfilled, the drive will induce complete Rabi oscillations resulting in a transi- tion probability P↓(t) ≡ h↓ ψ(t)i2 = sin2(Ωt/2), where Ω = gµBBac/(2) is called the Rabi frequency. Oth- erwise, i.e., in the case of a finite detuning δ = ω − gµBB/ between the drive frequency and the resonance frequency, one finds incomplete Rabi oscillations with a δ- dependent frequency: P↓(t) = P max with P max ↓ = Ω2/(Ω2 + δ2) < 1. ↓ sin2(√δ2 + Ω2t/2), Still focusing on the weak-driving regime Bac ≪ B, one can go beyond the RWA, e.g., by numerical sim- ulations or analytical techniques such as the Floquet perturbation theory2. Then, a richer phenomenology is revealed, including (i) subharmonic or 'multi-photon' resonances2, (ii) drive-strength-dependent Bloch-Siegert shifts1 (BSSs) of the resonance frequencies, and (iii) Bloch-Siegert oscillations modulating the simple Rabi oscillations1. We restrict our attention to (i) and (ii) here. (i) In the case of a transverse ac field, such as the example used in Eq. (1), odd subharmonic resonances appear2. Rabi oscillations are obtained not only for the fundamental resonance ω ≈ gµBB/, but also when In the case of ω ≈ gµBB/(N ) with N = 3, 5, 7, . . . . a misaligned, non-transversal, ac field, such as B(t) = (0,−Bac cos θ cos ωt, B − Bac sin θ cos ωt) with 0 < θ < π/2, both even and odd subharmonics are present. The Rabi frequency Ω(N ) res at the N -photon subharmonic res- onance is weaker than that of the fundamental one: Ω(N ) res ∝ BN (ii) The resonance frequencies ω(N ) res (i.e., the drive fre- quencies where complete Rabi oscillations are induced) increase with increasing drive strength, by an amount that depends on N , and is proportional to B2 ac/B(N −1). ac/B. In many situations, it is be more convenient to con- trol spins using an ac electric field rather than an ac magnetic field. For example, if an electron spin is elec- trostatically confined in a quantum dot (QD), then an ac electric field can be easily created by applying an ac voltage component of the confinement gate electrodes. Along these lines, electrically driven spin resonance5 -- 11 (EDSR) of individual electron spins was demonstrated in a variety of materials12 -- 23. As the ac electric field cou- ples to the orbital degree of freedom of the electron and has no direct effect on the spin, a sufficiently strong cou- pling mechanism between the orbit and spin is required for EDSR. Such a coupling can be supplied by spin-orbit interaction, hyperfine interaction, or an inhomogeneous magnetic field. Recent experimental20,23 -- 26 and theoretical27 -- 34 stud- ies addressed subharmonic resonances in EDSR. One mechanism that leads to subharmonic resonances in EDSR is the appearance of higher harmonics N ω of the drive frequency ω in the induced orbital dynamics29,31. In this case, the time-dependent effective magnetic fields caused by the orbital dynamics will also have compo- nents at frequency N ω, leading to Rabi oscillations as N ω matches the Zeeman splitting. Higher harmonics in the orbital dynamics arise naturally if the confinement potential is anharmonic, or if the driving electric field is inhomogeneous. Subharmonic EDSR resonances can also arise in the presence of harmonic confinement and homo- geneous ac electric field, if the gradient of the effective magnetic fields is inhomogeneous; this is the case, e.g., if the the effective magnetic field is spatially localized or disordered32. A third mechanism, able to cause strong subharmonics with large N , is provided by Landau-Zener dynamics in the vicinity of level anticrossings26,34. In this work, we theoretically describe the character- istics of the half-harmonic resonance in MR with a mis- aligned ac field, as well as in EDSR. First, we use Flo- quet perturbation theory2 to characterise the parame- ter dependence of the half-harmonic resonance frequency and the corresponding Rabi frequency in the case of MR; in particular, the BSS is calculated. As for EDSR, we study a model35 (see Fig. 1), where a single electron is parabolically confined in a one-dimensional (1D) quan- tum dot, and is subject to a dc magnetic field, an ac electric field, and spin-orbit interaction of Rashba type, the latter three being spatially homogeneous. We show that the half-harmonic resonance does arise in this model, despite the harmonic confinement and homogeneous ac electric field. In the perturbative regime of this model, i.e., when the orbital level spacing ω0 dominates over other energy scales, we analytically derive the parame- ter dependence of the half-harmonic resonance frequency and the corresponding Rabi frequency. This is achieved via a combination of time-dependent Schrieffer-Wolff per- turbation theory (TDSW), which is used to obtain a 2×2 effective 'two-level' or 'qubit' Hamiltonian Hq, and Flo- quet perturbation theory, applied to describe the qubit dynamics governed by Hq. This paper is structured as follows. In Sec. II we sum- marize our main results. Secs. III and IV are dedicated to the detailed discussion of MR and EDSR, respectively. ( ) (b) E(t)=E sin( ac  2 y n SO t)  B z SO SO B E E n=2 n=1 n=0 ℏ 0 ℏ 0
FIG. 1. Electrically driven spin resonance in a 1D quantum dot. (a) An electron occupying the ground state ψ0(z) of a parabolic confinement potential is excited by an ac electric field of amplitude Eac and frequency ω. The electron is sub- ject to a homogeneous magnetic field B and spin-orbit inter- action characterised by the direction nSO. (b) Left: Orbital levels labelled by the oscillator quantum number n, separated by the level spacing ω0. Right: Diagram representing one of the many fifth-order virtual processes contributing to the half-harmonic resonance. Horizontal lines represent the en- ergy eigenstates of the harmonic oscillator Hamiltonian. Ar- rows labelled by E, B and SO correspond to matrix elements of the ac electric field, magnetic field and spin-orbit Hamil- tonians, respectively. Note that the spin-orbit Hamiltonian provides both spin-flip and spin-conserving matrix elements only if both cos θ and sin θ are nonzero. [Cf. Eq. (36).] In them we formulate the problems, derive analytical so- lutions and compare these with numerical simulations where called for. In Sec. V we give a conclusion of our findings. II. SUMMARY OF THE RESULTS In this Section, we summarize the main results that are derived in later Sections. Let us start with the case of spin-1/2 MR with a misaligned ac field. The param- eters of the model are B, the energy scale of the static magnetic field along the z axis; Bac, the energy scale of the ac magnetic field oriented in the yz plane; and θ, the angle enclosed by the ac field and the y axis. (For more details, see Sec. III.) The fundamental (or single-photon) resonance frequency ω(1) res , that is, the drive frequency at which the Rabi oscillations are complete, deviates from the Zeeman splitting: ω(1) res = B + ω(1) BSS, (2) where the second term on the right hand side is the BSS and has the form ω(1) BSS = B2 ac cos2 θ 16 B . (3) The half-harmonic resonance frequency is shifted with re- spect to the half of the fundamental resonance frequency: 3 (4) ω(2) res = 1 2(cid:16) B + ω(1) g + ω(1) nlZ(cid:17) + ω(2) BSS, (13) where the drive-strength-dependent BSS is expressed as ω(2) BSS = 2 B α2 E2 ac cos2 θ 34ω4 0 . (14) The Rabi frequency at the half-harmonic resonance is Ω(2) res = B α2 E2 ac sin(2θ) (ω0)4 . (15) Note that the EDSR resonance and Rabi frequencies above are expressed up to fifth order in the small energy scales α, B, Eac ≪ ω0. A detailed discussion of these results, and their comparison with numerical solutions of the time-dependent Schrodinger equation, is included in Sec. IV D. III. MAGNETIC RESONANCE WITH A MISALIGNED AC FIELD In this Section, using Floquet perturbation theory, we derive and discuss the properties of the fundamen- tal (single-photon) and half-harmonic (two-photon) res- onances in MR, for the spin-1/2 case. In particular, the results (2) -- (7) are derived. The Rabi frequency at the fundamental resonance is Ω(1) res = Bac 2 cos θ. Similarly to the fundamental the half- harmonic (two-photon) resonance also acquires a positive BSS: resonance, where ω(2) res = B 2 + ω(2) BSS, ω(2) BSS = B2 ac cos2 θ 6 B . (5) (6) The Rabi frequency at the half-harmonic resonance is Ω(2) res = B2 ac sin 2θ 4 B . (7) Note that the resonance frequencies above are expressed up to second order in the small energy scale Bac ≪ B. For a detailed discussion of these results, see Sec. III D. In the case of EDSR in a 1D parabolic QD, the param- eters characterizing the model are as follows. The orbital level spacing ω0 is the dominant energy scale; the static magnetic field, oriented along the z axis, is character- ized by the Zeeman-splitting energy scale B; spin-orbit interaction is described by the energy scale α and the unit vector nso = (0, cos θ, sin θ) which points along the spin-orbit field; and the energy scale Eac describing the strength of the driving ac electric field. We find the fol- lowing results for the fundamental resonance frequency: where the correction consists of a g-factor renormaliza- tion term, ω(1) g = − 2 B α2 cos2 θ 2ω2 0 (cid:18)1 − α2(1 + sin2 θ) 2ω2 0 (cid:19) , (9) a term describing the non-linear Zeeman effect, ω(1) nlZ = 2 B3 α2 cos2 θ 4ω4 0 , (10) and a correction that is second order in the drive strength Eac, hence analogous to the BSS: ω(1) BSS = − B α2 E2 ac cos2 θ 4ω4 0 . (11) Note that the sign of this BSS is negative, in contrast to the positive sign in the case of MR [Eqs. (3) and (6)]; an interpretation of this anomalous sign is given in Sec. V, paragraph (4). The Rabi frequency at the fundamental resonance is Ω(1) res = 2 B Eac α cos θ 2ω2 0 1 + 0 ! . B2 − 2 α2 2ω2 (12) ω(1) res = B + ω(1) g + ω(1) nlZ + ω(1) BSS, (8) A. Problem formulation (16) (17) We consider MR spin dynamics driven by a misaligned ac field. The Hamiltonian reads B(t) · σ , where the magnetic field has the form H(t) = − 1 2 B(t) =  0 − Bac cos θ cos ωt B − Bac sin θ cos ωt   . Here we introduced B = gµBB and Bac = gµBBac. Henceforth the parameters with tilde, e.g., B, have en- ergy dimension. The Hamiltonian in Eq. (16) has four parameters: the strength of the static field B, the strength of the driving field Bac, the frequency of the driving field ω, and the misalignment angle θ. Note that θ = 0 corresponds to a transverse ac field, and θ = π/2 corresponds to a longitudinal ac field. We consider the case of weak driving, Bac ≪ B. In particular, we want to solve the initial-value prob- I: the initial state is the ground lem described in Sec. state ↑i of the Hamiltonian without driving i.e., ψ(t = 0) = ↑i, driving is switched on abruptly at t = 0, and we are interested in the time evolution ψ(t) of this state. We calculate the transition probability describing the time-dependent occupation of the excited state ↓i at the fundamental and half-harmonic resonances, and from those we deduce the parameter dependencies of the resonance frequencies and the Rabi frequencies. In the rest of this Section, we use Floquet perturba- tion theory2,36 -- 38 to derive the results and to provide qualitative interpretations in simple limiting cases, such as the limits of transversal and longitudinal ac fields. Even though similar treatments can be found in the literature2,38, we present a detailed discussion of the MR problem for the following reason. The MR problem is rel- atively simple as compared to the EDSR problem, which can be appreciated, e.g., by comparing the driven two- level Hamiltonians of Eqs. (16) and (41), respectively. Moreover, as we will show, the Floquet method and the qualitative interpretations we describe here for the MR problem can be carried over to the EDSR problem, once the 2 × 2 effective qubit Hamiltonian in Eq. (41) is ob- tained for the latter. This allows us to provide a rather compact description of the EDSR in the forthcoming Sec- tions, by referencing this Section wherever possible. B. Floquet method The Floquet method allows one to find the solution of an initial-value problem of a periodically driven quan- tum system, described by the time-periodic Hamiltonian H(t) = H(t + T ). The period of the driving is de- noted by T , and the corresponding (angular) frequency by ω = 2π/T . The key ingredient of the method is the quantum-mechanical Floquet theorem36, which guaran- tees that the Schrodinger equation i Ψ(t) = H(t)Ψ(t) of a d-level system has d solutions Ψk(t) (k = 1, . . . , d) that are themselves periodic with period T , apart from a phase factor. Therefore, these special solutions have the form Ψk(t)i = e−iEkt/ d ∞ Xl=1 Xm=−∞ ck,lmeimωtψli , (18) where ψli is an arbitrary basis of the Hilbert space. Note that the result of the double sum is a periodic function of t with period T . In Eq. (18), the quantity Ek and the coefficients ck,lm are a priori unknown; the former is called quasi-energy. Once these special solutions Ψk(t)i are found, they provide the propagator U (t, 0) = d Xk=1 Ψk(t)ihΨk(0), (19) which in turn provides the solution of any initial-value problem via Ψ(t) = U (t, 0)Ψ(0). (20) The special solutions Ψk(t) are found by using Eq. (18) as an Ansatz, substituting it to the Schrodinger equation, 4 evaluating the scalar product of the equation with hψl′, multiplying the equation by e−im′ωt and integrating the equation in time between t = 0 and t = T . This proce- dure yields the following eigenvalue equation for Ek: d ∞ Xl=1 Xm=−∞ where Fl′m′,lmck,lm = Ekck,l′m′, (21) ∞ Fl′m′,lm= mωδl′lδm′m+ hψl′H(n)ψliδm′,n+m(22) Xn=−∞ is the Floquet matrix or Floquet Hamiltonian, and we introduced the Fourier components H(n) of the Hamilto- nian via H(t) = ∞ Xn=−∞ H(n)einωt. (23) We call two eigenvalue-eigenvector pairs of F equiva- lent, if the two time-dependent solutions they generate via Eq. (18) are the same. Importantly, even though the number of eigenvalue-eigenvector pairs of F is infinite, they form only d equivalence classes. In summary, we have transformed the time-dependent Schrodinger equation of the periodically driven d × d Hamiltonian H(t) into the time-independent Schrodinger equation (21) of the infinite-dimensional Floquet Hamil- tonian F . To construct the special solutions (18), and thereby the solution of any initial-value problem via Eqs. (19) and (20), the quasi-energies Ek and the correspond- ing eigenvectors ck should be found by solving the eigen- value problem of the Floquet Hamiltonian F . C. Perturbative description of the transition probability After reviewing the Floquet method in general, we now apply this to the MR problem defined in Eq. (16). Here, we have a two -- level system, therefore d = 2, and we use αi ≡ ↑i ≡ ψ1i and βi ≡ ↓i ≡ ψ2i to denote these levels. The Fourier components of the Hamiltonian read H(0) = − 1 H(±1) = 4 Bσz, 1 2 Bac(cos θσy + sin θσz), (24a) (24b) and the other Fourier components are zero. First, consider the case when the drive frequency is close to the fundamental resonance, ω ≈ B. Then, the diagonal elements of the Floquet Hamiltonian F (Floquet levels) form pairs: 1 2 Fαm,αm = mω− B ≈ (m−1)ω+ B = Fβ,m−1,β,m−1. (25) The distance between different pairs is approximately ω ≈ B, which is much larger than the energy scale Bac 1 2 characterising the off-diagonal elements of F . Therefore, the tools of quantum-mechanical perturbation theory can be used to provide an approximate solution of the eigen- value problem of the Floquet Hamiltonian. The structure of the Floquet Hamiltonian F is visu- alised for the case ω = B in the level diagram shown in Fig. 2. Horizontal lines represent the diagonal ma- trix elements Flm,lm of the Floquet Hamiltonian, their vertical positions correspond to their value, their colour (black, red) represents their spin index l ∈ (α, β) and their horizontal position stands for their Floquet index m = . . . ,−1, 0, 1, 2, . . . . The vertical spacing of the Flo- quet levels is ω = B. The blue arrows indicate the nonzero off-diagonal matrix elements of F , which are of the order of Bac and hence small compared to the level spacing. In the case ω = B shown in Fig. 2, the Floquet lev- els form degenerate pairs. The pair formed by Fβ,−1,β,−1 and Fα,0,α,0 is highlighted in Fig. 2 by the blue box. The subspace of this pair is weakly coupled to the other Flo- quet levels, hence this coupling can be treated perturba- tively using (time-independent) Schrieffer-Wolff pertur- bation theory39, which is also known as quasi-degenerate perturbation theory40. This perturbative treatment is also applicable if there is a finite, but small detuning 5 δ = ω − B/ ≪ B/ from the resonance condition. The small dimensionless parameter characterising the strength of the perturbation is ǫ = Bac/ B. In this case, the Floquet Hamiltonian reads θ= FIG. 2. Magnetic resonance in a misaligned ac field: struc- ture of the Floquet Hamiltonian at the fundamental reso- nance. Panels show cases when the ac field is perpendicular to the static field (a), is parallel to the static field (b), has finite perpendicular and parallel components (c). Horizon- tal lines (blue arrows) correspond to diagonal (off-diagonal) matrix elements of the Floquet Hamiltonian F. The vertical position of each horizontal line corresponds to the value of the diagonal matrix element. F = β−1 ↓ ... 0 α−1 ↓ ... B − ω 0 1 2 4 Bac sin θ − i Bac cos θ − 1 4 B − ω Bac cos θ Bac sin θ 0 0 ... 2 1 4 α−1 → . . . − 1 β−1 → . . . α0 → . . . β0 → . . . α1 → . . . β1 → . . . i 4 0 0 ... α0 ↓ ... β0 ↓ ... 4 Bac sin θ − i Bac cos θ − 1 − 1 0 B 4 2 Bac cos θ Bac sin θ 0 B 1 2 1 4 i 4 4 Bac sin θ − i Bac cos θ − 1 4 Bac cos θ − 1 Bac sin θ 2 ... ... 1 4 i 4 1 4 i 4 α1 ↓ ... 0 0 β1 ↓ ... 0 0 . . . . . . (26) Bac cos θ . . . Bac sin θ . . . 4 Bac sin θ − i Bac cos θ − 1 B + ω 4 0 . . . . . . 0 ... B + ω 1 2 ... 1. Fundamental resonance within RWA Using first-order perturbation theory, the two non- equivalent eigenvalues and eigenvectors of F can be found approximately. For this we introduce F0 and F1 so that F = F0 + F1. F0 is the diagonal component of F at ω = B/., i.e. at δ = 0. First-order perturbation theory in F1 amounts to di- agonalizing the 2 × 2 block highlighted in purple in Eq. (26) and Fig. 2. For future reference, we recast this 2× 2 block to the form F =(cid:20) ǫ0 + ∆ iλ 2 ( B + δ), ∆ = −δ/2, and λ = 1 ǫ0 − ∆(cid:21) , −iλ 4 where ǫ0 = − 1 (27) Bac cos θ. The matrix F has eigenvalues E± = ǫ0 ±p∆2 + λ2 , and corresponding eigenvectors c± = N±[ i λ (∆ ±p∆2 + λ2), 1] , where N± is a normalization constant. Note that in- stead of using the numerical index k ∈ (1, 2) labelling the solutions (18), in Eqs (28) and after we use the val- ues k ∈ (+,−). The results (28) and (29) imply that the two non- equivalent approximate eigenvalue-eigenvector pairs of F are ( E±, c±), where (28) (29) c±,1 if (l, m) = (β,−1), c±,2 if (l, m) = (α, 0), 0 otherwise. , (30) c±,lm =  and c±,1 and c±,2 are the components of c± in Eq. (29). This result allows us to construct the transition probabil- ity Pβ←α(t) = hβΨ(t)i2 from the initial spin (ground) state ↑i ≡ αi to the excited state ↓i ≡ βi via Eqs. (18), (19) and (20). A straightforward calculation yields Pβ←α(t) = λ2 λ2 + ∆2 sin2(cid:18) 1 p∆2 + λ2 t(cid:19) . (31) According to Eq. (31), the spin makes complete Rabi oscillations if ∆ = 0, that is, δ = ω− B/ = 0. Hence the single-photon resonance frequency is ω(1) res = B/. The Rabi frequency upon resonant driving is Ω(1) res = 2λ/ = Bac 2 cos θ; thus only the transverse component of the ac field contributes to the Rabi frequency at the fundamen- tal resonance. In fact, the result (31) is equivalent to the one obtained by neglecting the longitudinal ac field and performing RWA. 2. Fundamental resonance beyond the RWA: Bloch-Siegert shift of the resonance frequency res and Ω(1) Let us discuss the corrections to ω(1) res beyond the RWA. To this end, we incorporate in the analysis the effect of those matrix elements of F1 that connect the two highlighted Floquet levels [see Eq. (26) and Fig. (2)] to the complementary subspace. This is done via a (time-independent) Schrieffer-Wolff transformation that is second order in F1. The resulting effective 2×2 Floquet Hamiltonian F has the form given in Eq. (27), with ∆ = − 1 2 δ + B2 ac cos2 θ 32 B λ = Bac cos θ 4 . (32a) (32b) Recall that the eigenvalues and eigenvectors of F are given by Eqs. (28) and (29). From these, we conclude c±,lm =  that the two non-equivalent approximate eigenvalue- eigenvector pairs of F are ( E±, c±), where 6 c±,1 + o(ǫ2) if (l, m) = (β,−1), c±,2 + o(ǫ2) if (l, m) = (α, 0), o(ǫ) otherwise. (33) We neglect the perturbative corrections ∼ o(ǫ), o(ǫ2) in the eigenvectors c±, and this implies that the approxi- mate transition probability is given by Eq. (31). Equa- tion (31) predicts that complete Rabi oscillations are in- duced when ∆ = 0; solving Eq. (32a) for ω (recall that δ = ω − B/) provides the resonance frequency shown in Eq. (2). The second term of Eq. (2) corresponds to the Bloch-Siegert shift of the resonance frequency: as the drive strength Bac is increased, the resonance frequency shifts upwards. This feature is further discussed in Sec. III D. Finally, the Rabi frequency at the fundamental res- onance, which is given by Ω(1) res = 2λ, is expressed using Eq. (32b) in Eq. (4); the result is the same as in the RWA. 3. Half-harmonic resonance Let us now consider the spin dynamics at half- harmonic resonance, when ω ≈ B/2. The level di- agram visualising the Floquet Hamiltonian in the case ω = B/2 is shown in Fig. 3. Again, we can iden- tify degenerate pairs of Floquet levels, e.g., the pair (Fβ,−1,β,−1,Fα,1,α,1) highlighted with the blue box in Fig. 3. θ= FIG. 3. Magnetic resonance in a misaligned ac field: struc- ture of the Floquet Hamiltonian at the half-harmonic reso- nance. Panels show cases when the ac field is perpendicular to the static field (a), is parallel to the static field (b), has finite perpendicular and parallel components (c). Horizon- tal lines (blue arrows) correspond to diagonal (off-diagonal) matrix elements of the Floquet Hamiltonian F. The vertical position of each horizontal line corresponds to the value of the diagonal matrix element. Note that in this case, there is no direct matrix ele- ment (blue arrow) connecting these two Floquet levels. This implies that by repeating the first-order perturba- tion theory (equivalent to RWA) done in Sec. III C 1, we would conclude that two-photon Rabi oscillations do not happen. However, this result is not correct; two-photon Rabi oscillations can happen. To see that, we perform a second-order Schrieffer-Wolff transformation on F , as done in Sec. III C 2. Furthermore we use the appropriate notation δ = ω− B 2 . The obtained effective 2× 2 Floquet Hamiltonian F has the form given in Eq. (27), with B2 ac cos2 θ 6 B , ac sin 2θ . ∆ = δ − B2 λ = 8 B (34a) (34b) Following the approach used in Sec. III C 2, and solv- ing ∆ = 0 we find that the half-harmonic resonance fre- quency ω(2) (5). Furthermore, using Ω(2) res = 2λ and Eq. (34b), the Rabi frequency at the half-harmonic resonance is obtained as shown in Eq. (7). res is given by Eq. D. Discussion Let us now discuss the main features of the results (2), (4), (5), and (7). Consider first the fundamental resonance frequency ω(1) res expressed in Eq. (2). The second term in Eq. (2) implies that ω(1) res has a positive drive-strength-dependent ac/ B with respect to the nominal Zeeman correction ∝ B2 splitting B. This correction is known as the BSS, which can be regarded as a special case of the ac Stark shift41. Note that the parameter λ and hence the Rabi fre- quency Ω(1) res sets the frequency broadening of the funda- mental transition, as indicated by the prefactor λ2/(λ2 + ∆2) on the right hand side of Eq. (31). According to Eq. (4), this power broadening of the fundamental resonance is greater by a factor of B/ Bac than the BSS. Equation (2) also shows that the BSS is finite in the limit of purely transversal drive (θ = 0), and vanishes in the limit of purely longitudinal drive (θ = π/2). The respective Floquet level diagrams in Fig. 2a and b pro- vide a straightforward interpretation: the BSS can be regarded as a consequence of coupling-induced repulsion between the Floquet levels. In Fig. 2a (θ = 0), the Floquet level Fβ,−1,β,−1 is connected by a blue arrow (off-diagonal matrix elements of F ) to the lower-lying Floquet level Fα,−2,α,−2. The consequence of this cou- pling in second-order perturbation theory is level repul- sion; i.e., the lower-lying Floquet level pushes Fβ,−1,β,−1 upwards. Similarly, Fα0,α0 is pushed downwards by its coupling to the higher-lying Floquet level Fβ1,β1. These second-order level shifts appear in Eq. (32a) as the last term, and give rise to a finite BSS. In contrast, each of the highlighted Floquet levels in Fig. 2b (θ = π/2) is con- nected to one higher-lying and one lower-lying Floquet level, and the corresponding downward and upward level repulsions cancel each other, giving rise to a vanishing BSS in this case. 7 Consider now the half-harmonic resonance. Equation (7) provides the corresponding Rabi frequency, and it in- dicates the existence of Rabi oscillations unless θ = 0 or θ = π/2. I.e., Rabi oscillations appear at half-harmonic excitation only if the transversal and longitudinal compo- nents of the driving field are both nonzero. The Floquet level diagrams shown in Fig. 3 provide a visual interpre- tation of this feature: Rabi oscillations arise if the blue arrows (off-diagonal matrix elements of F ) draw at least one path between the two Floquet levels highlighted by the purple box, via virtual intermediate Floquet levels outside the box. In the special cases θ = 0 and θ = π/2 depicted in Figs. 3a and b, respectively, no such paths exist. However, there exist infinitely many such paths for 0 < θ < π/2 (Fig. 3c), due to the coexistence of spin- conserving and spin-flip off-diagonal matrix elements. In particular, in our second-order Schrieffer-Wolff transfor- mation leading to the result (7), the two two-step paths via Fα,0,α,0 and Fβ,0,β,0 are incorporated. In the case of the half-harmonic resonance, the re- lation between the power broadening and the BSS is qualitatively different from the case of the fundamental resonance. For the half-harmonic resonance, the power broadening is given by Eq. (7), whereas the BSS is given by the second term of Eq. (5), i.e., the two quantities ac/ B. Hence we are of the same order, both being ∼ B2 expect that for the half-harmonic resonance, the BSS is relatively easily resolvable experimentally, at least if the dissipative frequency scales are smaller than the power broadening. Equation (5) also shows that the BSS is finite in the limit of purely transversal excitation (θ = 0), and vanishes in the limit of purely longitudinal excitation (θ = π/2). An interpretation completely analogous to the case of the fundamental resonance can be given based on the Floquet level diagrams in Fig. 3a and b. IV. ELECTRICALLY DRIVEN SPIN RESONANCE A. The model From now on, we describe EDSR mediated by spin- orbit interaction in a 1D parabolic quantum dot. The setup is shown in Fig. 1. The Hamiltonian H = H0 + HE + HB + HSO (35) includes the harmonic-oscillator Hamiltonian (H0) con- sisting of the kinetic energy of the electron and the parabolic confinement potential, the ac electric potential arising from the driving electric field (HE), the static Zeeman effect caused by a homogeneous magnetic field (HB), and the spin-orbit term (HSO). The explicit forms of these terms, respectively, are as follows: B. Effective qubit Hamiltonian 8 1 + p2 z 2m 1 2 (36a) mω2 2(cid:19) , 0z2 = ω0(cid:18)a†a + H0 = HE = ezEac sin(ωt) = Eac sin(ωt)(a† + a ), (36b) (36c) HB = − HSO = αpz nso · σ = i α(a† − a)nso · σ (36d) Here, a and a† are the ladder operators of the harmonic oscillator Hamiltonian, and nso = (0, cos θ, sin θ) is the direction of the effective magnetic field arising from spin- orbit coupling. Furthermore, we defined g∗µBBσz = − Bσz 1 2 1 2 B = g∗µBB α = αr mω0 Eac = eEacr  2 2mω0 These quantitites have the dimension of energy. (37) (38) (39) Note that we use the same notation θ for two different quantities: θ appears in Eq. (17) as the ac field misalign- ment angle in MR, and it also appears in this Section and in Fig. 1, as the angle characterising the direction of the spin-orbit term. We use the same notation for these quantities as they play very similar roles in the spin dy- namics. It is natural to represent the Hamiltonian terms (36) in the product basis of the orbital and spin degrees of freedom, {nσi n = 0, 1, 2, . . . ; σ =↑,↓}, where n is the harmonic-oscillator orbital quantum number and σ is the spin quantum number with quantization along z. We will refer to the two lowest-energy eigenenstates of our static Hamiltonian H0 + HB + HSO as the qubit basis states. The qubit basis state with the lower (higher) energy will be denoted by Gi (Ei). The electron is initialized in state Gi at t = 0. Our aim is to describe the time evolution of the state upon driving. In particular, we are interested in the time- dependent occupation probability PE(t) of state Ei. It is expected that at resonant driving ω ≈ B, the dy- namics resembles Rabi oscillations. Subharmonic (multi- photon or N -photon) resonances at ω ≈ B/N (where N = 1, 2, . . . ) are also expected. In this work we focus on the fundamental (single-photon, N = 1) and half- harmonic (two-photon, N = 2) resonances. We aim at an analytical, perturbative description of spin transitions induced by the ac electric field. In partic- ular, we calculate the resonance frequency and the Rabi frequency at resonant driving. We consider the parame- ter range where the energy scale ω0 of the confinement potential dominates the other four energy scales, the lat- ter ones being assumed to be comparable in magnitude: ω ∼ α ∼ Eac ∼ B ≪ ω0. (40) This hierarchy of energy scales will allow for a perturba- tive description of the dynamics, with the small parame- ter ǫ ∼ ω ω0 ∼ B ω0 ≪ 1. ω0 ∼ α ω0 ∼ Eac In the EDSR problem defined in Sec. IV A, the hier- archy of the energy scales is given by Eq. (40). Because of this hierarchy, an effective time-dependent two-level Hamiltonian [see Eq. (41) below] can be derived for the qubit dynamics, using TDSW perturbation theory, which we outline in Appendix A. This qubit Hamiltonian can then be used to express the resonance frequencies ω(1) res and ω(2) res , and the corresponding Rabi frequencies at these resonances, Ω(1) res, corresponding to the funda- mental and half-harmonic resonances, respectively [see Eqs. (8), (8), (13), and (15) below]. res and Ω(2) We use the orbital-spin product basis {nσi n = 0, 1, 2, . . . ; σ =↑,↓}, as the starting point of TDSW, and take the two-dimensional subspace of 0↑i and 0↓i as the relevant subspace in TDSW. We carry out a fifth-order TDSW (in the small parameter ǫ), which is expected to describe both the fundamental and the half-harmonic res- onances. The TDSW procedure yields the effective qubit Hamiltonian Hq ≈ H(0) q + H(1) q + H(2) q + H(3) q + H(4) q + H(5) q , (41) where the six terms, representing terms from different orders in the perturbation, are listed below in Eq. (42). Note that the terms H(0) q are proportional to the 2 × 2 unit matrix σ0, therefore they do not influ- ence the dynamics, and hence we disregard them in the forthcoming calculations; nevertheless we include them here for completeness: q , and H(4) q , H(2) σ0 ω0 2 B 2 α2 + E2 σ3 H(0) q = H(1) q = − H(2) q = − H(3) q = − ac sin2(ωt) ω0 B Eac α cos θ σ0 sin(ωt)σ1 2ω2 0 (42a) (42b) (42c) (42d) ( Eacω cos(ωt) + B α sin θ)σ2 − + α cos θ 2ω2 0 α 2ω2 0 ( B α cos2 θ − Eacω sin θ cos(ωt))σ3 ( B α cos θ)2 H(4) q = − q = −(cid:16)h(5) H(5) σ0 , 3ω3 0 x σx + h(5) y σy + h(5) z σz(cid:17) , (42e) (42f) In Eq. (42f), we used h(5) x = h(5) y = B α Eac cos θ 4ω4 0 (2 α2 − B2) sin(ωt) (43a) Eac αω3 cos θ ω4 0 cos(ωt) + B α2 sin 2θ 24ω4 0 ( B2 − α2 + E2 ac sin2(ωt)) (43b) h(5) z = Eac αω3 sin θ ω4 0 cos(ωt) B α2 cos2 θ + q 4ω4 0 refers to the order q and H(1) ac sin2(ωt)). (43c) of perturbation theory in which the term appears. ( B2 − α2 + E2 Note that the upper index in, e.g., H(3) Out of the six terms in Eq. (41), H(0) q are sim- ply the projected parts of H0 and H1 ≡ HE +HB +HSO, respectively. H(2) contains a static and a time-dependent second-order energy shift, due to the spin-orbit inter- has action and the ac electric field, respectively. five terms. The first, second and fifth terms are spin- and time-dependent, hence these all contribute to the qubit dynamics. The third and fourth terms are static; they describe the spin-orbit-induced g-tensor renormal- ization. The fourth-order term H(4) of the qubit Hamil- tonian, being diagonal, does not influence spin dynam- ics. The static parts of the fifth-order term H(5) de- scribe higher-order g-tensor renormalisation (those pro- portional to α4 B), or nonlinear Zeeman splitting (those proportional to α2 B3). H(3) q q q q q tive Hamiltonian H(3) Already at this point, there are reasons to expect that in this EDSR model, a half-harmonic resonance occurs, and that the half-harmonic resonance frequency is driving-strength dependent: (i) The third-order effec- incorporates both longitudinal and transverse ac components, in analogy with the case of III. (ii) The the misaligned-field MR discussed in Sec. fifth-order effective Hamiltonian H(5) incorporates terms q proportional to E2 E2 ac(1 − cos 2ωt). The longitudinal static part ∝ E2 acσz can be interpreted as a drive-strength-dependent effective g-tensor renormalisa- tion, which contributes to the BSS, whereas the dynami- cal part ∝ E2 ac cos 2ωt σy is expected to drive Rabi oscil- lations at half-harmonic excitation, i.e., when 2ω ≈ B. (42) fulfills the expectation that no spin transition occurs if the external B-field and the spin-orbit field are aligned, ie, when θ = π/2. We note that the effective Hamiltonian in Eq. ac sin2 ωt = 1 2 C. Floquet perturbation theory for EDSR We apply Floquet perturbation theory, outlined in Sec. III C, to describe the fundamental and half-harmonic res- 9 res , as well as the Rabi frequencies Ω(1) onances. In particular, we derive the parameter depen- dence of the corresponding resonance frequencies ω(1) res and ω(2) res, at these two resonances, up to terms of the order of ∼ Bǫ4. There are two significant differences in the derivation of the EDSR results with respect to that of the MR results; we outline these differences in the following. res and Ω(2) (1) The MR Hamiltonian (16) has a driving term that is proportional to sin ωt. In contrast, the effective qubit Hamiltonian (41) we obtained for EDSR has cos ωt terms as well as second-harmonic terms proportional to cos 2ωt. In practice, the latter fact implies that the Floquet ma- trix will contain off-diagonal matrix elements that con- nect Floquet levels with next-nearest-neighbor Floquet quantum numbers. B, Bǫ2 and (2) In the EDSR case, we repeat the same second-order time-indepedent Schrieffer-Wolff transformation on the Floquet Hamiltonian F that we applied in Secs. III C 2 and III C 3. The Floquet Hamiltonian itself contains Bǫ4, since it is terms of the order of constructed from the effective qubit Hamiltonian that is itself the result of a finite-order perturbative calculation. When we separate the Floquet Hamiltonian to diagonal (F0) and off-diagonal (F1) components, and apply time- independent Schrieffer-Wolff transformation up to second order in F1, the resulting 2×2 effective Floquet Hamilto- nian will involve higher-order terms, up to Bǫ8. As our original Hamiltonian was accurate only up to the ∼ Bǫ4 terms, we drop the terms that are of higher order than ∼ Bǫ4 from the effective Floquet Hamiltonian. D. Analytical vs. numerical solution The results we obtain from Floquet perturbation the- ory are shown in Sec. II as Eqs. (8) -- (15). In the rest of this subsection we discuss these results and compare them to numerical results. The terms describing the fundamental resonance fre- (8) are interpreted as nominal Zeeman quency in Eq. splitting, g-tensor renormalisation, nonlinear Zeeman ef- fect, and BSS, respectively. We call the last term a BSS as it is a power-dependent correction to the resonance frequency, that is second order in the drive amplitude, hence analogous to the BSS in MR. Remarkably, the BSS in Eq. (8) is a negative correction, whereas the BSS in MR is always positive. The last term of the half- harmonic resonance frequency [Eq. (13)] is also inter- preted as a BSS. Further similarities with the MR case: (i) For the fundamental resonance, the BSS is smaller (∼ Bǫ4) than the power broadening, the latter being given by Ω(1) res ∼ Bǫ2. (ii) For the half-harmonic reso- nance, the BSS, being ∼ Bǫ4, is comparable to the power broadening, the latter being given by Ω(2) res ∼ Bǫ4. (iii) The BSS for both the fundamental and the half-harmonic resonance is proportional to cos2 θ, i.e., it vanishes in the limit of purely longitudinal excitation, and finite for  0.49 0.485 B~ / ω -h 0.48 0.475 0.47 ω(2) res 2π/Ω(2) res 1 0.8 0.6 0.4 0.2 0 0.4925 0.4924 0.4923 ) t , ω ( E P B~ / ω -h 0 2 6 4 ~ t (103 -h / B ) 8 10 FIG. 4. (color online) Electrically driven Rabi spin dynam- ics at half-harmonic resonance. The excited-state occupation probability PE is shown as a function of the drive frequency ω and time t; the numerical data reveals the chevron pattern characteristic of magnetic resonance. Parameters: θ = π/4, α/ B = Eac/ B = 1, ω0/ B = 5. The analytical results for the half-harmonic resonance frequency ω(2) res and the Rabi fre- quency Ω(2) res are also displayed. For the above parameter val- ues, the latter one is related to the time period of the oscilla- tion via 2π/Ω(2) res = 2π × 625 / B . purely transversal excitation. These features can be ex- plained by the argument provided in Sec. III D for the case of MR, applied to the effective qubit Hamiltonian (41). Regarding the results (8) and (13) for the resonance frequencies, we note that their ratio is exactly two in the limit of vanishing driving power, i.e., lim Eac→0(cid:16) ω(1) res(cid:17) = res ω(2) 2. We have checked that the result (8) for the fundamen- res matches the corresponding result tal Rabi frequency Ω(1) of Ref. 5; see Appendix B for details. The analytical results are tested against numerically exact solutions of the time-dependent Schrodinger equa- tion defined by the Hamiltonian H in Eq. (35). The nu- merical results were obtained using the truncated Hilbert space spanned by the 8 lowest-energy eigenstates of H0 + HB, corresponding to the 4 lowest-lying levels of the harmonic oscillator. We have checked that there was no visible change in the numerical results upon extending the Hilbert space with further, higher-lying orbitals. In Fig. 4, we plot the numerically computed time evolution of the occupation probability of the excited state Ei, for a finite range of the driving frequency in the vicinity of the 'nominal' half-harmonic resonance fre- quency ω/ B = 0.5 (see caption for parameter values). The analytical result (13) predicts complete Rabi oscil- lations at ω = ω(2) res = 0.4809 B/, and the Rabi fre- quency at this resonance is predicted by Eq. (15) to 10 ) c a E~ ω , ( x a m E P 1 0.8 0.6 0.4 0.2 0 0.4922 0 0.5 1.5 2 1 ~ ~ ac / B E FIG. 5. (color online) Bloch-Siegert shift and power broad- ening of the half-harmonic resonance. The maximal excited- state occupation probability P max is shown as a function of the amplitude Eac of the driving ac electric field and drive frequency ω. Parameters: θ = π/4 α/ B = 1, and ω0/ B = 8. The red line indicates the analytical result for the resonance frequency as the function of electric field based on Eq. (13). E 625 res ≈ 1 B be Ω(2)  . These predictions are in line with the numerical data shown in Fig. 4. For a finite detuning from the resonance frequency, the Rabi oscillations be- come faster and reduced (i.e., they do not reach PE = 1), leading to the characteristic chevron pattern4,21 known from MR. The results of Fig. 4 therefore reveal simple Rabi dynamics at the half-harmonic resonance. E The density plot of Fig. 5 is a visual demonstration of the BSS, i.e., of that the resonance frequency increases with increasing drive strength. The figure shows the maximum P max of the excited-state probability PE(t) within a time span exceeding the Rabi period at the half- harmonic resonance, as a function of the amplitude Eac of the ac electric field and the drive frequency ω. (See cap- tion for parameters.) Therefore, vertical cuts of the den- sity plot correspond to resonance curves. The solid line represents the analytical result (13) for the half-harmonic resonance frequency. The agreement between the ana- lytical curve and the P max E ≈ 1 ridge of the numerical simulation reassures the validity and correspondence of the two approaches. Importantly, in Fig. 5, the BSS is comparable in magnitude to the power broadening, which makes the BSS relatively easily resolvable in experiments realizing the model we use. A further question is how the BSS depends on the an- gle θ characterizing the direction of the spin-orbit inter- action. This dependence is exemplified by Fig. 6, which shows P max as a function of θ and the drive frequency. The latter is measured from half of the calculated fun- damental resonance frequency ω(1) (8). The solid line, showing good agreement with the centre of the bright P max E ≈ 1 region of the underlying density plot, res , see Eq. E  0.004 0.002 B~ / ) 2 / s e r ) 1 ( ω - ω ( -h 0 -0.002 0 π/8 3π/8 π/2 π/4 θ 1 0.8 0.6 0.4 0.2 0 , ) θ ω x a m ( E P E FIG. 6. (color online) Anisotropy of the half-harmonic res- onance. The maximal excited-state occupation probability P max is shown as a function of the angle θ characterizing the spin-orbit interaction and the drive frequency ω measured from ω(1) res /2, see Eq. (8). The solid line corresponds to the analytically obtained half-harmonic resonance frequency, i.e., it shows ω(2) 2 , see Eq. (13). Parameters: E/ B = 1, α/ B = 1, and ω0/ B = 5. res − ω (1) res shows the analytical result for the half-harmonic reso- nance frequency ω(2) res [Eq. (13)]. V. DISCUSSION AND CONCLUSIONS (1) We provide a numerical example to estimate orders of magnitudes of the EDSR resonance shifts and Rabi frequencies. Let us take B = 1.7 T with the electronic g factor 2, yielding B ≈ 0.1 meV. We set α = 0.1 meV and Eac = 0.1 meV, and the orbital level spacing is chosen to be ω0 = 1 meV. Then the order of magnitude of the Rabi frequency at the fundamental resonance becomes Ω(1) res ∼ Bǫ2/ ≈ 1.5 × 109 1 s corresponding to a spin- flip time of ≈ 4.3 ns. For the half-harmonic resonance, Ω(2) res ∼ Bǫ4/ ≈ 1.5 × 107 1 s , corresponding to a spin- flip time of ≈ 430 ns. For both resonances, the BSS is comparable to the value of Ω(2) (2) The results presented in this work describe a per- turbative regime where spin-orbit interaction is assumed to be 'weak', in the sense that the spin-orbit energy scale in the QD is dominated by the QD level spacing, α ≪ ω0. In nanowire QD host materials such as InAs25 and InSb17, spin-orbit interaction is known to be 'strong' in the sense that it creates a strong g-factor renormali- sation, already in the bulk materials. A question arising from these facts is: are typical InAs and InSb nanowire QDs within the range of validity of our perturbative the- ory? One way to answer this question is via a compar- ison of the dependence of the fundamental EDSR res- res estimated above. 11 onance frequency obtained from the perturbative the- ory and from experiments. The experiments17,25 have found that the fundamental resonance frequency shows a similar angular dependence as the perturbative result (8), i.e., for a magnetic field with a fixed magni- Eq. tude, the resonance frequency is maximal if the magnetic field is aligned along a certain direction and minimal if it is aligned perpendicular to that direction. To be spe- cific, we take the data given in the first row of Table I. of Ref. 25, which indicates that the ratio of the mini- mal and maximal resonance frequencies in the considered case were ≈ 0.84. Using the first two terms in Eq. (8), we can identify that ratio with 1 − 2 α2/2ω2 0, yielding α/ω0 ≈ 0.28 for this particular InAs device. A similar analysis of the experimental data in Fig. 3c of Ref. 17 results in an estimate α/ω0 ≈ 0.37 for the measured InSb device. These estimates suggest that the InAs and InSb QDs are on the border between 'weak' and 'strong' spin-orbit interaction. (3) To our knowledge, three experiments have re- ported subharmonic EDSR resonances in semiconductor nanowire QDs, where our model based on the Rashba- type spin-orbit interaction could be appropriate to de- scribe the spin dynamics. The strong subharmonic reso- nances reported in Stehlik et al.26 are described by a the- ory developed for strongly driven double quantum dots34. Faint half-harmonic resonances are visible in the data of Refs. 25 (see Fig. 2b therein) and 17 (see Fig. 2b therein). A quantitative experimental analysis exploring the parameter dependencies of the corresponding reso- nance and Rabi frequencies would allow for a comparison with our predictions. (4) One of our conclusions was that the BSS of the fundamental EDSR resonance frequency has an anoma- lous, negative sign, see Eq. (8). Here, we provide a simple physical picture explaining this result, using the unitary transformation applied in Ref. 42. For simplic- ity, we focus on the case when the spin-orbit field is perpendicular to the magnetic field, i.e., θ = 0. Then, the unitary transformation S of Ref. 42 (not to be con- fused with the generator of the Schrieffer-Wolff transfor- mation in Appendix A) applied on our static Hamilto- nian H0 + HB + HSO eliminates the spin-orbit term and transforms the homogeneous magnetic field HB to an in- homogeneous, spiral-like magnetic field, H ′ B ≡ SHBS† ∝ B [σz cos(z/ξ) − σx sin(z/ξ)], where ξ ∝ 1/ α is the spin- orbit length [see Eq. (2) of Ref. 42]. The driving elec- tric field, incorporated in our model as HE, induces a spatial oscillation z(t) = −A sin ωt of the electron's cen- tre of mass with an amplitude A ∝ Eac. Inserting this time-dependent z(t) to the above expression for H ′ B, and expanding the terms up to second order in A/ξ, we find B(t) ∝ Bhσz(cid:16)1 − A2 H ′ the time-averaged z component of the time-dependent magnetic field in H ′ B(t) acquires a correction propor- tional to − B A2 ξ2 ∝ − B E2 ac α2. Notice that this correction is negative and has the same parameter dependence as ξ sin ωti. That is, ξ2 sin2 ωt(cid:17) + σx A 12 the BSS in, the last term of, Eq. (8). to the BSS in MR. (5) To our knowledge, BSS has not yet been experimen- tally or theoretically analyzed in the context of EDSR. However, we wish to point out that certain numerical results in Ref. 10, related to EDSR in a double quan- tum dot, are reminiscent of the BSS. Figures 4a and 4b in Ref. 10 show spin Rabi oscillations for different drive strengths. Therein, the drive strength is characterised by the dimensionless quantity f . In Figs. 4a and 4b of Ref. 10 it is shown that the complete Rabi oscillations at the fundamental resonance become incomplete upon increasing the drive strength from f = 0.02 to f = 0.15, while the driving frequency is maintained. This phe- nomenology is reminiscent of the effect of BSS: when the drive strength is increased, BSS provides a shift of the resonance frequency, hence a fixed drive frequency be- comes off-resonant, and the Rabi-oscillation amplitude decreases. It is therefore tempting to interpret these re- sults as consequences of BSS. However, the phenomenol- ogy of BSS would imply that (i) upon further increase of the drive strength, e.g., at f = 0.35, the amplitude of the Rabi oscillation further decreases, and (ii) the Rabi os- cillation speeds up gradually as f is increased from 0.02 to 0.15 and to 0.35. The results shown in Figs. 4a and 4b of Ref. 10 disagree with these expectations, hence we conclude that the BSS phenomenology is insufficient to describe the numerical results of Ref. 10. Importantly, Ref. 10 considers parameter settings where the undriven system consists of four, approximately equidistant levels (see their Fig. 2a), which is a key difference with respect to the effectively two-level setup considered in our present work, and can be responsible for the phenomenology de- viating from that of the BSS. (6) Even though EDSR experiments can be performed on single QDs13,21, many current experiments use the Pauli blockade setup for initialization and readout3. The latter setup consists of a double QD which is occupied by two electrons (in the simplest case) during EDSR. The in- herent anharmonicity of the double QD confinement po- tential, as well as the presence of the Coulomb interaction between the two (or more) electrons, can provide alter- native nonlinear EDSR mechanisms7,10,24,26,29,34, which compete with those presented in our work focusing on harmonic confinement and single-electron dynamics. For example, an apparently well understood34 case when the two-electron and double-QD features dominate the sub- harmonic EDSR resonances is the experiment of Ref. 26. In conclusion, we have studied the characteristics of EDSR in a 1D QD model with parabolic confinement, homogeneous Rashba spin-orbit interaction and homoge- neous driving electric field. We demonstrated the exis- tence of subharmonic (multi-photon) resonances in this model, and analysed the half-harmonic (two-photon) res- onance in detail. We have analytically described the pa- rameter dependence of the fundamental resonance fre- quency and the half-harmonic resonance frequency, and demonstrated that these resonance frequencies increase with increasing drive strength. This effect is analogous Our results describe a perturbative regime, where the orbital level spacing of the QD dominates the energy scales of the external magnetic field, spin-orbit interac- tion, and electrical drive. Therefore our results have di- rect experimental relevance for QDs with weak spin-orbit interaction. They can also serve as benchmarks for nu- merical studies departing from the perturbative regime. The model used here contains only minimal ingredients necessary to describe EDSR, suggesting that the subhar- monic resonances and the BSS discussed here are generic features of electrically driven spin dynamics. Note added: Upon completing this work, we became aware of a related experiment43 revealing half-harmonic EDSR in a single QD, mediated by an inhomogeneous magnetic field. ACKNOWLEDGMENTS We thank P. Scarlino and G. Sz´echenyi for useful dis- cussions. We acknowledge funding from the EU Marie Curie Career Integration Grant CIG-293834 (Carbon- Qubits), the OTKA Grants PD 100373 and 106047, and the EU ERC Starting Grant CooPairEnt 258789. GB acknowledges funding from the Deutsche Forschungsge- meinschaft (DFG) within SFB767 and from the EU Marie Curie ITN S3NANO. AP is supported by the J´anos Bolyai Scholarship of the Hungarian Academy of Sci- ences. Appendix A: Time-dependent Schrieffer-Wolff perturbation theory Here we introduce the time-dependent Schrieffer-Wolff perturbation theory (TDSW), the method we use to de- rive the effective 2 × 2 time-dependent Hamiltonian (41) governing the dynamics of the qubit. Let us first recall the basic idea of standard time- independent Schrieffer-Wolff (SW) perturbation the- ory.39,40 We consider a Hamiltonian H = H0 + H′ where H0 is diagonal and H′ is the perturbation. Furthermore, the basis states of H can be divided into relevant A and irrelevant B subspaces that have well separated energy scales. A and B are weakly interacting, i.e. the matrix elements connecting them are small compared to the en- ergy separation of the two subspaces. Ideally, we can introduce a unitary transformation e−S that brings H into a block diagonal form H = e−SHeS where the rele- vant and irrelevant subspaces are separated as illustrated in Fig. 7. However, in most of the cases we don't know the explicit form of the transformation e−S so we have to construct it bit-by-bit until the elements connecting the two subsets vanish up to the desired order of per- turbation. This is usually done by expanding e−S in a series and constructing the terms of different orders suc- cessively. A ϵ A 0 ϵ B 0 B 0 A 0 0 0 B A' 0 0 B' 0 ϵ ϵ 0 FIG. 7. Schematic form of the matrices we encounter during the SW perturbation theory. The label ǫ refers to blocks with matrix elements that are much smaller than the energy difference between the relevant and irrelevant subspaces. In the EDSR problem, the energy scale of the those blocks is B ∼ ǫω0. A great advantage of SW with respect to conventional perturbation theory is that here we don't need to distin- guish between the degenerate and non-degenerate cases. Now we introduce the time-dependent SW pertur- bation theory as a natural extension of the time- independent case. Similar approaches have been applied for particular problems in Refs. 44 -- 46; here, we provide a general description of the method, which we utilised in the main text for deriving the effective qubit Hamiltonian (41) of the EDSR problem. Consider the time-dependent Hamiltonian H(t) = H0 + H′(t), where the perturbation is divided into a block-diagonal and block-off-diagonal part H′(t) = H1 + H2 as shown in Fig 7. IV A), H1 = HB and H2(t) = HE(t) + HSO. Note that there H1 happens to In our problem (see Sec. 13 be a time-independent perturbation, but the treatment outlined here is readily applicable to a time-dependent block-diagonal perturbation as well. Similarly to the SW we successively build the unitary transformation U (t) = e−S(t) that separates the sub- spaces A and B, but here the matrix S(t) is now time- dependent. Note that any unitary transformation can be written in this form, and the matrix S(t) should be anti-Hermitian to ensure the unitary character of U (t). The matrix S(t) is chosen to be block-off-diagonal (see Fig 7). Note also that because of the weakness of the inter-subspace coupling, the unitary tranformation U (t) is close to unity, and hence S(t) is small and can be expressed as a power series with respect the perturbing terms. The transformation of time-dependent Schrodinger equation −i ∂ ∂t ψ(t)+H(t)ψ(t) = 0 with the above U (t) is canonical, i.e., it preserves the form of the time evolution equation. The transformed wave function and Hamilto- nian read as: ψ(t) = e−S(t)ψ(t), (A1) H(t) = e−S(t)H(t)eS(t) + i ∂e−S(t) ∂t eS(t). (A2) From now on, we might suppress the time argument and denote time derivatives such as ∂ ∂t ψ as ψ. Starting from Eq. (A2), we utilize the power series of the exponential function. The second term in Eq. (A2) is the heart of the time -- dependent SW transformation; in the time -- independent case this term vanishes as S is time-independent. The expansion of the first term in Eq. (A2) is known from SW formalism, therefore we do not discuss it here. The explicit form of the second term, after expanding the exponential function, has the follow- ing form: ∂e−S ∂t eS =(cid:20) ∂ ∂t(cid:18)−S + ×(I +S + 1 2! 1 2! = −[ S, S](0) − S3 + . . .(cid:19)(cid:21) (I +S + 1 S2− 3! 1 S3+. . . ) = (− S + S2 + 3! [ S, S](1) − 1 2! 1 S S− 2! [ S, S](2) ··· = − 1 2! 1 2! 1 3! S2+ S3+ . . .) = (− S + SS + SS2+ 1 2! S SS− [ S, S](j) . 1 3 1 3! 1 3! SS− Xj=0 ∞ 1 3! 1 (j + 1)! 1 3! 1 2! S S− S2 S + . . .) SS2− 1 3! S SS− 1 3! S2 S +. . . ) × (A3) The transformed Hamiltonian then equals to H = ∞ Xj=0 1 j! [H, S](j) − i 1 (j + 1)! [ S, S](j) . (A4) ∞ Xj=0 with [H, S](n+1) = h[H, S](n) , Si and [H, S](0) = H. Note that the second term in (A4) is new with respect to time-independent SW, and it is a consequence of the time dependence of the Hamiltonian and therefore that of the matrix S. Considering a time-independent Hamil- tonian the second term vanishes and we are left with the well-known SW transformation. We now exploit the block-off-diagonal property of S in order to separate the block-off-diagonal and block- diagonal parts of the transformed Hamiltonian: Hoff-diag = (2j + 1)! [H0 + H1, S](2j+1) 1 1 (2j)! [H2, S](2j) 1 (2j + 1)! [ S, S](2j) , (A5) ∞ ∞ + Xj=0 Xj=0 Xj=0 − i ∞ Hdiag = 1 (2j)! 1 ∞ ∞ + Xj=0 Xj=0 Xj=0 − i ∞ [H0 + H1, S](2j) ... 1 3 − [[H2, S2] , S1] + i S3 , (A9d) From the order-by-order expansion of Eq. (A7), we ob- tain the following hierarchy of simple algebraic equations for the Sj matrices: 14 (A9a) [H0, S1] = −H2 , [H0, S2] = − [H1, S1] + i S1 , [H0, S3] = − [H1, S2] − [H0, S4] = − [H1, S3] − 1 3 1 3 (A9b) [H2, S1](2) + i S2 , (A9c) [[H2, S1] , S2] Once the first equation (A9a) is solved for S1(t), the solution can be inserted to (A9a) which then forms an algebraic equation for S2(t), etc. Note that since we work in the eigenbasis of H0, the above procedure simplifies to subsequently solving single-variable linear equations, which is a trivial analytical task, well suited for symbolic computation. j=0 H(n), where After obtaining the Sj matrices and inserting them into Eq. (A6), we have an order-by-order expansion H = Hdiag =P∞ H(0) = H0 H(1) = H1 H(2) = [H2, S1] + H(3) = [H2, S2] + [[H0, S1] , S2] (A10b) (A10c) (A10a) 1 2 1 2 1 2 [H0, S1](2) [H1, S1](2) + 1 2 [[H0, S2] , S1] − i [ S1, S1] (A10d) + 1 2 ... With the use of Eqs. (A9) we can further simplify Eqs. (A10): (A11a) (A11b) (A11c) (A11d) (A11e) (A11f) [H2, S1] [H2, S2] [H2, S3] − [H2, S4] − 1 [[[H2, S1] , S2] , S1] − 4! 1 4! 1 4! [H2, S1](3) [[[H2, S1] , S1] , S2] 1 4! [[[H2, S2] , S1] , S1] H(0) = H0 H(1) = H1 1 H(2) = 2! 1 H(3) = 2! 1 H(4) = 2! 1 H(5) = 2! − ... (2j + 1)! [H2, S](2j+1) 1 (2j + 2)! [ S, S](2j+1) . (A6) Then, S is determined by solving Hoff-diag = 0 . (A7) The effective (now block-diagonal) Hamiltonian becomes H = Hdiag. Note that H as well as the term 'effective Hamiltonian' is also used to describe the block of H cor- responding to the relevant subpsace. So far no approximation has been made; now we make use of the smallness of the perturbation. Following the approach of time-independent SW perturbation theory, we aim at solving Eq. (A7) via expanding S as a power series in the perturbation, S = S1 + S2 + S3 + . . . , (A8) where Sj represents an operator of jth order in the per- turbation. Recall that in TDSW, S is time dependent, and its time derivative appears in its defining equation (A7) as well as in the effective Hamiltonian (A6). There- fore, to separate the terms of different order in perturbing parameter in Eq. (A7), it is necessary to make an a pri- ori assumption on the frequency scale characterizing the magnitude of Sj. As the drive frequency is ω, expectedly the frequency characterizing the time evolution of all Sj- s will be ∼ ω, hence we assume Sj ∼ ωSj. In the EDSR IV A, the relevant subspace is problem defined in Sec. the subspace of the ground-state orbital spanned by 0↑i and 0↓i. Furthermore, the energy scales of the drive fre- quency, drive strength, Zeeman splitting and spin-orbit coupling are much lower than the splitting between the oscillator levels ∼ ω0, and all of them are treated as per- turbation. This implies that Sj is of the order of (j + 1) in perturbation. Obviously, after solving Eq. (A7) with this assump- tion, we need to check if the obtained Sj functions are consistent with our assumption above. 15 Finally, we need to check the consistency of our as- sumption for the time evolution of Sj with the actual so- lution we obtained for Sj using that assumption. From (A9a), S1 inherits harmonic time-dependence from H2 with frequency ω. This implies that the time derivative is S1 ∼ ωS1, as assumed. From Eq. (A9b), the matrix S2 might contain frequency components at ω, as well as at zero frequency and 2ω (if H1 is time-dependent with frequency ω); nevertheless, the S2 ∼ ωS2 relation still holds, etc. In conclusion, TDSW allows for obtaining an effective time-dependent Hamiltonian for the relevant subspace. The procedure is to evaluate the transformation matrices Sj up to the desired order via solving Eq. (A9), and substituting the resulting Sj matrices into Eq. (A11). y − z plane, and the E-field is along the x axis. Further- more, we project the Hamiltonian on the y-ground-state orbital of the harmonic oscillator, yielding HGBL = p2 x 2m + 1 2 mω2 0x2 + αpxσy + 1 2 g∗µB B · σ + eEacx sin ωt (B1) For simplicity, we focus on the special case B = (0, 0, B) from now on. Then, the Hamiltonian in Eq. (B1) is equivalent to our Hamiltonian H at θ = 0. To deduce the Rabi frequency calculated by GBL for the special case above, we start from their Eqs. (13) and (14), where they provide the time-dependent part of the effective qubit Hamiltonian as Appendix B: Rabi frequency of the fundamental resonance: relation to the results of Ref. 5 where HGBL = 1 2 h(t) · σ, (B2) (B3) EDSR in a QD in a two-dimensional electron gas due to Rashba and Dresselhaus spin-orbit interactions has been described by Golovach, Borhani and Loss (GBL) in Ref. 5. Therein, the Rabi frequency of the fundamental reso- nance as a function of system parameters (magnetic field strength, magnetic field direction, spin-orbit interaction strengths and ac electric field amplitude and direction) has been calculated. Even though the dimensionality and the spin-orbit Hamiltonian in the model of GBL differs from our model, the calculated Rabi frequencies can be compared after a special case of the model of GBL has been reduced to one dimension. Here we show that after this dimension reduction our result for the fundamental Rabi frequency equals that of GBL. In the model of GBL, the 2DEG lies in the x-y plane. We consider the special case when the confinement po- tential is parabolic and has a cylindrical symmetry, the Dresselhaus coupling vanishes, β = 0, the B-field is in the 1 F. Bloch and A. Siegert, Phys. Rev. 57, 522 (1940). 2 J. H. Shirley, Phys. Rev. 138, B979 (1965). 3 F. H. L. Koppens, C. Buizert, K. J. Tielrooij, I. T. Vink, and K. C. Nowack, T. Meunier, L. P. Kouwenhoven, L. M. K. Vandersypen, Nature 442, 766 (2006). 4 M. Veldhorst, J. C. C. Hwang, C. H. Yang, A. W. Leenstra, B. de Ronde, J. P. Dehollain, J. T. Muhonen, F. E. Hudson, K. M. Itoh, A. Morello, and A. S. Dzurak, ArXiv e-prints (2014), arXiv:1407.1950 [cond-mat.mes-hall]. 5 V. N. Golovach, M. Borhani, Phys. Rev. B 74, 165319 (2006). 6 C. Flindt, A. S. Sørensen, and D. Loss, and K. Flensberg, Phys. Rev. Lett. 97, 240501 (2006). 7 Y. Tokura, W. G. van der Wiel, T. Obata, and S. Tarucha, Phys. Rev. Lett. 96, 047202 (2006). 8 E. I. Rashba, Phys. Rev. B 78, 195302 (2008). 9 J. D. Walls, Phys. Rev. B 76, 195307 (2007). 10 D. V. Khomitsky, L. V. Gulyaev, and E. Y. Sherman, A straightforward calculation shows that h(t) = 2µBB × Ω(t). αeEacg∗µBB 1 2 h(t) = sin(ωt)ex (B4) = 2 sin(ωt)ex (B5) ω2 0 α Eac B 2ω2 0 Note that this effective ac magnetic field is perpendicular to the static magnetic field, which is applied in the z direction. The Rabi frequency due to this ac magnetic field at the fundamental resonance frequency reads Ω(1) res,GBL = 2 α Eac B 2ω2 0 , (B6) which is identical to our result in Eq. (8), if the latter is evaluated at θ = 0 and terms above third order are dropped. Phys. Rev. B 85, 125312 (2012). 11 R. Li, J. Q. You, C. P. Sun, Phys. Rev. Lett. 111, 086805 (2013). and F. Nori, 12 Y. Kato, R. C. Myers, D. C. Driscoll, A. C. Gossard, J. Levy, and D. D. Awschalom, Science 299, 1201 (2003). 13 K. C. Nowack, F. H. L. Koppens, Y. V. Nazarov, and L. M. K. Vandersypen, Science 318, 1430 (2007). 14 M. Pioro-Ladriere, T. Obata, Y. Tokura, Y.-S. Shin, and S. Tarucha, T. Kubo, K. Yoshida, T. Taniyama, Nat. Phys. 4, 776 (2008). 15 E. A. Laird, C. Barthel, E. Marcus, M. P. Hanson, Phys. Rev. Lett. 99, 246601 (2007). I. Rashba, C. M. and A. C. Gossard, 16 S. Nadj-Perge, S. M. Frolov, E. P. A. M. Bakkers, and L. P. Kouwenhoven, Nature 468, 1084 (2010). 17 S. Nadj-Perge, V. S. Pribiag, J. W. G. van den S. R. Plissard, E. P. A. M. and L. P. Kouwenhoven, Berg, K. Zuo, Bakkers, S. M. Frolov, 32 G. Sz´echenyi Phys. Rev. B 88, 165302 (2013). and Phys. Rev. B 89, 115409 (2014). 33 E. N. Osika, A. Mre´nca, Phys. Rev. B 90, 125302 (2014). 16 A. P´alyi, and B. Szafran, Phys. Rev. Lett. 108, 166801 (2012). 18 F. Pei, E. A. Laird, G. A. Steele, and L. P. Kouwenhoven, Nat. Nanotech. 7, 630 (2012). 19 V. S. Pribiag, S. Nadj-Perge, S. M. Frolov, J. W. G. S. R. Plissard, and L. P. Kouwenhoven, I. van Weperen, van den Berg, E. P. A. M. Bakkers, Nat Nano 8, 170 (2013). 20 E. A. Laird, F. Pei, Nat. Nanotech. 8, 565 (2013). 21 E. Kawakami, P. Scarlino, D. R. Ward, F. R. Braakman, D. E. Savage, M. G. Lagally, M. Friesen, S. N. Cop- persmith, M. A. Eriksson, and L. M. K. Vandersypen, Nat. Nanotech 9, 666 (2014). 22 P. V. Klimov, A. L. Falk, B. B. Buckley, and D. D. Awschalom, Phys. Rev. Lett. 112, 087601 (2014). 23 F. Forster, M. Muhlbacher, D. Schuh, W. Wegscheider, and S. Ludwig, Phys. Rev. B 91, 195417 (2015). 24 E. A. Laird, C. Barthel, E. I. Rashba, C. M. Marcus, M. P. Hanson, and A. C. Gossard, Semiconductor Science and Technology 24, 064004 (2009). 25 M. D. Schroer, K. D. Petersson, M. Jung, and J. R. Petta, Phys. Rev. Lett. 107, 176811 (2011). 26 J. Stehlik, M. D. Schroer, M. Z. Maialle, M. H. Degani, and J. R. Petta, Phys. Rev. Lett. 112, 227601 (2014). 27 A. De, C. E. Pryor, and M. E. Flatt´e, Phys. Rev. Lett. 102, 017603 (2009). 28 J. Pingenot, C. E. Pryor, and M. E. Flatt´e, Phys. Rev. B 84, 195403 (2011). 29 E. I. Rashba, Phys. Rev. B 84, 241305 (2011). 30 M. P. Nowak, B. Szafran, and F. M. Peeters, Phys. Rev. B 86, 125428 (2012). 31 E. N. Osika, B. Szafran, and M. P. Nowak, and L. P. Kouwenhoven, 35 M. Trif, V. N. Golovach, and D. Loss, Phys. Rev. Lett. 113, 247002 (2014). 34 J. Danon and M. S. Rudner, Phys. Rev. B 77, 045434 (2008). 36 T. Dittrich, Quantum Transport and Dissipation (Wiley- WCH, 1998). 37 P. K. Aravind and J. O. Hirschfelder, The Journal of Physical Chemistry 88, 4788 (1984). 38 I. Gromov and A. Schweiger, J. Magn. Res. 146, 110 (2000). 39 J. R. Schrieffer and P. A. Wolff, Phys. Rev. 149, 491 (1966). 40 R. Winkler, Spin-orbit Coupling Effects in Two-Dimensional Electron and Hole Systems, Springer Tracts in Modern Physics, Vol. 191 (Springer, 2003). 41 C. Wei, A. S. M. Windsor, and N. B. Manson, J. Phys. B: At. Mol. Opt. Phys. 30, 4877 (1997). 42 L. S. Levitov and E. I. Rashba, Phys. Rev. B 67, 115324 (2003). 43 P. Scarlino, E. Kawakami, D. R. Ward, D. E. Savage, M. G. Lagally, M. Friesen, S. N. Coppersmith, M. A. Eriks- son, and L. M. K. Vandersypen, ArXiv:1504.06436 (un- published). 44 F. Schwabl, Advanced Quantum Mechanics, Chapter 9. (Springer, 2008). 45 A. Kaminski, Y. V. Nazarov, and L. I. Glazman, Phys. Rev. B 62, 8154 (2000). 46 Y. Goldin and Y. Avishai, Phys. Rev. B 61, 16750 (2000). 

1710.10909

1

1710

Spin-transfer Antiferromagnetic Resonance

Currents can induce spin excitations in antiferromagnets, even when they are insulating. We investigate how spin transfer can cause antiferromagnetic resonance in bilayers and trilayers that consist of one antiferromagnetic insulator and one or two metals. An ac voltage applied to the metal generates a spin Hall current that drives the magnetic moments in the antiferromagnet. We consider excitation of the macrospin mode and of transverse standing-spin-wave modes. By solving the Landau-Lifshitz-Gilbert equation in the antiferromagnetic insulator and the spin-diffusion equation in the normal metal, we derive analytical expressions for the spin-Hall-magnetoresistance and spin-pumping inverse-spin-Hall dc voltages. In bilayers, the two contributions compensate each other and cannot easily be distinguished. We present numerical results for a MnF$_2|$Pt bilayer. Trilayers facilitate separation of the spin-Hall-magnetoresistance and spin-pumping voltages, thereby revealing more information about the spin excitations. We also compute the decay of the pumped spin current through the antiferromagnetic layer as a function of frequency and the thickness of the antiferromagnetic layer.

cond-mat.mes-hall

cond-mat

Spin-transfer Antiferromagnetic Resonance Øyvind Johansen, Hans Skarsvag, and Arne Brataas Center for Quantum Spintronics, Department of Physics, Norwegian University of Science and Technology, NO-7491 Trondheim, Norway (Dated: October 31, 2017) Currents can induce spin excitations in antiferromagnets, even when they are insulating. We investigate how spin transfer can cause antiferromagnetic resonance in bilayers and trilayers that consist of one antiferromagnetic insulator and one or two metals. An ac voltage applied to the metal generates a spin Hall current that drives the magnetic moments in the antiferromagnet. We consider excitation of the macrospin mode and of transverse standing-spin-wave modes. By solv- ing the Landau–Lifshitz–Gilbert equation in the antiferromagnetic insulator and the spin-diffusion equation in the normal metal, we derive analytical expressions for the spin-Hall-magnetoresistance and spin-pumping inverse-spin-Hall dc voltages. In bilayers, the two contributions compensate each other and cannot easily be distinguished. We present numerical results for a MnF2Pt bilayer. Tri- layers facilitate separation of the spin-Hall-magnetoresistance and spin-pumping voltages, thereby revealing more information about the spin excitations. We also compute the decay of the pumped spin current through the antiferromagnetic layer as a function of frequency and the thickness of the antiferromagnetic layer. I. INTRODUCTION Antiferromagnets have many qualities that make them attractive for use in spintronic devices. For example, the absence of stray fields allows for a dense storage of com- ponents without undesired crosstalk between the active elements. The most interesting feature of antiferromag- nets is that their high resonance frequencies pave the way toward terahertz circuits [1]. Current-induced spin-transfer torques (STTs) can in- duce ferromagnetic resonance [2] in both metallic and insulating ferromagnets [3, 4]. Antidamping-like STT is even under magnetization reversal [5]. Consequently, the magnetic moments in the two sublattices of a collinear antiferromagnetic insulator experience the same STT, which enables STT-driven spin dynamics in antiferro- magnets. Current-induced STT is a powerful method for probing the magnetization dynamics in magnetic layers [6, 7]. An electric signal can simultaneously drive and de- tect the magnetization dynamics. An ac voltage leads to an alternating spin current through the spin Hall effect [8], which drives the magnetic moments at resonance. Subsequently, the spin Hall magnetoresistance (SMR) and spin pumping (SP) induce dc voltages through the inverse spin Hall effect (ISHE) [9] that can be measured using a bias tee. SMR [10, 11] is the dependence of the normal metal resistance on the orientation of the magnetic moments in an adjacent magnetic layer relative to the applied cur- rent. When the magnetic moments precess, the resis- tance of the metal correspondingly oscillates. The mix- ing of the oscillating resistance and charge current gener- ates a dc voltage bias that can provide insights into the magnetization dynamics. Recent experiments have indi- cated that SMR also occurs in antiferromagnetic insula- tor/normal metal (AFN) bilayers [12–16]. Theoretical predictions have also been made for conducting antifer- romagnets [17]. Similar to ferromagnets, SP is also active in antifer- romagnets [18]. However, to the best of our knowledge, there are no direct experimental detections of antiferro- magnetic SP. The lack of direct experimental signatures is possibly due to the high resonance frequencies and low susceptibilities of the magnetic moments in antifer- romagnets, which make experimental detection challeng- ing. However, we have recently theoretically shown that the susceptibilities and thus the dc SP substantially in- crease near the spin-flop transition, where the resonance frequency is low [19]. Therefore, we expect that antifer- romagnetic SP will be a prominent effect if we drive our system close to the spin-flop transition. In this paper, we compute the dc voltages resulting from SMR and SP in antiferromagnetic insulator/normal metal bilayers. The driving source is an ac voltage bias on the normal metal. In addition to the macrospin mode, we also consider the excitation of transverse standing spin waves. These standing waves have a higher resonance fre- quency than that of the uniform precession modes, and these waves can be excited by tuning the frequency of the applied voltage bias. The detection of such waves would reveal a wide variety of properties of the antiferromag- netic material. The resonance frequencies can be used to determine contributions to the free energy of the antifer- romagnet, such as exchange and anisotropy frequencies, and the exchange lengths of the sublattices. The am- plitudes and linewidths of the resonance peaks can also be used to determine both the intrinsic and SP-induced damping and thus the transverse spin conductance of the AFN interface. Finally, we also show how the SMR and SP dc voltages can be separated by sandwiching the anti- ferromagnetic material between two metals and measur- ing the dc biases in the metals independently. This ap- proach requires that the dissipation of the pumped spin current through the antiferromagnet is negligible. We therefore study for what thicknesses of the antiferromag- netic layer and for what resonance frequencies this is a 7 1 0 2 t c O 0 3 ] l l a h - s e m . t a m - d n o c [ 1 v 9 0 9 0 1 . 0 1 7 1 : v i X r a valid assumption. A. Equations of motion 2 II. MODEL We first consider a bilayer that consists of an antifer- romagnetic insulator in contact with a heavy metal, as shown in Fig. 1. An ac voltage applied to the metal causes spin excitations in the antiferromagnet via the spin Hall effect and the STT. Subsequently, SP in com- bination with the ISHE and SMR cause a dc voltage in the normal metal. For the bilayer system, our main result is the analyti- cal expressions for the dc voltages resulting from SP and SMR in Eq. (31). These expressions hold for a uniaxial antiferromagnet under the influence of an external mag- netic field that can control the resonance frequency and enhance the voltage signals [19]. For the NAFN trilayer, our main contribution is illustrating how this system can be used to measure the SP and SMR voltages indepen- dently. These contributions cannot be distinguished in the bilayer system because they have the same frequency dependence. N AF x(cid:48) x n(t) θ H0 z(cid:48) z m(t) Ic(t) Vac y y = dN y = 0 y = −dAF FIG. 1. An ac voltage applied to a normal metal with strong spin-orbit coupling generates spin currents that flow into the antiferromagnetic insulator, exciting the magnetic moments. The direction of the applied voltage and the easy axis of the antiferromagnet are parallel to the AFN interface, and there is an angle of θ between them. An external field H0 along the easy axis of the antiferromagnet controls the resonance frequency and the magnetic susceptibilities. The sublattice magnetizations of the antiferromagnetic insulator are M1 and M2. We describe the dynamics of these magnetizations in terms of the dimensionless aver- age magnetization and N´eel order parameter vectors m and n, which are defined as Lm = (M1 + M2)/2 and Ln = (M1 − M2)/2, where L is the saturation mag- netization of each sublattice. These vectors satisfy the constraints m2 + n2 = 1 and m · n = 0. The cou- pled equations of motion for m and n are given by the Landau–Lifshitz–Gilbert (LLG) equations m = n = 1 2 1 2 (ωm × m + ωn × n) + τ GD m + τ SP m + τ STT m , (1a) (ωm × n + ωn × m) + τ GD n + τ SP n + τ STT n . (1b) In the LLG equation (1), the Gilbert damping torques are m = α0 (m × m + n × n) , τ GD n = α0 (m × n + n × m) , τ GD the interfacial SP torques are m = α(cid:48)dAFδ(y) (m × m + n × n) , τ SP n = α(cid:48)dAFδ(y) (m × n + n × m) , τ SP and the STTs are τ STT m = α(cid:48)  dAFδ(y) τ STT n = α(cid:48)  dAFδ(y) (cid:104) m ×(cid:0)m × µN s (y, t)(cid:1) s (y, t)(cid:1)(cid:105) + n ×(cid:0)n × µN (cid:104) s (y, t)(cid:1) m ×(cid:0)n × µN s (y, t)(cid:1)(cid:105) + n ×(cid:0)m × µN , (2a) (2b) (3a) (3b) (4a) . (4b) Here, we have introduced the SP-induced enhanced damping parameter α(cid:48) = γg⊥/(4πLAdAF), and α0 is the intrinsic Gilbert damping parameter. A is the AFN interface area. The STTs depend on the spin accumula- tion µN The frequencies ωm,n corresponding to the effective fields are ωm = −(γ/L) · δf /δm and ωn = −(γ/L) · δf /δn, where γ is the gyromagnetic ratio and f is the free energy density, s in the normal metal. (cid:104) −ω(cid:107)(cid:0)m2 ωE L γ f = (cid:0)m2 − n2(cid:1) − 2ωH mx x − (λm∇m)2 − (λn∇n)2(cid:1)(cid:105) x + n2 Here, ωE is the exchange frequency, ω(cid:107) is the easy-axis anisotropy frequency, ωH is the frequency that describes . (5) the external magnetic field along the easy axis, and λm,n is the exchange length for m and n. B. Spin accumulation 3 We will compute the induced dc voltages to the second order in the spin excitations. For this purpose, comput- ing the spin excitations to the first order in their devia- tions from equilibrium is sufficient. For simplicity, we as- sume an ideal compensated antiferromagnetic insulator- metal interface. In this case, we can only excite stand- ing waves in the transversal direction, along the interface normal. Impurities, an uneven interface, or a sufficiently high temperature can also facilitate the excitations of waves in other directions. Within our assumptions, we linearize the LLG equations and use a harmonic transver- sal standing-wave ansatz of the solutions: (cid:0)δm(y)eiωt + δm∗(y)e−iωt(cid:1) , (cid:0)δn(y)eiωt + δn∗(y)e−iωt(cid:1) , m(y, t) = 1 2 n(y, t) = n0 + 1 2 where n0 = x and δm(y) = δmy cos[ky + δmz cos[kz δn(y) = δny cos[ky + δnz cos[kz m(dAF + y) + φy m(dAF + y) + φz n(dAF + y) + φy n(dAF + y) + φz m]y m]z , n]y n]z . The spin-diffusion equation determines the spatiotem- s (r, t) in the poral evolution of the spin accumulation µN normal metal, ∂µN s (r, t) ∂t = γNH0 × µN s + DN ∂2µN s ∂y2 − µN s τ N sf , (8) where γN is the gyromagnetic ratio in the normal metal, H0 is an external magnetic field, DN is the diffusion con- stant, and τ N is the spin-flip relaxation time. When sf τ N is considerably smaller than the other time scales of sf the system (applied ac voltage frequency and character- istic magnetic field frequency ωH = γNH0), the spin- diffusion equation can be approximated to be static. This approximation is true for metals such as Pt, which has a spin-flip relaxation time that is as low as 0.01 ps [20]. We use this simplification in our calculations, leaving us with a 1D Helmholtz equation characterized by the spin- diffusion length λN sd = sf , with solutions given by hyperbolic functions. (cid:113) DNτ N The source of the spin accumulation in the normal metal is an ac voltage. This ac voltage leads to an ac c (t)x(cid:48) in the metal, which gen- charge current I0 erates an oscillating spin current through the spin Hall effect [8]. We consider a harmonic ac current with fre- quency ω. The total spin current in the normal metal is then c(t) = I 0 (6a) (6b) (7a) (7b) Fig. 2 illustrates the different standing waves. IN s (y, t) = θSH 2e c (t)z(cid:48) − σA I 0 4e2 ∂µN s (y, t) ∂y . (9) where θSH is the spin Hall angle and σ is the conductivity of the normal metal. N=0 N=1 N=2 FIG. 2. the transversal direction in the limit when φy,z ky,z m,ndAF = N π. Standing waves of the N´eel order parameter in m,n = 0 and The boundary conditions that µN s must satisfy are that the spin current across the normal metal-vacuum interface must vanish (IN s (y = dN, t) = 0) and that the spin current across the AFN interface is continu- ous (IN (y = 0, t)). The spin current in the antiferromagnetic insulator is given by contributions from SP and STTs, and it is approximated as s (y = 0, t) = IAF s (cid:2) (n × n) + n ×(cid:0)n × µN (cid:1)(cid:3) s y=0 (10) IAF s (t) ≈ g⊥ 2π to the leading order in the applied ac voltage bias, where g⊥ is the transverse spin conductance. We have dis- regarded contributions from the imaginary part of the transverse spin conductance since it is small in most ma- terials. We also consider the exchange limit (ω(cid:107) (cid:28) ωE), which is a good approximation for many antiferromag- netic materials. In the exchange limit, the antiferromag- net is approximately collinear also at resonance, which means that the net magnetization is negligible. Any SP contributions from the magnetization will therefore be insignificant compared to the SP from the N´eel order pa- rameter. By solving the spin-diffusion equation (8) with the boundary conditions, we find that sinh(cid:2)(2y − dN)/(2λN sd)(cid:3) sinh(cid:2)dN/(2λN sd)(cid:3) z(cid:48) + 1 1 + ξ µN s (y, t) = µs0(t) cosh(cid:2)(y − dN)/λN (cid:3) cosh(cid:2)dN/λN sd sd 4 (cid:3) × [ (n × n) − µs0(t)n × (n × z(cid:48))]y=0 , (11) where we have introduced the dimensionless parameter ξ =(cid:2)πσA tanh(cid:0)dN/λN sd tanh(cid:0)dN/2λN (cid:1)(cid:3) /(2g⊥e2λN (cid:1) I 0 sd sd sd) c (t)/(Aσ) . µs0(t) = 2θSHeλN and a characteristic spin accumulation (12) (13) C. Magnetization dynamics The magnetization dynamics in the antiferromagnet can be divided into two separate regions: the dynamics at the interfaces and the dynamics in the bulk. At the AFN interface, the STTs τ STT m,n drive the dynamics, and there are also dissipative SP torques (τ SP m,n). By integrating the LLG equation (1) in a small volume around the AFN interface, we find the boundary conditions for n: dAFα(cid:48)(cid:20) +ω(cid:107)(cid:2)λ2 1  n × n + nn × ∂yn(cid:3) s )(cid:1)(cid:21) (cid:0)n × (n × µN y=0 y=0 = 0 . (14) We assume that the other interface (y = −dAF) connects to vacuum or a substrate with neither SP nor spin trans- fer. Subsequently, there is only a contribution to the boundary conditions from the exchange stiffness, which requires that the spatial derivative in the transversal di- rection vanishes. By linearizing the boundary condition in Eq. (14) with the ansatz in Eq. (6), we obtain the following constraint on the wave number kz n: kz ndAF tan(kz ndAF) = i AFα(cid:48)ω d2 nω(cid:107) λ2 κ , (15) (cid:20) where we have introduced κ = 1 + 2e2g⊥λN sd coth(dN/λN sd) Aπσ (cid:21)−1 = ξ ξ + 1 . (16) (15) is When the term on the right-hand side of Eq. small, we can expand the solution around the roots where m,ndAF ≈ N π (N = 0, 1, 2, . . .) to determine the wave ky,z numbers. This limit corresponds to the low-damping limit where the decay of the standing waves in the anti- ferromagnetic layer is negligible. Note that the opposite limit implies that the precessions at the interface (and thus the SP and SMR voltages) become small; therefore, this limit is of little interest. The constraint from the boundary conditions in Eq. n depends on an amplitude of (14) on the wave number ky δn, nky nω(cid:107) sin(ky δny = dAF cos θα(cid:48)κµs0/ . ndAF) − idAFα(cid:48)κω cos(ky (cid:2)λ2 ndAF)(cid:3) (17) Another equation is required to find solutions for δny and ky n; therefore, we must solve the LLG equations in the bulk of the antiferromagnet. In the bulk (−dAF < y < 0), the LLG equation be- comes a 4× 4 matrix equation. A non-trivial solution re- quires the determinant of this matrix to be zero because there is no dynamical source in the bulk. The dynamics enters through the boundary conditions at the interface. Because the determinant is independent of the preces- sion amplitudes δm and δn, we can use this condition to determine the solutions for ky n that allow a non-trivial so- lution of the precession amplitudes. The amplitude δny can then be determined from Eq. (17), and the remain- ing amplitudes δmy, δmz, and δnz can be determined from the eigenvectors of the LLG bulk equations. III. SPIN-TRANSFER-TORQUE-INDUCED ANTIFERROMAGNETIC RESONANCE A. Frequency spectrum and susceptibilities In the low-damping and exchange limits, the reso- nance frequencies of the N -node standing wave mode are ω(N )± = ω(N ) 0 ± ωH, where (cid:115) (cid:18) N πλn (cid:19)2 , ω(N ) 0 = dAF (18) ω2 0 + ω(cid:107) n dAF = N π roots are approxi- of Eq. (17) around the ky,z mately complex Lorentzians, and ω0 ≈(cid:112)2ωEω(cid:107) is the gap frequency. The solutions ωE/(cid:0)2(cid:112)2ω(cid:107)(cid:1) (cid:114) ωE y,±(ω) ≈ (−1)N +1α(cid:48)(N )κ cos θµs0√ (cid:17)ω(N )± ω − ω(N )± + i∆ω(N )± /2 where we have introduced the linewidth α0 + α(cid:48)(N )κ ∆ω(N )± = 2 δn(N ) (cid:16) (20) (19) , . 2ω(cid:107) We have also introduced the effective SP damping param- eter α(cid:48)(N ) for the N -mode spin wave, where α(cid:48)(N =0) = α(cid:48) and α(cid:48)(N(cid:54)=0) = 2α(cid:48), which is analogous to the result for ferromagnetic spin waves [21]. The Lorentzian approximation in Eqs. (19) and (20) is valid to the lowest order in α0 and α(cid:48) under the as- sumption that λm,n/dAF (cid:28) 1. If the antiferromagnet is so thin that the thickness becomes comparable to the exchange length, then the gap between the resonance frequencies of the macrospin mode and the higher-order standing wave modes approaches the exchange frequency ωE. This is the upper bound of the resonance frequency; thus, in the limit dAF ∼ λm,n, we can only excite the macrospin mode. Because we also want to study the higher-order standing waves, we only consider the limit where dAF (cid:29) λm,n. cessions is not suppressed by the inverse of this factor. This differs from the case where the source of the dynam- ics is a magnetic field, where the amplitudes are sup- factor of (cid:112)ωE/ω(cid:107), the maximum amplitude of the pre- pressed by a factor of (cid:112)ω(cid:107)/ωE. For the spin-transfer- case in addition to the(cid:112)ω(cid:107)/ωE factor. driven case, the only suppression arises from the high resonance frequencies of the antiferromagnet, and this suppression is also present in the magnetic-field-driven Note that even though the linewidth is enhanced by a The z-component of the N´eel order parameter is re- lated to the y-component by cos (kz ndAF) δn(N ) z,± = ∓sign (ωH ) i cos (ky ndAF) δn(N ) y,± . (21) This is obtained through the eigenvectors of the bulk LLG equations in Eq. (1) in the region where −dAF < y < 0. The magnetization δm also has a circular polar- ization for uniaxial antiferromagnets, and its amplitude is suppressed by a factor of ∝(cid:112)ω(cid:107)/ωE compared to the N´eel order parameter. This suppression factor justifies our discarding of the contributions from the magnetiza- tion in the antiferromagnet to the spin accumulation in the metal. We now assume that the separation between the resonance frequencies ω(N )± is considerably greater than the linewidths ∆ω(N )± and that the real part of the eigenfrequency is much greater than the imaginary part. The frequency spectrum for, e.g., δny(ω), can then be approximated by a sum of the complex Lorentzians in Eq. (19): δny(ω) = δn(N ) y,i (ω) , (22) ∞(cid:88) (cid:88) N =0 i=± and similarly for the other amplitudes. B. Spin Hall magnetoresistance and spin pumping dc voltages We can now use our solutions of the spin accumulation in Eq. (11) and precession amplitudes in Eqs. (19) and (21) to determine the total charge current resulting from the applied voltage and interaction with the antiferro- magnet driven at resonance. This charge current density is [11] 5 I 0 c (t) A x(cid:48) + θSHσ 2e y × ∂µN s (y, t) ∂y . (23) over jc(t) = jc(y, t)dy, we find that the contributions to the the normal metal, jc(y, t) = (cid:82) dN By averaging d−1 x(cid:48)-direction are N 0 jc(t) · x(cid:48) = jSMR c,x(cid:48) (t) + jSP c,x(cid:48)(t), where (cid:20) c,x(cid:48)(t) = − θSHσ I 0 c (t) A jSMR c,x(cid:48) (t) = jSP 2dNe Here, we have introduced ∆ρ0 = −ρθ2 SH η = 1 1 + ξ tanh 1 − ∆ρ0 ρ − ∆ρS ρ (1 − n2 z(cid:48)) η [(n × n)z(cid:48)]y=0 . (cid:19) (cid:18) dN (cid:18) dN (cid:19) (cid:18) dN 2λN sd tanh , 2λN sd dN tanh 2λN sd λN sd (24) (cid:21) , (25) y=0 (26) (cid:19) (27) , (28) and the SMR ∆ρS = −η∆ρ0/2. ρ = 1/σ is the resistiv- ity of the normal metal. The contributions from both the SMR and the SP induce a dc component in the result- ing ISHE voltage in the normal metal. Assuming that c (t) = I 0 I 0 c cos(ωt), we find that (cid:104)jSMR c,x(cid:48) (t)(cid:105)t = ∆ρSI 0 c 2ρA sin 2θRe [δnz cos(kz ndAF)] . (29) To find the dc contributions from SP, we study the dc component of (cid:104)(n × n)z(cid:48)(cid:105)t, and we compute that (cid:104) (δny cos(ky ×δnz cos(kz (cid:105) ndAF))∗ ndAF) sin θ. (30) (cid:104)(n × n)z(cid:48)(cid:105)t = −ωIm Let us now compare the results for the dc components of jSMR c,x(cid:48) and jSP c,x(cid:48) in Eqs. (25) and (26) to the ferromag- netic case [3]. We observe that the results are exactly the same when n ↔ M and Gr → 2Gr, where M is the magnetization unit vector in the ferromagnet and Gr is the real transverse spin conductance in Ref. 3. Experiments measuring the SMR in NiOPt het- erostructures indicate that the SMR is negative for anti- ferromagnets [13–16]. Because the only key difference be- tween the antiferromagnetic case and the ferromagnetic case is that the N´eel order parameter, not the magneti- zation, causes the SMR, the negative sign must be due to some property of the N´eel order parameter. This is in agreement with the reasoning in Ref. 14, where the neg- ative SMR is explained by the coupling of the N´eel order parameter to the magnetic field. They typically couple perpendicularly to each other, whereas for ferromagnets, the magnetization couples along the magnetic field. The perpendicular coupling gives rise to a π/2 phase shift rel- ative to the ferromagnetic case and a negative sign in the measured SMR. If we consider the case in which the susceptibility of the N´eel order parameter is of the same order of magni- tude as the susceptibility of the magnetization in a ferro- magnet, then the SMR and SP voltages in an antiferro- magnet should be comparable to those in a ferromagnet. Eq. (19) shows that the susceptibility scales with the inverse of the resonance frequency. The susceptibility of the N´eel order parameter therefore becomes comparable to that of the magnetization in a ferromagnet when the system is driven close to the spin-flop transition, where the resonance frequency is small [19]. Inserting our solutions of the frequency-dependent am- plitudes in Eqs. (19) and (21), the dc voltages as a function of applied ac voltage frequency become approx- imately 6 V SMR dc (ω) = sign (ωH ) K sin 2θ cos θ dc (ω) = −sign (ωH ) κK sin 2θ cos θ V SP (cid:18) I 0 (cid:19)2 ∞(cid:88) (cid:88) 2(cid:0)α0 + κα(cid:48)(N )(cid:1) i · L(N ) 2(cid:0)α0 + κα(cid:48)(N )(cid:1)(cid:35)2 (cid:34) (cid:19)2 ∞(cid:88) (cid:18) I 0 (cid:88) α(cid:48)(N ) α(cid:48)(N ) i=± ω(N ) c A c A N =0 i i N =0 i=± , (ω) i · L(N ) ω(N ) i i (ω) (31a) , (31b) (t)(cid:105)t and l is the length of where V SMR/SP the bilayer in the direction of the applied voltage. We have also introduced the symmetric Lorentzian = lρ(cid:104)jSMR/SP c,x dc L(N ) i (ω) = and the constant (cid:17)2 (cid:16) ∆ω(N ) /2 (cid:16) (cid:16)ω − ω(N ) SHe(cid:0)λN lκηθ3 sd i dNσ2 i (cid:17)2 (cid:1)2 (32) /2 (cid:17)2 (cid:21) . (33) + ∆ω(N ) i (cid:20) dN 2λN sd K = tanh2 In our model, the SMR and SP voltages as functions of frequency are described via symmetric Lorentzians. However, we have not included the contributions from the Oersted field to the dynamics. The charge current causes an oscillating magnetic field that leads to an antisym- metric Lorentzian component [3]. One therefore needs to filter out the antisymmetric component before compar- ing experimental data with our model. We did not take the Oersted field in the free energy into account since the susceptibility associated with this magnetic field is a factor of ∼ (cid:112)ω(cid:107)/ωE smaller than the susceptibility associated with the spin accumulation [22]. Moreover, because the Oersted field is approximately uniform in a sufficiently thin antiferromagnetic film, it can only couple to the N = 0 mode. The symmetric Lorentzian can there- fore be expected to be the dominant component of the signal for most antiferromagnetic materials and should be the only component for the N (cid:54)= 0 modes. Next, we will compute the dc voltages for MnF2 using the parameters in Table I and Table II. Direct measure- ments of some of the parameters are lacking. We there- fore use these missing material parameters from simi- lar systems. We use the Gilbert damping of NiO, α0 = TABLE I. Material parameters for MnF2. ωE (s−1) [23] ω(cid:107) (s−1) [23] L (A/m) [24] 9.3·1012 1.5 ·1011 47 862 TABLE II. Material parameters for Pt. sd (nm) [26] σ ([Ωm]−1) [27] θSH [25] λPt 1.5 0.12 5·106 2.1· 10−4 [28], and the typical transverse conductance for ferromagnet-normal metal systems, g⊥/A ∼ 3 · 1018 m−2 (ferromagnetPt [29]). The exchange length is ∝ a(cid:112)ωE/ω(cid:107), where a is the typical lattice spacing. We define λn = az(cid:112)ωE/ω(cid:107), where z is the number of nearest-neighbor sites on each sublattice. Because MnF2 has a tetragonal crystal struc- ture and therefore two lattice constants [30], we use the average of the two lattice constants for our characteristic length a. MnF2 has eight nearest neighbors. This results in an estimated exchange length of 26 nm. We note, however, that the real value might differ by as much as an order of magnitude from this value. The exchange length should therefore be estimated by measuring the separation between the resonance peaks. sd , I 0 The results obtained using these parameters are pre- sented in Figure 3, where we have used l = 100 µm, dN = 2λPt c /A = 1010 A/m2, and θ = 35o. The SMR and SP dc voltages always have opposite signs, whereas their frequency dependence is exactly the same. The par- tial cancellation leads to a smaller net signal. The contri- butions from SMR and SP cannot be distinguished from one another in this bilayer system. We also observe that the SP voltage is always smaller than the SMR voltage. For a given direction of the external magnetic field, here assuming that ωH > 0, the signs of the dc signals de- pend solely on whether the precessions are right handed (ωres > 0, + mode) or left handed (ωres < 0, − mode). The dc voltages resulting from the higher-energy modes are not particularly large for our choice of parameters. This is primarily due to the high resonance frequencies of these modes. The standing waves will be easier to detect in materials with a lower gap frequency ω0 than MnF2 or in materials with a shorter exchange length, which leads to a lower resonance frequency for these modes. Examples of antiferromagnets with a low gap frequency are RbMnF3, which has a gap of ω0/2π = 9 GHz [31], and GdFe3(BO3)4, which has a gap of ω0/2π = 29 GHz [32]. For comparison, the gap frequency of MnF2 is ω0/2π = 267 GHz. Because experimental data for the exchange lengths in antiferromagnets are lacking, it is difficult to suggest material candidates for this category. We expect, however, that the exchange length is propor- tional to (cid:112)ωE/ω(cid:107); thus, the exchange length is small in materials where the easy-axis anisotropy is significant compared to the exchange energy. The disadvantage of this is that the exchange energy is generally large in anti- ferromagnets; thus, the gap frequency and spin-flop fields will typically also be large for these materials. An exam- ple of this case is FeF2, which has ω(cid:107) = 0.37ωE [33]. The gap frequency for this material is as large as 1.41 THz, and the spin-flop field is 50.4 T. Spin-transfer antiferro- magnetic resonance experiments with this material will therefore be very challenging. As an alternative to find- ing materials with lower resonance frequencies, one can apply a stronger voltage to enhance the signals because the measured dc voltages are quadratic in the applied ac voltage. In the next section, we will discuss a trilayer system that allows separating the SMR and SP voltages. Sepa- rating these voltages yields more information about the system, such as the ratio between the intrinsic damping to the SP-enhanced damping. IV. SEPARATION OF SPIN HALL MAGNETORESISTANCE AND SPIN-PUMPING VOLTAGES A. NAFN system We now extend and generalize our considerations to an antiferromagnetic insulator sandwiched between two normal metals, as illustrated in Fig. 4. We apply an ac voltage with a constant amplitude to an active normal metal as in the previous sections. Additionally, we use a passive normal metal to detect the SP contributions from the antiferromagnet. Because the passive normal metal does not exhibit any SMR dc voltage to the leading order in the applied voltage source, we measure the SMR and SP voltages independently. 7 (a) (b) 1− 2− 0+ 3− FIG. 3. Resonance spectrum for a MnF2 film of thickness dAF = 100 nm in an external magnetic field ωH = 0.97ω0. The dc voltages for the low-energy macrospin mode (0−) are shown in (a), while some of the higher-energy left-handed N− modes and the right-handed macrospin mode (0+) are shown in (b). n(y, t) IL c (t) IR c (t) Vac NL AF NR FIG. 4. An antiferromagnet sandwiched between two normal metals. Magnetization dynamics in the antiferromagnet is induced by applying an ac voltage with a constant amplitude on the normal metal to the right, which leads to a spin current into the antiferromagnet. A spin current is then pumped into the left normal metal, which induces a charge current through the inverse spin Hall effect. Taking advantage of the symmetry of our system, the spin accumulation in the passive normal metal can read- ily be obtained from Eq. (11). Because the only source for the spin accumulation is SP from the antiferromagnet, we find the spin accumulation to be µNL s (y, t) =  1 + ξ × cosh (cid:104) [n × n]y=−dAF (cid:16) (cid:104) cosh dNL /λNL sd (cid:105)(cid:17)−1 (cid:105) (y + dNL + dAF)/λNL sd . (34) Consequently, the average charge current in the passive normal metal along the x(cid:48)-direction becomes c,x(cid:48)(t) = jL jL c (t) · x(cid:48) = θSHσ 2dNLe η [(n × n)z]y=−dAF . (35) c,x(cid:48)(t)(cid:105)t = −(cid:104)jSP If we, for simplicity, assume no decay of the spin cur- rent in the antiferromagnet and let the properties and dimensions of both normal metals be identical, then we can observe from Eq. (26) that (cid:104)jL c,x(cid:48)(t)(cid:105)t. In other words, we can indirectly measure the SMR dc voltage by measuring the ratio of the dc voltage in the passive normal metal NL relative to the dc voltage in the active normal metal NR. This indirect measurement of the SMR voltage assumes that the decay of the pumped spin current is insignificant. We will now determine in what region this approximation holds. B. Spin-current decay The non-zero spin current across the left AF interface implies that the boundary conditions at y = −dAF must be extended to include SP and STTs: dAFα(cid:48)(cid:20) −ω(cid:107)(cid:2)λ2 1  n × n + nn × ∂yn(cid:3) s )(cid:1)(cid:21) (cid:0)n × (n × µN y=−dAF = 0 . (36) y=−dAF The boundary conditions at y = 0 remain unchanged and are given by Eq. (14). As a result of the new boundary conditions at y = −dAF, the solutions for the phases φy,z m,n in our linear response ansatz in Eq. (7) are no longer zero as they were in the bilayer system. The phases will now have a finite correction in α(cid:48). We can rewrite our bound- ary conditions at y = 0,−dAF to the following constraints on the wave numbers and phases: AFα(cid:48)ω d2 n = −i nω(cid:107) λ2 AFα(cid:48)ω d2 nω(cid:107) λ2 ky,z n dAF tan φy,z kz ndAF tan (kz ndAF + φz n) = i (37b) (37a) κ , κ . We can decouple the above equations to obtain con- straints that are only dependent on the wave number kz n, (cid:34) tan (kz ndAF) kz ndAF + (cid:19)2(cid:35) (cid:18) d2 AFα(cid:48)ω nω(cid:107) λ2 κ 1 kz ndAF AFα(cid:48)ω d2 nω(cid:107) λ2 ≈ kz ndAF tan (kz ndAF) = 2i κ . (38) 8 This constraint is similar to the constraint for the AFN bilayer in Eq. (15) to the lowest order in α(cid:48), except that α(cid:48) → 2α(cid:48). The doubling of the damping due to SP is because we now pump spins across two interfaces rather than one interface. The last constraint on ky n is equivalent to Eq. (17), where we now also have to take the non- zero phase φy n into account; thus, the boundary condition becomes dAF cos θα(cid:48)κµs0/ = δny −idAFα(cid:48)κω cos(ky ndAF + φy nω(cid:107) sin(ky ndAF + φy n) (39) (cid:2)λ2 n)(cid:3) . nky The decay of the spin current in the antiferromag- netic insulator is related to the imaginary components n . At resonance and to the lowest order in α(cid:48) of dAFky,z and α0, we find these to be dAFky (cid:12)(cid:12)Im(cid:0)dAFkz (cid:12)(cid:12)(cid:12)Im (cid:16) (cid:12)(cid:12)Im(cid:0)dAFkz (cid:12)(cid:12)(cid:12)Im (cid:16) dAFky n,N =0 n,N =0 n,N >0 n,N >0 (cid:115) (cid:115) (cid:1)(cid:12)(cid:12) = (cid:17)(cid:12)(cid:12)(cid:12) = (cid:1)(cid:12)(cid:12) = (cid:17)(cid:12)(cid:12)(cid:12) = for the macrospin mode and , AFκα(cid:48)ω d2 nω(cid:107) λ2 AF (κα(cid:48) + α0) ω d2 nω(cid:107) λ2 (40a) , (40b) AFκα(cid:48)ω 2d2 nN πω(cid:107) λ2 AF (2κα(cid:48) + α0) ω d2 nN πω(cid:107) λ2 (41a) , (41b) for the standing-wave modes, respectively. n,N Let us now study how the imaginary components in Eqs. (40) and (41) scale with dAF and the resonance fre- quency ω. Since the SP-induced damping α(cid:48) ∝ 1/dAF, (cid:1) scales as ∝ (dAFω)ζ, where ζ = 1/2 for Im(cid:0)dAFkz AFω(cid:1)ζ n,N ) scales as Im(dAFkz n,N ) scales as ∝ (cid:0)d2 the macrospin mode and ζ = 1 for the standing waves In the limit where α0 (cid:28) α(cid:48), when the bulk (N > 0). damping is small compared to the interface damping, Im(dAFky n,N ). However, when α0 becomes large compared to α(cid:48), the bulk damping dom- inates, and Im(dAFky . We can then observe, as expected, that the spin current decays faster as a function of dAF for thicker films where the bulk damping dominates. Based on these scaling rela- tions, we observe that we can minimize the decay of the spin current, thereby keeping the magnitude of the SP at the two interfaces similar to each other, by (i) keep- ing the antiferromagnetic layer sufficiently thin and (ii) reducing the resonance frequency by driving the system close to the spin-flop transition. Assuming that the SP is dominated by the dynamics of n, which is a good assumption for most collinear antifer- romagnets, the transmission of the pumped spin current through the antiferromagnetic layer can be defined as T (N ) SP = (cid:104)(n × n)y=−dAF (cid:104)(n × n)y=0(cid:105)t (cid:105)t . (42) 9 This describes the ratio of the SP at the passive inter- face relative to the active interface where we excite the dynamics by injecting a spin current. When this ratio is close to unity, it is a good assumption that we are in the low-decay regime, and the pumped spin current across the two interfaces will be approximately the same. In the low-decay regime, the SMR and SP dc voltages can be separated by measuring the dc voltage in both normal metals independently. We plot the transmission of the pumped spin current as a function of dAF in Fig. 5 for the three lowest en- ergy modes. As shown, the pumped spin current of the can also be significant if dAF is very large compared to the exchange length λn or if the resonance frequency is high. However, this result clearly indicates that the transmis- sion of the spin current will quickly decay with increasing dAF as the bulk damping starts to dominate, which is in agreement with our scaling analysis. SP ≈ 1 is n ) (cid:28) 1. We can a good approximation when Im(dAFky,z then utilize the analytical expressions for these imaginary components in Eqs. (40) and (41) to evaluate whether we are in a low-decay regime, where the SMR and SP dc voltages can be separated. Based on our results, one can observe that T (N ) V. CONCLUSIONS We have studied STT-induced antiferromagnetic res- onance in bilayers that consist of an antiferromag- netic insulator and a normal metal and in a metal- antiferromagnetic insulator-metal trilayer. We consider excitations of the uniform mode and of the transverse standing waves. The dc voltages have contributions from the SMR and SP, similar to ferromagnetic systems. In the antiferromagnetic system, the dynamics of the N´eel order parameter causes these effects. A challenge in an antiferromagnetic system is the weak signals due to the low susceptibility of the N´eel order parameter. We demonstrate how the signals are enhanced by driving the system close to the spin-flop transition, where the reso- nance frequency is lower. In trilayer systems, the contri- butions due to SP and SMR can be separated when the antiferromagnetic layer is thin. ACKNOWLEDGMENTS This work was supported by the Research Council of Norway through its Centres of Excellence funding scheme, project number 262633 "QuSpin" and Grant No. 239926 "Super Insulator Spintronics", as well as the European Research Council via Advanced Grant no. 669442 "Insulatronics". Note – During the completion of this work, a recent independent study of spin-transfer antiferromagnetic res- onance was reported [22]. Ref. 22 computes the spin accumulation and frequency dependence of the conduc- tivity in the normal metal for the macrospin mode. A main point and difference in our work is that we consider a magnetic field that importantly reduces the frequency and enhances the output signal. This facilitates experi- mental detection in an experimentally feasible frequency range. Additionally, we study the excitation of standing spin waves and a trilayer system that can be utilized to separate the output signals resulting from SP and SMR. FIG. 5. Ratio of the spin pumping at the passive AFN inter- face relative to the spin pumping at the active AFN interface as a function of dAF for three of the low-energy left-handed N− modes at ωH = 0.9ω0. AFα(cid:48)ωκ/(λ2 AFα(cid:48)ωκ/(λ2 (40) and (41), as d2 modes with the lowest energy exhibits the fastest decay with increasing thickness of the antiferromagnetic layer. This result is as expected from the scaling behaviors in nω(cid:107)) is a small di- Eqs. mensionless number for the choice of parameters that we have previously considered. If d2 nω(cid:107)) is of or- der unity or larger, then we are in the large-damping limit, and we can therefore expect the decay of the spin wave amplitudes to become significant. We can also ob- serve that the N > 0 modes decay when dAF ∼ λn in ad- dition to large values of dAF, unlike the macrospin mode. This result is due to the high resonance frequencies of the standing waves in this limit. It is now interesting to determine at what thickness the bulk damping starts to become more important com- pared to the interface damping. For our choice of param- eters, we have that α(cid:48) = α0 at dAF = 440 nm. As shown in Fig. 5, below this value, the transmission is close to unity, and the pumped spin current quickly decays as we move into the region where the bulk damping dominates. This does not mean that the decay can be neglected in the limit where the interface damping dominates, as this [1] R. Cheng, D. Xiao, and A. Brataas, Phys. Rev. Lett. B 89, 060407 (2014). 10 116, 207603 (2016). [2] L. Liu, T. Moriyama, D. C. Ralph, and R. A. Buhrman, Phys. Rev. Lett. 106, 036601 (2011). [26] S. Meyer, M. Althammer, S. Geprags, M. Opel, R. Gross, and S. T. B. Goennenwein, Appl. Phys. Lett. 104 (2014). (2011), [27] L. Liu, R. A. Buhrman, and D. C. Ralph, [3] T. Chiba, G. E. W. Bauer, and S. Takahashi, Phys. Rev. 10.1088/1751-8113/44/8/085201, arXiv:1111.3702. [28] T. Kampfrath, A. Sell, G. Klatt, A. Pashkin, S. Mahrlein, T. Dekorsy, M. Wolf, M. Fiebig, A. Leitenstorfer, and R. Huber, Nat. Photonics 5, 31 (2011). [29] T. Yoshino, K. Ando, K. Harii, H. Nakayama, Y. Kaji- wara, and E. Saitoh, JPCS 266, 012115 (2011). [30] E. Dormann, J. R. D. Copley, and V. Jaccarino, J. Phys. C 10, 2767 (1977). [31] W. J. Ince, J. Appl. Phys. 37, 1132 (1966). [32] A. I. Pankrats, G. A. Petrakovskii, L. N. Bezmaternykh, and O. A. Bayukov, J. Exp. Theor. Phys. 99, 766 (2004). [33] R. C. Ohlmann and M. Tinkham, Phys. Rev. 123, 425 (1961). Applied 2, 034003 (2014). [4] M. Schreier, T. Chiba, A. Niedermayr, J. Lotze, H. Huebl, S. Geprags, S. Takahashi, G. E. W. Bauer, R. Gross, and S. T. B. Goennenwein, Phys. Rev. B 92, 144411 (2015). [5] J. Slonczewski, J. Magn. Magn. Mater. 159, L1 (1996). [6] J. Sklenar, W. Zhang, M. B. Jungfleisch, W. Jiang, H. Chang, J. E. Pearson, M. Wu, J. B. Ketterson, and A. Hoffmann, Phys. Rev. B 92, 174406 (2015). [7] C. He, A. Navabi, Q. Shao, G. Yu, D. Wu, W. Zhu, C. Zheng, X. Li, Q. L. He, S. A. Razavi, K. L. Wong, Z. Zhang, P. K. Amiri, and K. L. Wang, Appl. Phys. Lett. 109, 202404 (2016). [8] M. Dyakonov and V. Perel, Phys. Lett. A 35, 459 (1971). [9] E. Saitoh, M. Ueda, H. Miyajima, and G. Tatara, Appl. Phys. Lett. 88, 182509 (2006). [10] H. Nakayama, M. Althammer, Y.-T. Chen, K. Uchida, Y. Kajiwara, D. Kikuchi, T. Ohtani, S. Geprags, M. Opel, S. Takahashi, R. Gross, G. E. W. Bauer, S. T. B. Goennenwein, and E. Saitoh, Phys. Rev. Lett. 110, 206601 (2013). [11] Y.-T. Chen, S. Takahashi, H. Nakayama, M. Althammer, S. T. B. Goennenwein, E. Saitoh, and G. E. W. Bauer, Phys. Rev. B 87, 144411 (2013). [12] J. H. Han, C. Song, F. Li, Y. Y. Wang, G. Y. Wang, Q. H. Yang, and F. Pan, Phys. Rev. B 90, 144431 (2014). [13] D. Hou, Z. Qiu, J. Barker, K. Sato, K. Yamamoto, S. V´elez, J. M. Gomez-Perez, L. E. Hueso, F. Casanova, and E. Saitoh, Phys. Rev. Lett. 118, 147202 (2017). [14] G. R. Hoogeboom, A. Aqeel, T. Kuschel, T. T. M. Pal- stra, and B. J. van Wees, Appl. Phys. Lett. 111, 052409 (2017). [15] L. Baldrati, A. Ross, T. Niizeki, R. Ramos, J. Cramer, O. Gomonay, E. Saitoh, J. Sinova, and M. Klaui, (2017), arXiv:1709.00910. [16] J. Fischer, O. Gomonay, R. Schlitz, K. Ganzhorn, N. Vli- etstra, M. Althammer, H. Huebl, M. Opel, R. Gross, S. T. B. Goennenwein, (2017), arXiv:1709.04158. and S. Geprags, [17] A. Manchon, Phys. Status Solidi Rapid Res. Lett. 11, 1600409 (2017), 1600409. [18] R. Cheng, J. Xiao, Q. Niu, and A. Brataas, Phys. Rev. Lett. 113, 057601 (2014). [19] Ø. Johansen and A. Brataas, Phys. Rev. B 95, 220408 (2017). [20] H. J. Jiao and G. E. W. Bauer, Phys. Rev. Lett. 110, 217602 (2013). [21] A. Kapelrud and A. Brataas, Phys. Rev. Lett. 111, 097602 (2013). [22] V. Sluka, (2017), arXiv:1708.08965. [23] M. P. Ross, Spin Dynamics in an Antiferromagnet, Ph.D. thesis, Technische Universitat Munchen (2013). [24] J. P. Kotthaus and V. Jaccarino, Phys. Rev. Lett. 28, 1649 (1972). [25] M. Obstbaum, M. Hartinger, H. G. Bauer, T. Meier, F. Swientek, C. H. Back, and G. Woltersdorf, Phys. Rev. 

1210.1372

1

1210

Non-equilibrium renormalised contacts for transport in nanodevices with interaction: a quasi-particle approach

We present an application of a new formalism to treat the quantum transport properties of fully interacting nanoscale junctions. We consider a model single-molecule nanojunction in the presence of two kinds of electron-vibron interactions. In terms of the electron density matrix, one interaction is diagonal in the central region and the second off-diagonal between the central region and the left electrode. We use a non-equilibrium Green's function technique to calculate the system's properties in a self-consistent manner. The interaction self-energies are calculated at the Hartree-Fock level in the central region and within a dynamical mean-field-like approach for the crossing interaction. Our calculations are performed for different transport regimes ranging from the far off-resonance to the quasi-resonant regime, and for a wide range of parameters. They show that a non-equilibrium (i.e. bias dependent) dynamical (i.e. energy dependent) renormalisation is obtained for the contact between the left electrode and the central region in the form of a non-equilibrium renormalisation of the lead embedding potential. The conductance is affected by the renormalisation of the contact: the amplitude of the main resonance peak is modified as well as `the lineshape of the first vibron side-band.

cond-mat.mes-hall

cond-mat

Non-equilibrium renormalised contacts for transport in nanodevices with interaction: a quasi-particle approach H. Ness, L. K. Dash Department of Physics, University of York, Heslington, York YO10 5DD, UK European Theoretical Spectroscopy Facility (ETSF) Abstract. We present an application of a new formalism to treat the quantum transport properties of fully interacting nanoscale junctions We consider a model single- molecule nanojunction in the presence of two kinds of electron-vibron interactions. In terms of the electron density matrix, one interaction is diagonal in the central region and the second off-diagonal between the central region and the left electrode. We use a non-equilibrium Green's function technique to calculate the system's properties in a self-consistent manner. The interaction self-energies are calculated at the Hartree- Fock level in the central region and within a dynamical mean-field-like approach for the crossing interaction. Our calculations are performed for different transport regimes ranging from the far off-resonance to the quasi-resonant regime, and for a wide range of parameters. They show that a non-equilibrium (i.e. bias dependent) dynamical (i.e. energy dependent) renormalisation is obtained for the contact between the left electrode and the central region in the form of a non-equilibrium renormalisation of the lead embedding potential. The conductance is affected by the renormalisation of the contact: the amplitude of the main resonance peak is modified as well as 'the lineshape of the first vibron side-band. PACS numbers: 73.40.Gk, 71.38.-k, 73.63.-b, 85.65.+h 2 1 0 2 t c O 4 ] l l a h - s e m . t a m - d n o c [ 1 v 2 7 3 1 . 0 1 2 1 : v i X r a Non-equilibrium many-body transport 2 1. Introduction The theory of quantum electronic transport in nano-scale devices has evolved rapidly over the past decade, as advances in experimental techniques have made it possible to measure transport properties down to single-molecule devices. The development of accurate theoretical methods for the description of quantum transport at the single- molecule level is essential for continued progress in a number of areas including molecular electronics, spintronics, and thermoelectrics [1]. One of the longstanding problems in quantum charge transport is the establishment of a theoretical framework which allows for quantitatively accurate predictions of conductance from first principles. The need for methods going beyond the conventional approaches, based on equilibrium electronic structure calculations combined with Landauer-like elastic scattering, has been clear for a number of years. Only recently have more advanced methods to treat electronic interaction appeared (for example see Refs. [2, 3, 4]). Alternative frameworks to deal with the steady-state or time-dependent transport are given by many-body perturbation theory, based on the non-equilibrium (NE) Green's function (GF) formalism: in these approaches, the interactions and (initial) correlations are taken into account by using conserving approximations for the many-body self- energy [5, 6, 7, 8, 9, 10, 11, 12, 13, 14]. Other interactions, such as electron-phonon coupling, also play an important role in single-molecule quantum transport. These interactions are more important in nanoscale systems, as the electronic probability density is concentrated in a small region of space and thus normal screening mechanisms are ineffective. Such interactions are also crucial in inelastic electron tunneling spectroscopy. Such a technique constitutes an important basis for spectroscopy of molecular junctions, yielding insight into the vibrational modes (single molecule phonons called vibrons) and ultimately the atomic structure of the junction [15]. There have been many theoretical investigations focusing on the effect of electron- vibron coupling in molecular- and atomic-scale wires [16-51]. In all these studies, the interactions have always been considered to be present in the central region (i.e. the molecule) only, with the latter connected to two non-interacting terminals. Interactions are also assumed not to cross at the contracts between the central region and the leads. When electronic interactions are present throughout the system, as within density- functional theory calculations, they are treated only at the mean-field level and do not allow for any inelastic scattering events. However, such approximations are valid only in a very limited number of practical cases. The interactions, in principle, exist throughout the entire system. In a recent paper we derived a general expression for the current in nano-scale junctions with interaction present everywhere in the system [52]. The importance of such extended interactions in nano-scale devices has also been addressed, for electron- electron interaction, in recently developed approaches such as Refs. [2, 53]. Non-equilibrium many-body transport 3 In the present paper, we apply our formalism [52] to a specific model of single- molecule nanojunctions. We focus on a model system in the presence of electron-vibron interaction both within the molecule and between the molecule and one of the leads. We adopt a quasiparticle-like approach to treat the crossing interactions (i.e. there are some restrictions on the components of the self-energy for the crossing interaction). We show how the interaction crossing at one interface of the molecular nanojunctions affects the transport properties by renormalising the coupling at the interface in a dynamical and bias-dependent manner. The paper is organised as follows: In Sec. 2, we briefly recall the main result of our expression for the current in fully interacting systems. In Sec. 3, we present the model Hamiltonian for the system which includes two kinds of electron-vibron interaction: a Holstein-like Hamiltonian combined with a Su-Schrieffer-Heeger-like Hamiltonian. In this section, we also describe how the corresponding self-energies are calculated and the implications of such approximations on the current expression at the left and right interfaces. In Sec. 4, we show how the non-equilibrium dynamical renormalisation affects the generalised embedding potential of the left (L) lead, and how in turn this affects the conductance of the nanojunction. We finally conclude and discuss extensions of the present work in Sec. 5. 2. General theory for quantum transport We consider a two-terminal device, made of three regions left-central-right, labelled L − C − R, in the steady-state regime. In such a device the interaction -- which we specifically leave undefined (e.g. electron-electron or electron-phonon) -- is assumed to be well described in terms of the single-particle self-energy ΣMB and spreads throughout the entire system. We use a compact notation for the Green's function G and the self-energy Σ matrix elements M (ω). They are annotated MC (ML or MR) for the elements in the central region C (left L, right R region respectively), and MLC (or MCL) and MRC (or MCR) for the elements between region C and region L or R. There are no direct interactions between the two electrodes, i.e. ΣMB LR/RL = 0. In Refs. [52, 54], we showed that for a finite applied bias V the steady-state current IL(V ) flowing through the left LC interface is given by: (cid:90) dω (cid:104) Tr{C} 2π L G< ΣMB,> (cid:104) ΥL,l L − ΣMB,< Gr C L G> L (cid:105) IL = e  +Tr{L} C + Ga C( ΥL,l C )† + G< C( ΥL C − ( ΥL C)†) (cid:105) where the ΥX C quantities are ΥL C(ω) = Σa C)† = Σa ( ΥL ΥL,l C = Σ< CL(ω) ga L(ω) Σr L Σr CL gr LC, L − gr CL (ga L) Σr LC(ω), LC + Σr CL g< L Σr LC. (1) (2) Non-equilibrium many-body transport 4 By definition ΣLC(ω) = VLC + ΣMB LC (ω) (similarly for the CL components) where VLC/CL are the nominal coupling matrix elements between the L and C regions. gx L(ω) are the GF of the region L renormalised by the interaction inside that region, where x = r, a, < stands for the retarded, advanced and lesser GF components respectively. L (ω))−1 = For example, L (ω))−1 − ΣMB,r/a (gr/a for the advanded and retarded components, we have (gr/a (ω) where all quantities are defined only in the subspace L. L The second trace in Eq. (1) corresponds to inelastic events induced by the interaction in the L lead. At equilibrium, because of the detailed balance equation Σ>G< = Σ<G>, this term vanishes. At non-equilibrium, this is generally not the case. However, when a local detailed balance equation holds, the system is locally in a (quasi)equilibrium state, this terms vanishes since one recovers locally ΣMB,>G< = ΣMB,<G>. Hence, in practice, we do not have to worry about the infinite sum in the trace Tr{L}[...] since there will always be a region/boundary in the left lead L beyond which the system is at (quasi)equilibrium. i.e. The first trace in the current equation Eq. (1) corresponds to a generalisation of the result of Meir and Wingreen [55]. It encompasses the cases for which the interactions are present in the three L, C, R regions as well as in between the L/C and C/R regions. It also bears some resemblance to the expression derived by Meir and Wingreen [55] when written as: I MW L = e  Tr{C} Gr CΣL,< C + Ga C(ΣL,< C )† + G< C(ΣL,a C − ΣL,r C ) . (3) (cid:90) dω 2π (cid:104) (cid:105) where we use the standard definitions ΣL,< C − ΣL,r ΣL,a C = VCL(ga L)VLC = iΓL. L − gr † By comparing Eq. (1) and Eq. (3), we can see that the quantities ΥLC ( Υ LC) and L respectively LC are now playing the role of the L lead self-energy Σa Υl when the interactions cross at the LC interface. L) and Σ< L (Σr C = −(ΣL,< C )† = VCL g< L VLC = ifLΓL and 3. Model for the interaction 3.1. Hamiltonians We consider a single-molecule junction in the presence of electron-vibron interaction both inside the central region and crossing at the contacts. We further concentrate on a model for the central region which consists of a single molecular level coupled to a single vibrational mode. A full description of our methodology, for the interaction inside the region C, is provided in Refs. [56, 57, 58]. Moreover, we consider that some electron-vibron interaction exists also at one contact (the left L electrode for instance). This model typically corresponds to an experiment for a molecule chemisorbed onto a surface (the left electrode) with a tunneling barrier to the right R lead. In the following model, we consider two kinds of electron-vibron coupling: a local coupling in the sense of a Holstein-like coupling of the electron charge density with a internal degree of freedom of vibration inside the central region, and an off-diagonal Non-equilibrium many-body transport 5 Figure 1. (Color online) Schematic representation of a central scattering region C connected to the left L and right R electrodes. Interactions are given by the coupling of the region C to the L (R) electrode (VLC/CL and VRC/CR), and by the many- body effects within the central region (ΣMB LC ). ΣMB corresponds to the coupling of an electron with the vibron mode ω0 (with coupling strength γ0), and ΣMB LC corresponds to the coupling of a hopping electron with another vibron mode ωA (with coupling strength γA). C ) and at the LC interface (ΣMB C coupling in the sense of a Su-Schrieffer-Heeger-like coupling [59, 17] to another vibration mode involving the hopping of an electron between the central C region and the L electrode. A schematic representation of the molecular junction is given in Figure 1. The Hamiltonian for the region C is HC = ε0d†d + ω0a†a + γ0(a† + a)d†d, (4) where d† (d) creates (annihilates) an electron in the molecular level ε0. The electron charge density in the molecular level is coupled to the vibration mode of energy ω0 via the coupling constant γ0, and a† (a) creates (annihilates) a vibration quantum in the vibron mode ω0. The central region C is nominally connected to two (left and right) one- dimensional tight-binding chains via the hopping integral t0L and t0R. The corresponding 0αeikα(ω)/βα with the dispersion relation electrode α = L, R self-energy is Σr ω = εα + 2βα cos(kα(ω)) where εα and βα are the tight-binding on-site and off-diagonal elements of the electrode chains. α(ω) = t2 To describe the electron-vibron interaction existing across the left contact, we consider that the hopping integral t0L is actually dependent on some generalised coordinate X. The latter represents either the displacement of the centre-of-mass of the molecule or of some chemical group at the end of the molecule link to the L electrode. At the lowest order, the matrix element can be linearised as t0L(X) = t0L + t(cid:48) 0LX. Hence the hopping of an electron from the C region to the L region (and vice versa) is coupled to a vibration mode (of energy ωA) via the coupling constant γA (itself related to t(cid:48) 0L). The corresponding Hamiltonian is given by HLC = γA(b† + b)(c (5) where b† (b) creates (annihilates) a vibration quantum in the vibron mode ωA, the † L (cL) creates (annihilates) generalised coordinate is X =(cid:112)/(2mAωA)(b† + b), and c † Ld + d†cL) + ωAb†b, an electron in the level εL of the L electrode. Non-equilibrium many-body transport 6 The Hamiltonians Eq. (4) and Eq. (5) are used to calculate the corresponding electron self-energies at different orders of the interaction γ0 and γA using conventional non-equilibrium diagrammatic techniques [56, 58]. Furthermore, at equilibrium, the whole system has a single and well-defined Fermi level µeq. A finite bias V , applied across the junction, levels as µL,R = µeq + ηL,ReV . The fraction of potential drop [60] at the left contact is ηL and ηR = ηL − 1 at the right contact, with µL − µR = eV and ηL ∈ [0, 1]. lifts the Fermi 3.2. Self-energies for the interactions In this paper, we consider different approximations to the treat the interaction inside the central region and the interaction crossing at the left contact. First of all, calculating exactly the corresponding interaction self-energies is a tremendous task since, in principle, they depend on both the phonon progator and the electron Green's functions in all the different parts of the system. Hence, for the first application of our formalism, we proceed step by step in terms of the increasing complexity of the interaction; and we limit ourselves to approximations for the self-energies that are well known and well controlled. The electron-vibron self-energies in the central region C are calculated within the self-consistent Born approximation (i.e. diagrams with one phonon line). The details of these calculations have been reported elsewhere [56, 58]. For the crossing interaction, we consider a mean-field-like approach for the electron-vibron coupling at the LC interface. Within such an approximation, we can understand the results of the calculations in terms of renormalisation of the LC contact. Furthermore considering the crossing interaction occuring only at one interface permit us to check and test the consistency of our formalism. Indeed, with no interaction crossing at the right CR interface, the current IR is given by the conventional Meir and Wingreen formula, i.e. Eq. (3) for the CR interface. In order to have a consistent formalism, we need to have current conservation, and such a constraint is best tested with crossing interaction at only one interface. The corresponding results are shown in detail in 4.1. Within mean-field-like approximations, the effects of the crossing interaction correspond to a renormalisation of the coupling in a static or a dynamical manner. The LC and Σr/a corresponding self-energies have only retarded and advanced components Σr/a CL. The extra inelastic effects included in the components Σ C quantities defined in Eq. (2) become: ≷ CL are neglected altogether. Hence the ΥX CL ga CL gr CL g< ΥL C = Σa C)† = Σa ΥL,l C = Σr C )† = −Σa L Σr L Σr L Σr CL g< LC, LC, LC, L Σa LC. ( ΥL ( ΥL,l (6) with Σr/a LC = VLC + ΣMB,r/a LC . Non-equilibrium many-body transport 7 Correspondingly, the generalised embedding potentials for the left contact defined C = [Σr LC]x [52, 54] with x = r, a, ≷, are now given by CLgLΣr as Y L,x Y L,r C (ω) = Σr Y L,a C (ω) = Σa Y L,< C (ω) = Σr CLgr CLga CLg< LΣr LΣa L Σa LC, LC, LC. (7) In the static limit, the mean-field approach leads to the Hartree expression for the electron-vibron self-energies at the LC interface: where (cid:104)nLC(cid:105) = −i (cid:104)nLC(cid:105), G< LC(ω). (8) (9) 2π ΣMB,r/a LC = −2 γ2 A ωA (cid:90) dω LC The self-energy ΣMB,r/a is independent of the energy ω and leads to a static (nonetheless bias-dependent) renormalisation of the nominal coupling VCL = VLC = t0L between the L and C regions. This NE renormalisation will induce, among other effects, a bias-dependent modification of the broadening of the spectral features of the C region. We have analysed these effects in details in Ref.[51]. The renormalisation is such that the amplitude of the current is reduced in comparison with the current values obtained when the interaction is present only in the central region. The NE static renormalisation of the contact is highly non-linear and non-monotonic in function of the applied bias, and the larger effects occur at applied biases corresponding to resonance peaks in the dynamical conductance. The conductance is also affected by the NE renormalisation of the contact, showing asymmetric broadening around the resonance peaks and some slight displacement of the peaks at large bias in function of the coupling strengh γA. Beyond the static limit, we can develop a dynamical mean-field-like approach for the electron-vibron coupling following Ref. [62]. The retarded self-energy containing all orders of the electron-vibron coupling is expressed as a continued fraction as shown analytically in Refs. [63, 64]. We have already used such an approach to study the transport properties of organic molecular wires which are dominated by the propagation of polarons [16, 17] or solitons [18]. The expression for the corresponding self-energy Σr LC(ω) is given by: Σr LC(ω) = LC(ω − ωA)−1 − Gr γ2 A 2γ2 A LC(ω − 2ωA)−1 − Gr 3γ2 A LC(ω − 3ωA)−1 − ... Gr (10) where Gr and the left lead L. LC is the retarded component of the off-diagonal GF between the central region Its closed expression is obtained from the corresponding Dyson Non-equilibrium many-body transport (cid:104) C]−1 − Y R,r [gr C (cid:21)−1 (cid:105)−1 Σr CL Σr LC (cid:104) C]−1 − Y R,r [gr C 8 (cid:105)−1 , (11) equation Gr LC = [gΣG]r LC: Gr LC(ω) = L]−1 − Σr [gr LC (cid:20) C with gr C (ω) = VCRgr (ω)]−1 and Y R,r C(ω) = [ω − ε0 − ΣMB,r Such a dynamical mean-field-like approach, which corresponds to a quasi-particle LC/CL can be incorporated into a Schrodinger-like is expected to affect the transport properties via the non-equilibrium through the generalised embedding approach since the self-energy Σr/a equation, dynamical renormalisation of the contacts, i.e. potentials Y L,x C (ω) as well as through the corresponding Υx C(ω) quantities. R(ω)VRC. Finally, note that for the lowest-order expansion, the self-energy Σr LC takes a simple form: 4. Results Σr LC(ω) = γ2 AGr LC(ω − ωA) . (12) We have perfomed calculations, in a self-consistent manner, for many different values of the Hamiltonian parameters. We present below the most characteristic results for different transport regimes and for different coupling strengths γA, while the interaction in the region C is taken to be in the intermediate coupling regime ω0 = 0.2, γ0 = 0.15, i.e. γ0/ω0 = 0.75. The nominal couplings between the central region and the electrodes t0L,R, before NE renormalisation, are not too large, so that we can discriminate clearly between the different vibron side-band peaks in the spectral functions. The values chosen for the parameters are typical values for realistic molecular junctions [65, 58]. In the following the current is given in units of charge per time, the conductance in units of the quantum of conductance G0 = 2e2/ and the bias V and the embedding potential YC in natural units of energy where e = 1 and  = 1. 4.1. Conserving approximation One of the most important physical conditions that our formalism needs to fulfil is the constraint of current conservation. We use conserving approximations to calculate the interaction self-energies in the central region C and an quasi-particle approximation for crossing interaction at the left interface. With our choice for Σr LC(ω) given by Eq. (10), we find that the left lead generalised embedding potential Y L,r C (ω) is renormalised by the crossing interaction. This is clearly seen in Figure 2, which shows the imaginary part of Y L,r for different transport regimes, different coupling at the LC interface and for both equilibrium and non-equilibrium conditions. C At equilibrium, and in the absence of contact renormalisation, the imaginary is simply the imaginary part of the conventional L lead self-energy part of Y L,r C Non-equilibrium many-body transport 9 Σr L(ω) = t2 0LeikL(ω)/βL and corresponds to a semi-elliptic functions with non-zero values within the energy range −2βL ≤ ω ≤ +2βL (see dashed lines in figure 2). The spectral weight of (cid:61)m Σr L is given by(cid:82) dω (cid:61)m Σr (cid:82) dω (cid:61)m gr L(ω) = −πt2 0L. L(ω) = t2 0L In the presence of the renormalised L contact, we obtained a strong deviation from the semi-elliptic function as shown by the (red) thin lines and (black) thick lines in Figure 2, which correspond respectively to the equilibrium (bias V = 0) and the non- equilibrium (bias V = 1.0) conditions. This result indicates a strong reduction of the available transport channels in the renormalised L lead embedding potential for regions of energy where Σr/a However, a very important property is that the total spectral weight of (cid:61)m Σr that in all the cases (cid:82) dω (cid:61)m Y L,r L is conserved for all the calculations we have performed, i.e. for all the different transport regimes, coupling strength γA and ωA at the LC interface and all applied bias. We find 0L (with maximum deviation of 0.05% arising from numerical errors in the results shown in Figure 2). So in this sense, we can say that the approximation for Σr,a LC/LC(ω) has non-zero value. C (ω) = −πt2 LC/LC is conserving. Correspondingly, we have also carefully checked that the current is conserved for all the calculations presented in the present paper, i.e. that IL +IR/IL ∼ IL +IR/IR < 10−5, with the current IL at the LC interface is given by expression Eq. (1) and the current IR at the CR interface does not contain any contact renormalisation, and hence is given by a Meir and Wingreen like expression Eq. (3). 4.2. Dynamical non-equilibrium renormalisation C(ω))/2πi. The dynamical renormalisation of the L lead embedding potential Y L,r C presents features (dips) at some energies. Qualitatively speaking, there is a form of correlation between them and the features that exist in the spectral function of the central region AC(ω) = (Ga C(ω) − Gr Figure 3 shows the spectral function AC(ω), rescaled by some factor, for a set of parameters corresponding to panel (a) in figure 2. The bottom panel of figure 3 shows the imaginary part of Y L,r C (ω) from which the semi-elliptic background has been C − Σr subtracted, i.e. ∆Y L,r L. Qualitatively speaking, both quantities present similar features, shifted in energy by an amount corresponding to ωA (ωA = 0.2 in the calculations) as expected from the definition of Σr/a C = Y L,r LC/LC(ω). One can then expect a maximum effect of the dynamical renormalisation in an C (ω) coincide with integrated quantity such as the current when the features in Y L,r those in the spectral function AC(ω) within a given bias energy-window. We now consider the modification of the dynamical conductance G(V ) = dI/dV induced by the crossing interaction Σr/a LC/LC(ω) for different transport regimes and for different crossing interaction strength. Figures 4 and 5 show G(V ) for three different transport regimes for weak to strong crossing interaction strength γA, and for two values of ωA (ωA = 0.2 for Fig. 4 and ωA = 0.1 Fig. 5). The dynamical renormalisation of the left contact slightly affects the main resonance Non-equilibrium many-body transport 10 (Color online) Imaginary part of the left-lead generalised embedding Figure 2. potential Y L,r C (ω). The dashed lines represent the L lead spectral function in the absence of renormalisation at the LC interface (i.e. semi-elliptic density of states). The thin (red) lines are the spectral functions (cid:61)mY L,r C (ω) at equilibrium (V = 0). The thick (black) lines are the non-equilibrium spectral functions (cid:61)mY L,r C (ω) at finite bias (V = 1.0). These show a strong reduction of the spectral functions on energy range of ∆ω ≈ 1 around ε0. Panel (a) Off-resonant regime ε0 = 0.5 and strong coupling at the interface ωA = 0.20, γ0 = 0.28. (b) Off-resonant regime ε0 = 0.5 and weak coupling at the interface ωA = 0.20, γ0 = 0.07. (c) Resonant regime ε0 = 0.05 and medium coupling at the interface ωA = 0.20, γ0 = 0.14. (d) Intermediate regime ε0 = 0.2 and medium coupling at the interface ωA = 0.10, γ0 = 0.07. The other parameters are ω0 = 0.20, γ0 = 0.15, t0R = t0L = 0.22, βα = 2.0,
α = 0.0, ηV = 1. peak in the conductance, and more importantly the lineshape of the first vibron side- band peak above the main conductance peak. The most important effects are obtained for the strong crossing interaction strength γA = 0.28. Even if the ratio γA/ωA is conserved between the calculations shown in Figure 4 and Figure 5, the absolute value of γA is the crucial quantity that governs the effects of the dynamical renormalisation. Furthermore, for the different sets of parameters we have considered, our calculations show that the lower order expansion for Σr LC (see Eq. ()) provides a good approximation to the results obtained with a larger number of levels in the continued fraction expansion -4-2024-0.025-0.02-0.015-0.01-0.0050Im YCL,r(ω)strong coupling-2-1012-0.024-0.023-0.022-0.021-0.02V = 1.0V = 0.0weak coupling-4-2024ω-0.025-0.02-0.015-0.01-0.0050Im YCL,r(ω)medium coupling-2-1012ω-0.024-0.023-0.022-0.021medium coupling(a)(b)(c)(d)ε0 = 0.05ε0 = 0.2ε0 = 0.5ε0 = 0.5Non-equilibrium many-body transport 11 Figure 3. (Color online) Rescaled spectral function proportional to −(cid:61)mGr C for the central region (top panel) and imaginary part of the L lead generalised embedding C − Σr potential relative to the non-renormalised L lead self-energy: ∆Y L,r L (bottom panel). The calculations are performed for the same parameters as in panel (a) of Figure 2 and for both zero bias and finite bias V = 1. Here µR = 0 and µL = V . The qualitative correlations between the features in ∆Y L,r C and in the spectral function of the C region −(cid:61)mGr C = Y L,r C are clearly observed. of Σr LC (see panel (a) in fig. ). We now check the effects of varying the nominal coupling t0α on the conductance. The results are shown in Figure 6 for the off-resonant transport regime with and without dynamical renormalisation. With renormalisation of the left contact, the conductance values decrease with increasing coupling t0α to the leads while the conductance peaks broaden. This is seen more clearly in the right panel of Figure 6 where the rescale conductance G(V )/Gmax is plotted such that the main resonance peak has an amplitude of 1. In the presence of dynamical renormalisation of the L contact, the main conductance peak follows the same behaviour as in the absence of renormalisation. However, one observes a highly non-linear behaviour of the modification of the first vibron side-band peak. The complete and detailed understanding of such modification is rather complex, and beyond the scope of the present paper. However it is strongly related to the new features in the renormalised L lead embedding potential which depart from the non-interacting case. -0.4-0.200.20.40.60.811.21.4ω00.0050.010.0150.02Im ∆YCL,r(ω)V = 1.0V = 0.0-Im GCr(ω)V = 1.0V = 0arbitrary unitsNon-equilibrium many-body transport 12 Figure 4. (Color online) Dynamical conductance G(V ) for different transport regimes and different coupling strength γA, and ωA = 0.20. Panel (a) Resonant regime ε0 = 0.05. Panel (b) Off-resonant regime ε0 = 0.2. Panel (c) Off-resonant regime ε0 = 0.5. The non-equilibrium dynamical renormalisation of the LC contact affects both the main conductance peak and the first vibron side-band peak. The other parameters are ω0 = 0.20, γ0 = 0.15, t0R = t0L = 0.22, βα = 2.0,
α = 0.0, ηV = 1. 5. Conclusion We have studied the transport properties through a two-terminal nanoscale device with interactions present not only in the central region but also with interaction crossing at the interface between the left lead and the central region. To calculate the current for such a fully-interacting system, we have used our recently developed quantum transport formula [52] based on the NEGF formalism. As a first practical application, we have considered a prototypical single-molecule nanojunction with electron-vibron interaction. In terms of the electron density matrix, the interaction is diagonal in the central region for the first vibron mode and off-diagonal between the central region and the left electrode for the second vibron mode. The interaction self-energies are calculated in a self-consistent manner using the lowest order Hartree-Fock-like diagram in the central region and a quasi-particle (dynamical mean-field-like) approach for the 00.20.40.60.8V bias00.050.10.150.20.25G(V)γA = 0γA = 0.28 - 1 levelγA = 0.28 - 9 levelsγA = 0.20 - 1 levelγA = 0.14 - 1 levelγA = 0.07 - 1 level0.10.20.30.4V bias0.010.020.030.040.050.06(a)00.10.20.30.40.5V bias00.10.20.30.40.50.60.70.8G(V)γA = 0.0γA = 0.28 - 1 levelγA = 0.20 - 1 levelγA = 0.14 - 1 levelγA = 0.07 - 1 level(b)00.10.20.30.40.50.60.70.80.91V bias00.10.20.30.40.50.60.70.8G(V)γA = 0.0γA = 0.28 - 1 levelγA = 0.20 - 1 levelγA = 0.14 - 1 levelγA = 0.07 - 1 level(c)Non-equilibrium many-body transport 13 Figure 5. (Color online) Dynamical conductance G(V ) for different transport regimes and different coupling strength γA, and ωA = 0.10. Panel (a) Resonant regime ε0 = 0.05. Panel (b) Off-resonant regime ε0 = 0.2. Panel (c) Off-resonant regime ε0 = 0.5. The non-equilibrium dynamical renormalisation of the LC contact affects both the main conductance peak and the first vibron side-band peak. The other parameters are ω0 = 0.20, γ0 = 0.15, t0R = t0L = 0.22, βα = 2.0,
α = 0.0, ηV = 1. crossing interaction. Our calculations were performed for different transport regimes ranging from the far off-resonance to the quasi-resonant regime, and for a wide range of parameter values. They show that, for this model, we obtain a non-equilibrium (i.e. bias dependent) dynamical (i.e. energy dependent) renormalisation of the generalised embedding potential of the L left lead. The renormalisation is such that the amplitude of the corresponding spectral function is reduced around the molecular level energy ε0 over an energy range roughly equal to the energy support of the spectral density of the central region C. This corresponds a reduction of the number of transport channels in the left lead at given energy, even if the total spectral weight of the L embedding potential is conserved. The non-equilibrium dynamical renormalisation of the L contact is highly non- linear and non-monotonic in function of the applied bias, and the larger effects occur at 00.20.40.60.8V bias00.050.10.150.2G(V)γA = 0.0γA = 0.14 - 1 levelγA = 0.10 - 1 levelγA = 0.07 - 1 levelγA = 0.035 - 1 level00.050.10.150.2V bias0.050.10.150.2(a)00.10.20.30.40.50.60.70.8V bias00.20.40.6G(V)γA = 0.0γA = 0.14 - 1 levelγA = 0.10 - 1 levelγA = 0.07 - 1 levelγA = 0.035 - 1 level0.20.30.4V bias0.050.1(b)00.10.20.30.40.50.60.70.80.91V bias00.20.40.6G(V)γA = 0.0γA = 0.14 - 1 levelγA = 0.10 - 1 levelγA = 0.07 - 1 levelγA = 0.035 - 1 level0.50.60.7V bias0.050.10.150.2(c)Non-equilibrium many-body transport 14 (Color online) (Left Panel) Dynamical conductance G(V ) for the off- Figure 6. resonant transport regime (ε0 = 0.50) and different norminal coupling strength t0R = t0L to the leads. (Right Panel) Dynamical conductance G(V ) rescaled such that the amplitude of the main resonance peak is 1 quantum of conductance. The other parameters are γA = 0.2, γA = 0.14 (unless otherwise indicated), ω0 = 0.20, γ0 = 0.15, βα = 2.0,
α = 0.0, ηV = 1. applied bias for which features are present in both the spectral function of the central region C and the generalised embedding potential of the L lead. The conductance is affected by the NE renormalisation of the contact: the amplitude of the main resonance peak is modified as well as the lineshape of the first vibron side-band peak. Finally, extensions of the present study are now considered to go beyond the quasi- particle approach for the crossing interaction self-energy by including as well as the lesser and greater components of the LC interface self-energy. This should lead to dynamical NE renormalisation of the contact involving inelastic scattering processes. References [1] Widawsky J R, Darancet P, Neaton J B and Venkataraman L 2012 Nano Letters 12 354 [2] Strange M, Rostgaard C, Hakkinen H and Thygesen K S 2011 Physical Review B 83 115108 [3] Rangel T, Ferretti A, Trevisanutto P E, Olevano V and Rignanese G M 2011 Phys. Rev. B 84 045426 [4] Darancet P, Ferretti A, Mayou D and Olevano V 2007 Phys. Rev. B 75 075102 [5] Baym G 1962 Physical Review 127 1391 [6] von Barth U, Dahlen N E, van Leeuwen R and Stefanucci G 2005 Physical Review B 72 235109 [7] van Leeuwen R, Dahlen N E, Stefanucci G, Almbladh C O and von Barth U 2006 Lecture Notes in Physics 706 33 [8] Kita T 2010 Progress of Theoretical Physics 123 581 [9] Tran M T 2008 Physical Review B 78 125103 [10] Myohanen P, Stan A, Stefanucci G and van Leeuwen R 2008 EuroPhysics Letters 84 67001 [11] Myohanen P, Stan A, Stefanucci G and van Leeuwen R 2010 Journal of Physics: Conference Series 220 012017 [12] Perfetto E, Stefanucci G and Cini M 2010 Physical Review Letters 105 156802 [13] Velick´y B, Kalvov´a A and Spicka V 2010 Physical Review B 81 235116 0.20.30.40.50.60.70.80.9V bias00.10.20.30.40.50.60.7G(V)t0L = 0.22 (γA = 0)t0L = 0.22t0L = 0.20 (γA = 0)t0L = 0.20t0L = 0.15 (γA = 0)t0L = 0.15t0L = 0.10 (γA = 0)t0L = 0.100.20.30.40.50.60.70.8V bias00.20.40.60.81G(V) / GmaxNon-equilibrium many-body transport 15 [14] von Friesen M P, Verdozzi C and Almbladh C O 2009 Physical Review Letters 103 176404 [15] Arroyo C R, Frederiksen T, Rubio-Bollinger G, V´elez M, Arnau A, S´anchez-Portal D and Agraıt N 2010 Physical Review B 81 075405 [16] Ness H and Fisher A J 1999 Physical Review Letters 83 452 [17] Ness H, Shevlin S A and Fisher A J 2001 Physical Review B 63 125422 [18] Ness H and Fisher A J 2002 Europhysics Letters 57 885 [19] Mii T, Tikhodeev S and Ueba H 2003 Physical Review B 68 205406 [20] Montgomery M J, Hoekstra J, Sutton A P and Todorov T N 2003 Journal of Physics: Condensed Matter 15 731 [21] Troisi A, Ratner M A and Nitzan A 2003 Journal of Chemical Physics 118 6072 [22] Chen Y C, Zwolak M and di Ventra M 2005 Nano Letters 4 1709 [23] Lorente N and Persson M 2000 Physical Review Letters 85 2997 [24] Frederiksen T, Brandbyge M, Lorente N and Jauho A P 2004 Physical Review Letters 93 256601 [25] Galperin M, Ratner M A and Nitzan A 2004 Nano Letters 4 1605 [26] Mitra A, Aleiner I and Millis A J 2004 Physical Review B 69 245302 [27] Pecchia A, di Carlo A, Gagliardi A, Sanna S, Frauenhein T and Gutierrez R 2004 Nano Letters 4 2109 [28] Chen Z, Lu R and Zhu B 2005 Physical Review B 71 165324 [29] Ryndyk D A and Keller J 2005 Physical Review B 71 073305 [30] Sergueev N, Roubtsov D and Guo H 2005 Physical Review Letters 95 146803 [31] Viljas J K, Cuevas J C, Pauly F and Hafner M 2005 Physical Review B 72 245415 [32] Yamamoto T, Watanabe K and Watanabe S 2005 Physical Review Letters 95 065501 [33] Cresti A, Grosso G and Parravicini G P 2006 Journal of Physics: Condensed Matter 18 10059 [34] Han J E 2006 Physical Review B 73 125319 [35] Troisi A and Ratner M A 2006 Nano Letters 6 1784 [36] de la Vega L, Mart´ın-Rodero A, Agraıt N and Levy-Yeyati A 2006 Physical Review B 73 075428 [37] Frederiksen T, Paulsson M, Brandbyge M and Jauho A P 2007 Physical Review B 75 205413 [38] Galperin M, Ratner M A and Nitzan A 2007 Journal of Physics: Condensed Matter 19 103201 [39] Ryndyk D A and Cuniberti G 2007 Physical Review B 76 155430 [40] Schmidt B B, Hettler M H and Schon G 2007 Physical Review B 75 115125 [41] Troisi A, Beebe J M, Picraux L B, van Zee R D, Stewart D R, Ratner M A and Kushmerick J G 2007 Proceedings of the National Academy of Sciences of the USA 104 14255 [42] Asai Y 2008 Physical Review B 78 045434 [43] Benesch C, C´ızek M, Klimes J, Kondov I, Thoss M and Domcke W 2008 Journal of Physical Chemistry C 112 9880 [44] Paulsson M, Frederiksen T, Ueba H, Lorente N and Brandbyge M 2008 Physical Review Letters 100 226604 [45] Egger R and Gogolin A O 2008 Physical Review B 77 113405 [46] Monturet S and Lorente N 2008 Physical Review B 78 035445 [47] McEniry E J, Frederiksen T, Todorov T N, Dundas D and Horsfield A P 2008 Physical Review B 78 035446 [48] Schmidt B B, Hettler M H and Schon G 2008 Physical Review B (Condensed Matter and Materials Physics) 77 165337 [49] Secker D, Wagner S, Ballmann S, Hartle R, Thoss M and Weber H B 2011 Physical Review Letters 106 136807 [50] Hartle R, Butzin M, Rubio-Pons O and Thoss M 2011 Physical Review Letters 107 046802 [51] Ness H and Dash L K 2012 Physical Review Letters 108 126401 [52] Ness H and Dash L K 2011 Physical Review B 84 235428 [53] Perfetto E, Stefanucci G and Cini M 2012 Physical Review B 85 165437 [54] Ness H and Dash L K 2012 Journal of Physics A: Mathematical and Theoretical 45 195301 [55] Meir Y and Wingreen N S 1992 Physical Review Letters 68 2512 Non-equilibrium many-body transport 16 [56] Dash L K, Ness H and Godby R W 2010 Journal of Chemical Physics 132 104113 [57] Ness H, Dash L K and Godby R W 2010 Physical Review B 82 085426 [58] Dash L K, Ness H and Godby R W 2011 Physical Review B 84 085433 [59] Heeger A J, Kivelson S, Schrieffer J R and Su W P 1988 Review of Modern Physics 60 781 [60] Datta S, Tian W D, Hong S H, Reifenberger R, Henderson J I and Kubiak C P 1997 Physical Review Letters 79 2530 [61] Ness H and Dash L K 2012 submitted [62] Ciuchi S, de Pasquale F, Fratini S and Feinberg D 1997 Physical Review B 56 4494 -- 4512 [63] Ness H 2006 Journal of Physics: Condensed Matter 18 6307 [64] Cini M and D'Andrea A 1988 Journal of Physics C: Solid State Physics 21 193 [65] Dash L K, Ness H, Verstraete M and Godby R W 2012 Journal of Chemical Physics 136 064708 

1307.6550

2

1307

Photonic density of states enhancement in finite graphene multilayers

We consider the optical properties of finite systems composed of a series of graphene sheets separated by thin dielectric layers. Because these systems respond as conductors to electric fields in the plane of the graphene sheets and as insulators to perpendicular electric fields, they can be expected to have properties similar to those of hyperbolic metamaterials. We show that under typical experimental conditions graphene/dielectric multilayers have enhanced Purcell factors, and enhanced photonic densities-of-states in both the THz and mid-IR frequency range. These behaviors can be traced to the coupled plasmon modes of the multi-layer graphene system. We show that these results can be obtained with just a few layers of graphene.

cond-mat.mes-hall

cond-mat

Photonic density of states enhancement in finite graphene multilayers Ashley M. DaSilva,1 You-Chia Chang,2, 3 Ted Norris,3, 4 and Allan H. MacDonald1 1Department of Physics, The University of Texas at Austin, Austin, Texas 78712-1192, USA 2Applied Physics Program, The University of Michigan, Ann Arbor, Michigan 48109, USA 3Center for Ultrafast Optical Science, The University of Michigan, Ann Arbor, Michigan 48109, USA 4Department of Electrical Engineering and Computer Science, The University of Michigan, Ann Arbor, Michigan 48109, USA We consider the optical properties of finite systems composed of a series of graphene sheets separated by thin dielectric layers. Because these systems respond as conductors to electric fields in the plane of the graphene sheets and as insulators to perpendicular electric fields, they can be expected to have properties similar to those of hyperbolic metamaterials. We show that under typical experimental conditions graphene/dielectric multilayers have enhanced Purcell factors, and enhanced photonic densities-of-states in both the THz and mid-IR frequency range. These behaviors can be traced to the coupled plasmon modes of the multi-layer graphene system. We show that these results can be obtained with just a few layers of graphene. I. INTRODUCTION Hyperbolic metamaterials (HMMs) are artificially structured materials which have hyperbolic light dis- persion, leading to an enhanced photonic density-of- states.1 -- 5 One approach that is used to design a HMM is to consider superlattices with alternating metal and dielectric layers and sub-wavelength periods. When de- scribed as a homogenous material using an effective medium approximation, the dielectric constants of such a superlattice are:6,7 ǫk =ρǫd + (1 − ρ)ǫm ǫm (cid:19)−1 ǫ⊥ =(cid:18) ρ 1 − ρ ǫd + (1) (2) where k and ⊥ refer to the directions parallel and per- pendicular to the interfaces, ρ is the dielectric to metal thickness ratio, and ǫd(m) is the dielectric function of the dielectric (metal) constituent. By proper choice of mate- rials, thickness ratio, and frequency, one can engineer a material with a hyperbolic dispersion relation, i.e. a sys- tem in which ǫk and ǫ⊥ have opposite signs over a wide range of frequencies. HMMs are important in photonic engineering,8 espe- cially in applications to sub-wavelength imaging9,10 and confinement.11 Near-field thermal properties may also be engineered using HMMs for applications such as energy harvesting and thermal management.12 One can also use HMMs to control luminescence via the Purcell effect,4,13 which reflects the dependence of spontaneous emission on the surrounding density of photonic states. In free space, the density of photonic states is proportional to ω2. In the presence of an interface, the density of states can be enhanced by evanescent modes close to the inter- face. The size of the enhancement can be a few orders of magnitude, with most of the states localized close to the interface, leading to a more rapid excited state decay and enhanced photoluminescence of nearby atoms; this is the origin of the Purcell enhancement factor. HMMs can have particularly strongly enhanced photonic densities- of-states. There are two types of HMMs depending on which components of the dielectric tensor are negative. Type I HMMs have a metal-like perpendicular dielectric con- stant (ǫ⊥ < 0, ǫk > 0) while type II HMMs have a metal-like parallel dielectric constant (ǫk < 0, ǫ⊥ > 0). As an alternative to the alternating layer strategy,5,14 HMMs can also be constructed by embedding metallic nano-wires in a dielectric medium.15,16 Graphene, a monolayer of graphite, has long-lived long-wavelength plasmons which are tunable via gate voltage.17 -- 21 The possibility of gate tuning is one attrac- tive feature of using graphene for the metallic layers in a HMM. Indeed, infinite graphene/dielectric stacks have recently been predicted to have a large Purcell factor and a negative22 ǫk and arrays of graphene ribbons have been predicted to perform favorably as a hyperlens.23 Recent calculations of Fresnel coefficients and power spectra in the THz frequency regime provide further evidence of graphene's suitability as a component of HMMs.24 Here we report a study of finite stacks of graphene layers in both the THz and mid-IR frequency ranges, and show that even for a small number of layers graphene/deielctric stacks retain desirable HMM properties, in particular an enhanced photonic density of states.25 We empha- size that only in systems composed of a small number of graphene sheets will it actually be possible to modify the carrier densities and hence the plasmon frequencies of individual graphene layers via the electric field effect.26 (Screening prevents a back gate from influencing layers far from the substrate.27) We have also found that while having a dielectric be- tween the graphene layers is important in order to pre- vent interlayer tunneling, its direct role in modulating optical properties in tuning HMM effects for electromag- netic radiation in the THz regime is minimal. We find that graphene HMMs boast a large photonic density of states enhancement for a wide range of frequencies, and that the properties are robust to the dielectric spacer thickness, Fermi energy, and elastic mean free path. We focus on wavevector-resolved transmission coefficient and photonic density of states, which show the presence of modes within the metamaterial which are evanescent in free space. The enlarged photonic density of states leads to a Purcell enhancement that is greatly improved rela- tive to metal/dielectric layered HMMs. We note that we have assumed equal carrier densities in all of the graphene layers. This is an unrealistic ap- proximation for multilayer graphene samples utilizing the elctric field effect to control the carrier density.28 How- ever, is possible to achieve approximately equal carrier densities by using doping techniques,29,30 and in this case we expect our assumption to be valid. Our paper is organized as follows. We first describe the electromagnetic Green's functions and transfer matrices we use to perform calculations. We then characterize graphene HMMs by evaluating the reflection coefficients, showing that the anticipated HMM features are already realized at quite small graphene layer numbers. We then calculate the transmission coefficient and wavevector- resolved photonic density of states for a N = 6-layer graphene-based HMM to illustrate the dependence of var- ious properties on controllable parameters. Finally, we evaluate the photonic density of states and Purcell coef- ficient for finite graphene-based HMMs in both the THz and the mid-IR regime. II. THEORETICAL FORMULATION Maxwell's equations in a uniform medium with dielec- tric constant ǫ and permittivity µ = 1 are conveniently solved using electromagnetic Green's functions defined by the differential equation31,32 ∇ × ∇ × EM ↔ G 0 (r) − ǫω2 c2 ↔ G EM 0 (r) = δ(r). (3) This function is called the Diadic Green's function to reflect the property that a source oriented in one direction 2 can in general result in electric and magnetic fields in any direction. In free space, the electric and magnetic Green's functions are the same. The electric and magnetic fields are obtained by integrating the electromagnetic Green's functions over the sources: E(r) = 4πiω c2 Z dr′ EM H(r) =Z dr′ ↔ G 0 EM ↔ G 0 (r − r′)J(r′) (r − r′)M(r′) (4) (5) where J includes all free currents and M(r) = (4π/c)∇× J(r) can be thought of as the magnetization they pro- duce. The electromagnetic local density of states (LDOS) can also be calculated from the Green's function,31 -- 33 ω πc2 Im tr ↔ G ρ(ω, z) = (6) where Im denotes the imaginary part and tr denotes the trace. In our planar geometry ρ depends only on the coordinate z which measures position relative to the graphene/dielectric multilayer. In a non-uniform medium, the electric and magnetic Green's functions will be different, and the total density of states is the sum of the electric and magnetic components, ρ(ω, z) = ρE(ω, z) + ρH (ω, z).31 The Green's functions defined in Eqn. (3) are those of a uniform medium, while the expression for the LDOS (Eqn. (6)) depends on the Green's function in the non- uniform medium. We will assume that the top surface of an HMM of total thickness L is located at z = 0, and the regions z > 0 and z < −L is free space (ǫ = 1). The presence of the HMM can then be accounted for by writing the total electromagnetic field at z > 0 as the sum of the incident and reflected parts. We find that the electric Green's function for z > 0 can then be written as E ↔ G (k, z, z′; ω) = i 2K h(ss + p− p−)e−iK(z−z ′)θ(z′ − z) + (ss + p+ p+)eiK(z−z ′)θ(z − z′) + (rs ss + rp p+ p−)eiK(z+z ′)i(7) where k is the two-dimensional in-plane wavevector, K = pǫω2/c2 − k2 is the out-of-plane wavevector, and rα for α = s, p are the reflection coefficients for the two polar- izations of EM waves. Here and below we set ǫ to 1 for a HMM embedded in a vacuum. The polarization vectors are s = p± = 1 (ky x − kx y) k √k2 + K 2 1 (∓K k + kz) denoting s (TE) and p (TM) polarized light. The ± index on p distinguishes upward moving and downward 3 is the transport time (which depends on the mobility), and θ(x) is the step function which specifies the thresh- old for interband transitions at large ω. This expres- sion ignores the nonlocal response, and is appropriate for k → 0. In practice, this expression is found to work well away from the onset of interband transitions, which oc- curs when ω ≈ 2EF .40 Inserting a transfer matrix for each graphene layer, and one propagation matrix Pi =(cid:18) e−iQidi 0 0 eiQidi (cid:19) (16) for each dielectric layer of thickness di, the total transfer matrix is a product of the component matrices, M s(p) =Yj PjM s(p) j−1,j (17) The reflection coefficient, rs(p), and transmission coeffi- cient, ts(p), are obtained from these expressions by solv- ing 1 (cid:19) ts(p) (cid:19) = M s(p)(cid:18) rs(p) (cid:18) 0 (18) III. OPTICAL PROPERTIES The Purcell enhancement factor is defined as the ratio of the total radiation rate of a unit dipole source to the radiation rate of the dipole in vacuum:7,41 b =1 + 3 2ω/c Re(cid:26)Z ∞ 0 dk k K (cid:20)f 2 ⊥ k2rp ω2/c2 + + 1 2 f 2 k (cid:18)rs − K 2 ω2/c2 rp(cid:19)(cid:21) e2iKz(cid:27) (19) where z is the surface to dipole distance, and fk and f⊥ are the components of the dipole along the directions par- allel and perpendicular to the HMM layers, respectively. In the effective medium approximation, the enhanced Purcell effect can be traced to the nonzero imaginary part of the reflection coefficient in the large k limit, k/(ω/c) → ∞.7 In real systems, the finite period of the HMM limits the maximum value of k, but the signature of a nonzero imaginary part of the reflection coefficient for k > ω/c remains. These modes are evanescent in the vacuum on either side of the system, but propagate within the structure, as demonstrated by the enhanced transmission coefficient.42 In Fig 1 we plot the reflection coefficients for s- and p- polarized light as a function of wavevector normalized to frequency. For s-polarized light, the number of graphene layers has little effect. For p-polarized light, a greater number of layers leads to more peaks of smaller magni- tude while the general features remain intact, including the presence of a nonzero imaginary part of the reflection coefficient up to k ∼ 200ω/c. For the same parameters, figure 2 shows the transmittance vs parallel wave vector. σ(ω) = 2e2 h (cid:20)i εF ω + i/τ + i 4 (15) Here vD is the Dirac velocity of graphene, εF = vD√πn is the Fermi energy as a function of carrier density n, τ moving waves. The magnetic Green's function can be obtained from the electric Green's function by replacing rs ↔ rp.31 We now see that the Green's function for z > 0, and therefore the LDOS, is determined solely by the reflection coefficients. The LDOS is 4k2 k ω 2π ρ(ω, z) = 2K (cid:20)4 + πc2 Re(cid:26)Z dk ǫω2/c2 (rs + rp) e2iKz(cid:21)(cid:27) where K = pω2/c2 − k2 and ǫ = 1 when the HMM is in vacuum. The reflection coefficients, rs and rp, are functions of k and are explicitly provided below. (8) Below we also consider the wave-vector resolved LDOS, ρ(q, ω, z), which separates contributions to the LDOS from different wavevectors: ρ(k, ω, z) = ω πc2 Re(cid:26) 2 K (cid:20)1 + c2k2 ǫω2 (rs + rp)e2iKz(cid:21)(cid:27)(9) The vacuum LDOS can be recovered by setting the re- flection coefficients to zero: ρ0(k, ω, z) ≡ ρ0(k, ω) = ω2 π2c3 ρ0(ω, z) ≡ ρ0(ω) = 2ω πc2 θ(ω/c − k) pǫω2/c2 − k2 (10) (11) which are independent of z. Note that in the absence of an interface, there is no contribution to the DOS from wavevectors k > ω/c. The reflection coefficients rs and rp for s- and p- polarized light, respectively, are defined as the ratio of the reflected to the incident electromagnetic field at the interface. They are determined entirely by the boundary conditions imposed by Maxwell's equations: E2↓ (cid:19) = M s(p) (cid:18) E2↑ E1↓ (cid:19) 12 (cid:18) E1↑ (12) where the indices 1 and 2 refer to the region above and below a graphene sheet, and ↑ and ↓ denote the upward moving and downward moving modes, respectively. The matrices connecting the fields are, M s 12 = M p 12 = 1 Q2 2 1 + Q1 1 − Q1 2 1 + ǫ2Q1 1 − ǫ2Q1 Q2 − 4πσω c2Q2 + 4πσω c2Q2 + 4πσQ1 ωǫ1 + 4πσQ1 ωǫ1 ǫ1Q2 ǫ1Q2 1 1 − Q1 1 + Q1 Q2 c2Q2 ! Q2 − 4πσω c2Q2 + 4πσω ǫ1Q2 − 4πσQ1 ǫ1Q2 − 4πσQ1 1 − ǫ2Q1 1 + ǫ2Q1 ωǫ1 ωǫ1 ! . (13) (14) The two-dimensional conductivity of a graphene is given by34 -- 40 ln(cid:12)(cid:12)(cid:12)(cid:12) ω − 2εF ω + 2εF(cid:12)(cid:12)(cid:12)(cid:12) + π 4 θ(ω − 2εF )(cid:21) (a) s r e R (c) s r m I 0.0 −0.2 −0.4 −0.6 −0.8 0.0 −0.1 −0.2 −0.3 −0.4 −0.5 0 (b) p r e R 1.20 1.15 1.10 1.05 1.00 0.95 (d) 0.06 p r m I 0.04 0.02 N=1 N=4 N=6 N=12 20 10 k/(ω/c) 30 0.00 0 100 200 300 k/(ω/c) FIG. 1. Real (top panels) and imaginary (bottom panels) parts of the reflection amplitudes for s (left panels) and p (right panels) polarized EM waves on graphene HMMs with different numbers of layers, N . The parameters used for this calculation were frequency f = 1.0 THz (ω = 4.1 meV), n = 4 × 1012 cm−2 (Fermi energy EF = 0.23 eV) in every layer, µ = 50, 000 cm2/Vs for graphene, d = 10 nm for the dielectric layer thicknesses and ǫ = 3.9 for their dielectric constant. Various values of frequency ω are considered. For both s- and p-polarized light, a significant fraction of the elec- tromagnetic wave is transferred through the structure at k < ω/c. This is due to the fact that the wavelength is much larger than the total thickness, λ ≫ N d, for N = 6 layers each of thickness d = 10 nm. We also observe fi- nite transmission when k > ω/c for both polarizations. As the frequency increases, the high-k transmission coef- ficient decays more rapidly for s-polarized radiation com- pared to p-polarized radiation. This decay is due to the decreased wavelength, and is similar to the behavior of a uniform dielectric. On the other hand, the transmis- sion coefficient for p-polarized radiation has several sharp peaks when k > ω/c. A large value of tp2 corresponds to t = E2↓/E1↓ > 1. When k > ω/c, the ↑ (↓) labels cor- respond to the evanescent modes which decay to zero for z → −∞ and z → +∞, respectively. Any electromag- netic mode with k > ω/c must decay in the free space on either side of the HMM, so E1↑ → 0, yielding peaks at electromagnetic modes of the HMM. The peaks in transmission for p-polarized radiation are obtained in the same regime where there are peaks in the imaginary part of the reflection coefficient. (See Fig. 2.) Such features are not observed for s-polarized ra- diation which does not excite plasmon resonances since the electric field and parallel wavevector are perpendic- ular. These peaks reflect the coupled plasmon modes in the graphene layers,43 which have energies bounded by the dashed black lines in Fig 2 (b). These modes have been predicted and observed in 2DEG superlattices (a) 2 p t 103 101 10−1 10−3 10−5 (b) / F ħ ω ħ 0 100 200 300 k/(ω/c) 1.2 1.0 0.8 0.6 0.4 0.2 0.0 0.0 0.3 0.6 0.9 1.2 k/kF 4 2 p t g o l 2 1 0 −1 −2 −3 −4 −5 FIG. 2. (a) Magnitude squared of the transmission ampli- tudes for p-polarized light on 6 layers of graphene HMM with the same material parameters as in Fig. 1 and ω ranging from ω = 4.0 meV (f = 1.0 THz) to ω = 165 meV (f = 40 THz), corresponding to the frequencies shown as colored lines in (b). (b) The logarithm of the magnitude squared of the transmis- sion coefficient is shown as a function of frequency in units of Fermi energy and wavevector in units of Fermi wavevec- tor. The Fermi energy is 233 meV. Overlayed on this plot are the threshold frequencies for inter band transitions (solid grey lines) and the frequency range of bulk plasmons in graphene superlattices (solid black lines). previously,44 -- 48 and are usually discussed in terms of in- stantaneous intra and inter layer Coulomb interactions, an approximation that is reliable in the large k regime. In the context of HMMs these are sometimes called high-k propagating modes42 and have application in subwave- length confinement and imaging.9 As the frequency is in- creased, the transmission of p-polarized radiation is en- hanced, with an optimal value at around 40 THz (165 meV), close to the Fermi energy. For frequencies above the Fermi energy, the photon energy is above the maxi- mum plasmon energy,43,45,46,48 and all high-k modes dis- appear. The desirable properties of HMMs stem from their enhanced LDOS at wavevector k > ω/c. The LDOS will be a function of frequency ω and distance z above the graphene HMM. The wavevector-resolved LDOS is shown in Fig. 3 for the same parameters as in Fig. 1 and Fig. 2 and for two different values of the distance z. As expected, the LDOS at large z (dashed lines) de- cays more rapidly at large wavevector, due to the weak influence of evanescent modes far from the surface of the HMM. We also notice that the LDOS enhancement is greater for smaller frequencies, in spite of the opposite trend in the transmission coefficient for p-polarization (See Fig. 2 (b)). This is also an expected trend, and is due to the 1/ω dependence of the LDOS which is ap- parent in the Purcell factor, Eqn. (19). In addition to the overall larger LDOS, the peaks in the LDOS are smeared out for smaller frequencies because τ −1 ≈ 0.7 meV is held constant. One must be careful at large wavevectors (k ≈ kF ) where the local approximation for the conductivity of graphene, Eqn. (15), becomes questionable.34 We have found that the LDOS decays before reaching k = kF for frequencies above ≈ 4 meV at 10 nm, and for all frequen- cies studied here at 1 µm. At smaller frequencies and dis- 10 8 6 4 2 ) m n ( d 200 150 100 50 ) V e m ( F E ) V e m ( 1 − τ ħ 12 10 8 6 4 2 (a) (b) (c) ) ω ( 0 ρ / ) z , ω , k ( ρ 106 105 104 103 102 101 100 ħω (meV) 1.03 2.07 3.10 4.13 6.20 10.34 200 250 300 0 50 100 150 k/(ω/c) FIG. 3. The LDOS for a 6 layer graphene HMM with the same material parameters as Fig. 1 and different values of ω ranging from from ω = 1.03 meV (f = 0.25 THz) to ω = 10.34 meV (f = 2.5 THz), corresponding to the frequencies shown in the legend. The solid and dashed lines correspond to distances of z = 10 nm and z = 1000 nm above the surface of the HMM, respectively. Values of (ω/c)ℓmf p ≈ 0.003 to 0.03 for frequencies shown on this plot. tances close to the interface the calculation may become unreliable unless nonlocal corrections to the conductivity are made. Nonlocal effects are known to provide a large wavevector cutoff of the LDOS enhancement.49 In the ballistic limit, kℓmf p ≫ 1 (ℓmf p is the elastic mean free path for electrons), nonlocal effects are important when ω > vDk where vD is the Dirac velocity of electrons. This limit reduces to k/(ω/c) < 300. The other parameters in the model are the period of the graphene/dielectric superlattice, the carrier density, and the mobility of the graphene sheets. For simplicity, we have assumed that all graphene sheets have the same carrier density and mobility. The former assumption is unrealistic when carriers are induced by gates, as men- tioned previously, but does not influence properties in an essential way. Fig. 4 (a) shows the dependence of the LDOS on both wavevector and period, d. The calcula- tions were performed for a distance of z = 50 nm from the surface. The LDOS is only weakly dependent on d for features at small k; this is because for the THz regime under consideration, d is always much smaller than λ (λ ∼ 100 µm for the frequencies studied here). However there is a noticeable enhancement of the high- k features in the LDOS as d decreases due to the de- pendence of the transfer matrix on the combination kd, which becomes comparable to 1 at k/(ω/c) ∼ 500 for d = 10 nm and k/(ω/c) ∼ 5000 for d = 1 nm. Fig. 4 (b) shows the logarithm of the LDOS vs. Fermi energy and k/(ω/c). The dependence again is weak, but as the Fermi energy decreases, there is an enhancement when ω be- comes comparable to EF . For smaller Fermi energies, we expect that our local approximation for σ(ω) breaks down. In realistic multilayers that have carriers induced 5 5 0 ρ / ρ g o l 4 50 100 150 200 250 300 k/(ω/c) 3 5 0 ρ / ρ g o l 4 50 100 150 200 250 300 k/(ω/c) 3 5 0 ρ / ρ g o l 4 3 0 50 100 150 200 250 300 k/(ω/c) FIG. 4. The logarithm of the LDOS at z = 50 nm as a function of parallel wavevector divided by vacuum wavevec- tor, k/(ω/c), and (a) the distance between graphene sheets, d, for fixed Fermi energy EF = 234 meV and relaxation time τ = 10−12 s; (b) Fermi energy, EF , for fixed d = 10 nm and τ = 10−12 s; (c) inverse relaxation time, τ −1 for fixed EF = 234 eV and d = 10 nm. For all plots, ω = 4.1 meV and there are six layers of graphene separated by a dielectric with ǫ = 3.9. The dashed line in (c) corresponds to kℓmf p = 1. In (a) and (b), kℓmf p = 3.7 on the right hand side of the figure. by gates, the density will normally be low in some layers. Fig. 4 (c) shows the logarithm of the LDOS vs τ −1 and k/(ω/c). Again, the dependence is relatively weak, how- ever there is a noticeable enhancement of the LDOS as τ decreases. This is probably due to stronger absorption in the graphene planes as the real part of the conductivity becomes larger. ) ω ( 0 ρ / ) z , ω ( ρ 109 108 107 106 105 104 103 102 101 100 102 101 ) V e m ( ω ħ 0 200 400 600 z (nm) 100 200 400 600 z (nm) 1.03 10.34 ħω (meV) 51.68 103.36 206.72 800 1000 8 7 6 5 4 3 2 1 0 0 ρ / ρ g o l 800 FIG. 5. (a) The integrated LDOS as a function of z for several values of ω as indicated in the legend. (b) The logarithm of the integrated LDOS as a function of ω and z. For both (a) and (b), the system is six layers of graphene with n = 4 ×1012 cm−2 and µ = 50, 000 cm2/Vs separated by 10 nm of dielectric with ǫ = 3.9. 107 106 105 104 103 r o t c a f l l e c r u P 102 0 200 r o t c a f l l e c r u P 102 101 100 metal/dielectric HMM metal film 0 20 40 60 z (nm) 80 100 400 600 800 1000 z (nm) FIG. 6. The Purcell factor as a function of z for a graphene HMM. The inset shows the Purcell factor as a function of z for a metal/dielectric HMM and a metallic film. All three have 100 nm total thickness. The graphene HMM is composed of 10 layers of 10 nm each, having the same parameters as in Fig. 5 and at ω = 4.1 meV. The metal/dielectric HMM has 10 unit cells each composed of 5 nm of metal (ǫm = −27.5 + 0.31i) and 5 nm of dielectric (ǫd = 6.7), and is at a frequency of ω = 1.65 eV. The metallic film has the parameters of the metallic part of the metal/dielectric HMM and is at the same frequency. In both the main figure and the inset, the solid lines are for a dipole oriented parallel to the HMM layers and the dashed lines are for a dipole perpendicular to the HMM layers. 6 The Purcell factor depends not on the wavevector- resolved LDOS, but on the total LDOS, integrated over all wavevectors. In Fig. 5(a) we show the integrated LDOS normalized to the vacuum LDOS vs z for different values of ω. As expected, the normalized LDOS decays away from the surface, and will reach a value of 1 for z ≫ λ. We also see that the normalized LDOS is larger for smaller values of ω. Fig. 5(b) shows the dependence of the normalized LDOS as a function of both z and ω for six layers of graphene with n = 4 × 1012 cm−2. We again note that for small ω our local approximation for the conductivity of graphene is questionable. Next we calculate the Purcell factor a 10 layer graphene HMM (total thickness 100 nm) for the param- eters ω = 4.1 meV, n = 4 × 1012 cm−2, µ = 50, 000 cm2/Vs, and ǫ = 3.9. Fig. 6 shows the Purcell factor calculated for the two orientations of a unit dipole: per- pendicular to the surface (dashed) and parallel to the surface (solid lines). In the inset of Fig. 6 we show the calculation of the Purcell factor for 100 nm of both a metal/dielectric HMM (green) and metallic film (red). The metal/dielectric HMM is composed of 10 unit cells of 5 nm of metal with dielectric constant ǫm = −27.5+0.31i, and 5 nm of dielectric with dielectric constant ǫd = 6.7 at a wavelength of 750 nm, parameters which are rele- vant for Ag/TiO2 multilayers which have been used as typical HMMs for experiments and theory.42,50,51 The metallic film has the parameters for the metal at the same wavelength. Our results for the Purcell enhancement vs dipole distance are in agreement with those obtained in Ref. 24 for a semi-infinite graphene-based HMM. We find that the Purcell factor for graphene is ehnanced at small distances compared to both the metal/dielectric HMMs and the metal films. However, one must note that the wavelength of the metal/dielectric HMM is necessarily different than the wavelength of the graphene HMMs. Metal/dielectric HMMs are limited to a frequency regime where the metal has a negative dielectric response (typi- cally in the optical region of the electromagnetic spec- trum), while the graphene HMMs are limited by the Fermi energy (typically in the THz to mid-IR region.) Note for example, at z = 1 nm the metal/dielectric HMM has a Purcell factor ∼ 102. Keeping the ratio of z to free space wavelength λ = 2πc/ω for the two metamaterials the same, the corresponding value of z for the graphene- based HMM is z ≈ 400 nm. We observe that the Purcell factor is ∼ 103, a factor of 10 enhancement compared to the metal/dielectric HMM. One obtains a large Purcell enhancement at small dis- tances for small numbers of graphene layers. Fig. 7 shows a comparison of the Purcell factor for N = 1, 2, 4, 6, and 10 layers of graphene. For one or two graphene lay- ers the Purcell factor decays more rapidly. By the time there are 4 graphene layers, the curves are almost identi- cal up to z = 1000 nm. This shows that graphene-based hyperbolic metamaterials are possible for few layers of graphene, as low as N = 4. So far our calculations are for high mobility graphene ) ∥ ( r o t c a f l l e c r u P 107 106 105 104 103 102 101 100 10−1 0 200 number of layers 6 10 1 2 4 109 108 107 106 105 104 103 102 101 1000 7 177 ħω (meV) 248 413 620 (a) 105 ) ω ( 0 ρ / ) ω ( ρ 104 103 102 101 ) ⟂ ( r o t c a f l l e c r u P 400 z (nm) 600 800 FIG. 7. The Purcell factor as a function of z for a graphene HMM of 1, 2, 4, 6, and 10 layers, as denoted in the legend. The Purcell factors for parallel and perpendicular dipoles have been plotted against different axes for clarity. Solid lines cor- respond to a dipole oriented parallel to the surface, with Pur- cell enhancement given on the left axis, while dashed lines correspond to a dipole oriented perpendicular to the surface, with Purcell enhancement given on the right axis. The pa- rameters of the graphene and dielectric are the same as in Fig. 5 and at ω = 4.1 meV. layers typical of exfoliated samples. Graphene grown by chemical vapor deposition (CVD) tends to have a lower mobility around 1000 cm2/Vs. For such samples, it is beneficial to operate in the mid-infrared (mid-IR) regime, where the conductivity given by Eqn. 15 will still have a significant imaginary part despite the smaller relaxation time. Since the frequency regime for the HMM is limited by the Fermi level, it is necessary for such samples to be highly doped, to 1013 cm−2 or larger carrier density. We have calculated the integrated LDOS and Purcell factor for highly-doped low mobility graphene at mid-IR wave- lengths. Fig. 8 shows that large LDOS and Purcell factor enhancements are predicted for these parameters. The Purcell factor is 2 orders of magnitude less than for the high mobility graphene in the THz regime; this is par- tially attributable to the ratio of z/λ which is larger for mid-IR wavelengths. The Purcell factor remains, how- ever, improved over that of the metal/dielectric HMM. We observe a crossover between wavelengths of 3 and 5 µm (177 and 248 meV) indicated by the much slower de- cay of the LDOS and Purcell factor with z. We attribute this cross-over to the transition from elliptical to hyper- bolic isofrequency contour.14 A simple Bloch theory52 for an infinite graphene-based metamaterial22 shows that the transition from an effective permittivity ǫk > 0 to ǫk < 0 should occur in exactly this regime, while ǫ⊥ remains positive. We should stress here the non-negligible role of loss in the enhancement of the Purcell factor. A lossy material need not be hyperbolic in order to produce a large Pur- cell enhancement: for example the metallic film in the (b) r o t c a f l l e c r u P 100 105 104 103 102 101 100 10−1 0 50 100 150 200 250 300 350 400 0 50 100 150 200 250 300 350 400 z (nm) FIG. 8. Properties of a highly doped, low mobility graphene HMM in the mid-IR regime. In both panels,the graphene layers are separated by 10 nm of dielectric with ǫ = 3.9. The graphene layers have a carrier density of 5 × 1013 cm−2 and a mobility of µ = 1000 cm2/Vs. (a) The integrated local density of states as a function of position z above the top graphene layer. The wavelengths range from 3 to 8 µm, with energies ω indicated in the legend. (b) The Purcell factor of a unit dipole oriented parallel (solid) and perpendicular (dashed) to the surface of a 10-layer graphene metamaterial. The frequency of the oscillating dipole corresponds to those noted in the legend of panel (a). inset of Fig. 6 has a larger Purcell enhancement than a metal/dielectric HMM at small enough distances. For hyperbolic systems, instead we see an enhanced Purcell coefficient over a longer distance. IV. SUMMARY In conclusion, we have found that thin graphene stacks are HMMs in the THz to mid-infrared regime for a wide range of parameters. We have studied the high-k prop- agating modes as well as the wavevector resolved local density of states for graphene stacks and find an en- hancement of both quantities at wavevectors which are evanescent in vacuum. This implies that enhanced near- field effects including sub-wavelength imaging and con- finement of light may be possible. We also calculate the Purcell factor for both our graphene HMMs, and HMMs composed of metal/dielectric stacks. We find that the graphene HMMs perform very well compared to this 8 Division of Materials Sciences and Engineering under grant DE-FG03-02ER45958. Michigan researchers were supported by the NSF MRSEC program under DMR 1120923. benchmark at both the THz and mid-IR wavelengths. The frequency range of the graphene HMM is limited by the Fermi energy, ω . εF , and so the graphene must be highly doped for mid-IR applications. We observe a transition from high Purcell enhancement to low Purcell enhancement around 3-5 µm for the low mobility, highly doped graphene which we attribute to a transition from hyperbolic to elliptical isofrequency contour. V. ACKNOWLEDGMENTS Work at the University of Texas was supported by the Welch foundation under grant TBF1473 and the DOE 1 D. D. Physical Review Letters 90, 077405 (2003). Smith and R. Schurig, 2 I. K. Microwave and Optical Technology Letters 31, 129133 (2001). Tretyakov, Ilvonen, Nikoskinen, Lindell, V. I. and A. S. S. 21 Y. Nature Nanotechnology 7, 330 (2012). Ferreira, M. Bludov, A. V. Peres, International Journal of Modern Physics B 27, 1341001 (2013). and I. 22 I. V. Iorsh, I. S. Mukhin, I. V. Shadrivov, P. A. Belov, and N. M. R. Vasilevskiy, 3 I. I. Smolyaninov and E. E. Narimanov, Y. S. Kivshar, Physical Review B 87, 075416 (2013). Physical Review Letters 105, 067402 (2010). 23 A. Andryieuski, A. V. Lavrinenko, and D. N. Chigrin, 4 Z. Jacob, I. Smolyaninov, and E. Narimanov, Physical Review B 86, 121108 (2012). Applied Physics Letters 100, 181105 (2012). 24 M. A. K. Othman, C. Guclu, and F. Capolino, 5 Z. Jacob, J. Y. Kim, G. V. Naik, A. Boltas- and V. M. Shalaev, seva, E. E. Narimanov, Applied Physics B 100, 215 (2010). 6 A. Fang, T. Koschny, and C. M. Soukoulis, Physical Review B 79, 245127 (2009). 7 O. Kidwai, S. V. Zhukovsky, and J. E. Sipe, Physical Review A 85, 053842 (2012). V. Kildishev and V. 8 A. Optics Letters 33, 43 (2008). 9 A. A. V. Physical Review B 73, 155108 (2006). Govyadinov and M. Shalaev, Optics Express 21, 7614 (2013). 25 A. DaSilva and A. MacDonald, "Graphene multilayers as hyperbolic metamaterials," (2013), contributed talk at the APS March Meeting, http://meetings.aps.org/link/BAPS.2013.MAR.R5.12 26 K. S. Novoselov, A. K. Geim, S. V. Morozov, D. Jiang, Y. Zhang, S. V. Dubonos, I. V. Grigorieva, and A. A. Firsov, Science 306, 666 (2004). 27 S. S. Datta, D. R. Strachan, E. J. Mele, and A. T. C. 28 D. C. Sun, de Heer, Physical Review Letters 104, 136802 (2010). P. N. First, Berger, W. A. and T. B. Norris, Divin, C. A. Podolskiy, Johnson, Nano Letters 9, 7 (2009). 10 I. I. Smolyaninov, Y.-J. Hung, and C. C. Davis, Science 315, 1699 (2007). 11 Z. Liu, H. Lee, Y. Xiong, C. Sun, and X. Zhang, 29 T. Ohta, A. Bostwick, T. Seyller, K. Horn, and E. Roten- Science 315, 1686 (2007). 12 Y. Guo, C. L. Cortes, S. Molesky, and Z. Jacob, Applied Physics Letters 101, 131106 (2012). 13 J. Kim, V. P. Drachev, Z. Jacob, G. V. Naik, and V. M. Shalaev, A. Boltasseva, E. E. Narimanov, Optics Express 20, 8100 (2012). 14 H. N. S. Krishnamoorthy, Z. Jacob, E. Narimanov, I. Kret- zschmar, and V. M. Menon, Science 336, 205 (2012). 15 T. E. Applied Physics Letters 100, 161103 (2012). U. E. Narimanov, Kitur, and M. A. Noginov, Tumkur, Gu, K. L. J. berg, Science 313, 951 (2006). 30 J.-H. Chen, C. Jang, S. Adam, M. S. Fuhrer, E. D. Williams, and M. Ishigami, Nature Physics 4, 377 (2008). 31 K. Joulain, R. Carminati, J.-P. Mulet, and J.-J. Greffet, Physical Review B 68, 245405 (2003). 32 G. S. Agarwal, Physical Review A 11, 253 (1975). 33 C. Girard, Reports on Progress in Physics 68, 1883 (2005). 34 E. Sarma, S. Physical Review B 75, 205418 (2007). Hwang Das and H. 35 L. A. Falkovsky and S. S. Pershoguba, Physical Review B 76, 153410 (2007). 16 J. Kanungo and J. Schilling, 36 T. Stauber, N. M. R. Peres, and A. K. Geim, Applied Physics Letters 97, 021903 (2010). 17 L. Ju, B. Geng, J. Horng, C. Girit, M. Martin, Z. Hao, and H. A. Bechtel, X. Liang, A. Zettl, Y. R. Shen, F. Wang, Nature Nanotechnology 6, 630 (2011). 18 Q. Bao and K. P. Loh, ACS Nano 6, 3677 (2012). 19 A. N. Grigorenko, M. Polini, Nature Photonics 6, 749 (2012). and K. S. Novoselov, 20 H. Yan, X. Li, B. Chandra, G. Tulevski, Y. Wu, and F. Xia, M. Freitag, W. Zhu, P. Avouris, Physical Review B 78, 085432 (2008). 37 K. F. Mak, M. Y. J. A. Misewich, Lui, Physical Review Letters 101, 196405 (2008). and T. Sfeir, Y. Wu, C. H. F. Heinz, 38 S. H. Abedinpour, M. Polini, W.-K. Tse, Physical Review B 84, 045429 (2011). G. Vignale, A. Principi, and A. H. MacDonald, 39 M. Jablan, H. Buljan, and M. Soljacic, Physical Review B 80, 245435 (2009). 40 F. H. L. Koppens, D. E. Chang, and F. J. Garcia de Abajo, 48 G. F. Giuliani and J. J. Quinn, Nano Letters 11, 3370 (2011). 41 O. Kidwai, S. V. Zhukovsky, and J. E. Sipe, Optics Letters 36, 2530 (2011). Physical Review Letters 51, 919 (1983). 49 W. Yan, M. Wubs, arXiv:1204.5413 (2012). and N. A. Mortensen, 42 C. L. Cortes, W. Newman, S. Molesky, and Z. Jacob, Journal of Optics 14, 063001 (2012). 43 Kenneth W.-K. Shung, Physical Review B 34, 979 (1986). 44 D. Olego, A. Pinczuk, A. C. Gossard, and W. Wiegmann, 50 A. F. d. Silva, I. Pepe, C. Persson, J. S. d. Almeida, C. M. Arajo, R. Ahuja, B. Johansson, C. Y. An, and J.-H. Guo, Physica Scripta 2004, 180 (2004). 51 P. B. Johnson and R. W. Christy, 9 Physical Review B 25, 7867 (1982). 45 S. Das Sarma and J. Physical Review B 25, 7603 (1982). 46 J. K. Jain and P. Physical Review Letters 54, 2437 (1985). J. B. Quinn, Allen, 47 A. Pinczuk, M. G. Lamont, and A. C. Gossard, Physical Review Letters 56, 2092 (1986). Physical Review B 6, 4370 (1972). 52 N. and M. G. C. Cottam, Journal of Physics C: Solid State Physics 19, 739 (1986). Constantinou 

1801.05053

1

1801

Differential Poisson's ratio of a crystalline two-dimensional membrane

We compute the differential Poisson's ratio of a suspended two-dimensional crystalline membrane embedded into a space of large dimensionality $d \gg 1$. We demonstrate that, in the regime of anomalous Hooke's law, the differential Poisson's ratio approaches a universal value determined solely by the spatial dimensionality $d_c$, with a power-law expansion $\nu = -1/3 + 0.016/d_c + O(1/d_c^2)$, where $d_c=d-2$. Thus, the value $-1/3$ predicted in previous literature holds only in the limit $d_c\to \infty$.

cond-mat.mes-hall

cond-mat

Differential Poisson's ratio of a crystalline two-dimensional membrane I.S. Burmistrova,b,c,d,∗, V. Yu. Kachorovskiie,c,a, I.V. Gornyic,d,e,a, A.D. Mirlinc,d,f,a aL.D. Landau Institute for Theoretical Physics, Kosygina street 2, 119334 Moscow, Russia bLaboratory for Condensed Matter Physics, National Research University Higher School of Economics, 101000 Moscow, Russia cInstitut fur Nanotechnologie, Karlsruhe Institute of Technology, 76021 Karlsruhe, Germany dInstitut fur Theorie der kondensierten Materie, Karlsruhe Institute of Technology, 76128 Karlsruhe, Germany eA. F. Ioffe Physico-Technical Institute, 194021 St. Petersburg, Russia fPetersburg Nuclear Physics Institute, 188300, St.Petersburg, Russia Abstract We compute the differential Poisson's ratio of a suspended two-dimensional crystalline mem- brane embedded into a space of large dimensionality d (cid:29) 1. We demonstrate that, in the regime of anomalous Hooke's law, the differential Poisson's ratio approaches a univer- sal value determined solely by the spatial dimensionality dc, with a power-law expansion ν = −1/3 + 0.016/dc + O(1/d2 c), where dc = d − 2. Thus, the value −1/3 predicted in pre- vious literature holds only in the limit dc → ∞. Keywords: crystalline membrane, Poisson's ratio 1. Introduction Poisson's ratio is defined as the ratio of a transverse compression to a longitudinal stretching. In the classical theory of elasticity, the Poisson's ratio is given by νcl = λ 2µ + (D − 1)λ , where µ and λ are the Lam´e coefficients and D is the dimensionality of the elastic body [1]. General conditions of thermodynamic stability restrict the Poisson's ratio to the range between −1 and 1/(D−1). Conventionally, a material contracts in transverse directions when it is stretched in the longitudinal direction, such that the Poisson's ratio is positive. However, some exotic, so-called auxetic [2], materials have a negative Poisson's ratio. Although examples of such materials, e.g., pyrite, have been known for a long time [3], the interest to auxeticity started only at the end of 1980s after the observation of a stretching-induced transverse expansion of polyurethane foam [4]. Nowadays, a negative Poisson's ratio is found in various materials and artificially engineered structures (see Ref. [5] for a review). ∗Corresponding author. Fax: +7-495-7029317 Email address: burmi@itp.ac.ru (I.S. Burmistrov) Preprint submitted to Annals of Physics January 17, 2018 8 1 0 2 n a J 5 1 ] l l a h - s e m . t a m - d n o c [ 1 v 3 5 0 5 0 . 1 0 8 1 : v i X r a An interesting example of auxetic material is a crystalline membrane of dimension D em- bedded into the space of dimension d > D. The self-consistent theory of such crystalline mem- branes [6] predicts the negative Poisson's ratio in the thermodynamic limit. This limit is achieved in large membranes, when the membrane size L exceeds the Ginzburg length L∗ ∼ κ/ T µ, where κ is the bending rigidity and T stands for the temperature. A crystalline membrane hosts dc = d − D soft out-of-plane modes, the so-called flexural phonons, which are characterized by strong anharmonicity mediated by the coupling to conventional in-plane phonons [7]. As a consequence of such anharmonicity, the elastic moduli show a nontrivial power-law scaling with the system size, temperature, and tension. The scaling of all elastic moduli, λ, µ, κ is controlled by the universal exponent η which depends only on dc. The critical exponent η was determined within several approximate analytical schemes[6, 8–11], none of which being controllable in the physical case D = 2 and d = 3. Numerical simulations for the latter case yielded η = 0.60± 0.10 [12], η = 0.72 ± 0.04 [13], and η = 0.85 [14]. It is because of the nontrivial scaling of the elastic moduli that the linear Hooke's law fails in the regime of small tension [9, 15–20]. √ Le Doussal and Radzihovsky [6] found a negative Poisson's ratio of a two-dimensional crys- talline membrane within the self-consistent screening approximation. More specifically, they obtained an entirely universal value ν = −1/3 independent of the spatial dimensionality dc. In Ref. [18], this result of the self-consistent membrane theory was reproduced by Kosmrlj and Nelson by means of a renormalization-group analysis for a relatively large membrane size L (cid:29) L∗ and not too strong tension, σ (cid:28) σ∗ = κL−2∗ . On the other hand, as shown by the present authors together with Katsnelson and Los in a parallel paper [22], the Poisson's ratio is strongly dependent of boundary conditions in the range of lowest tensions (linear-response −η∗ . An independence on boundary conditions is reached only at regime), σ (cid:38) σL = κLη−2L stronger tensions, σ (cid:29) σL. However, also in this regime, one should exert a care when defin- ing the Poisson's ratio. Specifically, emergence of the anomalous, non-linear Hooke's law re- sults in an essential difference between the absolute and differential Poisson's ratio, as shown in Ref. [22]. In this paper, we consider the non-linear regime σL (cid:28) σ (cid:28) σ∗ and focus on the differential Poisson's ratio. In order to define the differential Poisson's ratio ν, one needs to consider the response to an infinitesimally small anisotropic tension: σ(cid:107) = σ + δσ and σ⊥ = σ. Then, the ratio of the infinitesimally small change in transverse, δε⊥, and longitudinal, δε(cid:107), stretching determines the differential Poisson's ratio ν = − δε⊥ δε(cid:107) . (1) We demonstrate that in the regime σL (cid:28) σ (cid:28) σ∗ the differential Poisson's ratio indeed acquires a universal value. However, contrary to the result of the self-consistent membrane theory, this universal value depends on the dimensionality dc of embedded space. We perform calculations which are controlled by the small parameter 1/dc (cid:28) 1 and find that the differential Poisson's ratio of the two-dimensional crystalline membrane is given by ν = −1 3 0.016 + (2) Thus, the differential Poisson's ratio at σL (cid:28) σ (cid:28) σ∗ is universal (in the sense of independence on material parameters) but represents a nontrivial function of dc. The paper is organized as follows. In Sec. 2 we present the general formalism for the compu- tation of the differential Poisson's ratio of a two-dimensional crystalline membrane. The details dc σL (cid:28) σ (cid:28) σ∗. + O(cid:0)d−2 (cid:1) , c 2 of evaluation of the differential Poisson's ratio to the first order in 1/dc are presented in Sec. 3. We end the paper with a summary of results, Sec. 4. Technical details are given in Appendices. 2. Formalism We start with the partition function of a two-dimensional crystalline membrane written in terms of the functional integral over in-plane, u = {ux, uy}, and out-of-plane, h = {h1, . . . , hdc} phonons (see Refs. [20, 21, 23]): Here the action in the imaginary time is given by (β = 1/T) (cid:90) (cid:40)(cid:104) µ dτ d2x β(cid:90) 0 (cid:90) Z = D[u, h] exp(−S ). (cid:17)(cid:16) β − 1 + Kβ ξ2 + µuαβuβα + λ 2 4 δαβ + κ 2 + λ 8 α − 1 + Kα ξ2 (cid:105)(cid:104)(cid:16) (∆h)2 + (∆u)2(cid:105) (cid:104) (cid:16) S = where and uαβ = Kα = 1 βL2 1 2 β(cid:90) 0 (cid:90) ξβ∂αuβ + ξα∂βuα + ∂αu∂βu + ∂αh∂βh , dτ d2x Kα, Kα = ∂αu∂αu + ∂αh∂αh. (cid:105) + ρ 2 (cid:104) (∂τu)2 + (∂τh)2(cid:105) (cid:17) − KαKβ (cid:41) uααuββ , (cid:17) (3) (4) (5) (6) The free energy per unit area, f = −T L−2 ln Z, is a function of the stretching factors ξx and ξy, i.e. f ≡ f (ξx, ξy). With the function f (ξx, ξy), the diagonal components of the tension tensor can be found as (7) (7) determines the dependence of the tension tensor {σx, σy} on the ∂ f 1 ξy ∂ξy We emphasize that Eq. stretching tensor {ξx, ξy}, i.e., Eq. (7) is the equation of state. In order to find the differential Poisson's ratio ν, we consider the case of slightly anisotropic stretching factors, ξα = ξ + δεα, and adjust the ratio ν = −δεy/δεx in such a way that the components of the tension tensor, σx = σ + δσ and σy = σ, differ only by an infinitesimal addition δσ in σx. Then, we find ∂ f ∂ξx σx = σy = 1 ξx , . (cid:32) ∂σy (cid:32) ∂σy ∂ξx (cid:33) (cid:33) ν = ξy = ∂2 f ∂ξy∂ξx ∂2 f − σ ∂ξ2 y . (8) Here the derivatives are taken at ξx = ξy = ξ. ∂ξy ξx 3 We note that instead of independent variables ξx and ξy, one can choose as independent variables the components of the tension tensor, σx and σy. Equation (8) can be then rewritten in an alternative form: (cid:33) (cid:33) (cid:32) ∂ξy (cid:32) ∂σx ∂ξx ∂σx ν = − ∂2g = − , ∂σx∂σy ∂2g ∂σ2 x (9) σy σy where the derivatives are assumed to be calculated for σx = σy = σ. As usual, the free energy g(σx, σy) is related to the free energy f (ξx, ξy) via the Legendre transform: y − 1)/2, g(σx, σy) = f (ξx, ξy) − σx(ξ2 x − 1)/2 − σy(ξ2 (10) where ξα is expressed in terms of σα with the help of the equation of state (7). We note that the expression (9) has been used for the numerical evaluation of the Poisson's ratio in Ref. [24] (though with the different form of the free energy). Although, both formulations (8) and (9) are completely equivalent, in what follows we will use the formulation in which the stretching factors ξα are the independent variables. Using the exact form (4) of the action, one finds the following expressions for the second derivatives of the partition function f : (cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)ξx=ξy=ξ ∂2 f ∂ξ2 y (cid:90) (cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)ξx=ξy=ξ = σ + ξ2(2µ + λ) + (2µ + λ)(cid:10)(∂yuy)2(cid:11) + µ(cid:10)(∂xuy)2(cid:11) − 2µ + λ (cid:48)dx(cid:48)(cid:104)(cid:104)Ly(x, τ) · Lx(x(cid:48) ∂2 f ∂ξy∂ξx (cid:48))(cid:105)(cid:105), , τ dτ = ξ2λ − ξ2 (cid:90) −2µ ξ (cid:104)uxy∂xuy(cid:105) − ξ2 (cid:48)dx(cid:48)(cid:104)(cid:104)Ly(x, τ) · Ly(x(cid:48) dτ (cid:48))(cid:105)(cid:105). , τ (cid:104)uyy∂yuy(cid:105) (11) (12) ξ Here the average (cid:104). . .(cid:105) is defined with respect to the action (4), (cid:104)(cid:104)A · B(cid:105)(cid:105) = (cid:104)AB(cid:105) − (cid:104)A(cid:105)(cid:104)B(cid:105) and Lx = Ly = 2µ + λ 2 2µ + λ 2 Kx + Ky + λ 2 λ 2 Ky + Kx + 2µ + λ 2ξ 2µ + λ 2ξ uxx∂xux + uyy∂yuy + 2µ ξ 2µ ξ uyx∂yux + uxy∂xuy + λ ξ λ ξ uyy∂xux, uxx∂yuy. (13) (14) Equations (8), (11), and (12) express the Poisson's ratio in terms of correlation functions of elastic deformations. The actual computation of these correlation functions of the in-plane and flexural phonons is complicated due to interaction between these phonon modes. Below we limit the analysis to the case of high temperature, T (cid:29) κ2/(µL2) in which one can consider the phonons to be quasistatic. In this regime, one can also neglect the term ∂αu∂βu in comparison with ∂αh∂βh in the expressions for uαβ and Kα. Then we can simplify Eqs. (11) and (12). Indeed, by making the following change of variables: uα → ξαuα, we can recast the partition function (3) as: (cid:90) Z = D[u, h] exp(− E/T), (15) 4 (16) Kα = ∂αh∂αh. (17) (cid:17) (cid:90) where (cid:90) (cid:40)(cid:34) d2x E = (cid:35)(cid:34)(cid:18) α − 1 + Kα ξ2 µ 4 δαβ + λ 8 (cid:19)(cid:18) (cid:35) (cid:19) − Kα Kβ + κ 2 (∆h)2 +µuαβ uβα + λ 2 uαα uββ (cid:41) β − 1 + Kβ ξ2 (cid:90) . Here we have introduced the following notations: ∂αuβ + ∂βuα + ∂αh∂βh , Kα = 1 L2 d2x Kα, (cid:16) uαβ = 1 2 Since the action (16) becomes quadratic in the in-plane phonons, we can integrate them out and express the partition function as an integral over static flexural phonons, Z = D[h] exp(−E/T), (18) where the energy E for a given configuration of the flexural phonon field h(x) is given by [20] E = (cid:34) (cid:90) d2x (cid:90) + κ 2 (cid:35)(cid:34)(cid:18) µ 4 δαβ + λ 8 d2x (∆h)2 + α − 1 + Kα ξ2 2µ(µ + λ) 4(2µ + λ) (cid:19)(cid:35) (cid:19)(cid:18) (cid:90) (cid:48) d2kd2k(cid:48)d2q β − 1 + Kβ ξ2 (cid:32)(cid:90) (cid:33)2 (cid:0)hk+qh−k + µ 2L2 [k × q]2 d2x ∂xh∂yh [k(cid:48) × q]2 (2π)6 q2 q2 (cid:1)(cid:0)h−k(cid:48)−qhk(cid:48)(cid:1). (19) The 'prime' sign in the last integral means that the interaction with q = 0 is excluded: the 'zero- mode' term with q = 0 from the contributions uαβ uβα and uαα uββ to the energy E in Eq. (16) has been combined with the term Kα Kβ, yielding exactly the term with ∂xh∂yh in Eq. (19). Since now ξα does not enter the interaction part of the free energy which depends on u, we obtain a much simpler equation of state: (cid:32) (cid:33) σx σy (cid:33) 1 2 x − 1 + (cid:104) Kx(cid:105) ξ2 y − 1 + (cid:104) Ky(cid:105) ξ2 (cid:32)2µ + λ (cid:33) λ = M (20) Here the average (cid:104). . .(cid:105) is with respect to the energy (19). The second derivatives of the free energy with respect to the stretching factors become 2µ + λ M = λ , . (cid:32) (cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)ξx=ξy=ξ (cid:90) = ξ2λ − ξ2β ∂2 f ∂ξy∂ξx (cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)ξx=ξy=ξ ∂2 f ∂ξ2 y dx(cid:48)(cid:104)(cid:104) Ly(x) · Lx(x(cid:48))(cid:105)(cid:105), (cid:90) dx(cid:48)(cid:104)(cid:104) Ly(x) · Ly(x(cid:48))(cid:105)(cid:105), = σ + ξ2(2µ + λ) − ξ2β (21) (22) where Lα = Mαβ Kβ/2. The energy functional E involves two types of interaction of flexural phonons. The terms in the first line of Eq. (19) correspond to the interaction with zero momentum transfer ('zero 5 In the case of large membrane size, σ (cid:29) σL, this interaction can be treated in the mode'). random phase approximation. Then, we find (cid:32) Π (cid:33)−1 d2x(cid:48)(cid:104)(cid:104) Kα(x) · Kβ(x(cid:48))(cid:105)(cid:105) = 1 + 1 2 MΠ , αβ (23) (cid:90) β 2 where Π denotes the polarization operator (at zero momentum) irreducible with respect to the interaction with the zero-momentum transfer: (cid:90) Παβ = β 2 d2x(cid:48)(cid:104)(cid:104) Kα(x) · Kβ(x(cid:48))(cid:105)(cid:105)irr. (24) We note that Παβ has two independent components: Πxx = Πyy and Πxy = Πyx. Using Eqs. (21) and (22), we express the differential Poisson's ratio in terms of the components of Π: Here ν0 − Y0Πxy/2 1 + Y0Πxx/2 . ν = ν0 = λ 2µ + λ , Y0 = 4µ(µ + λ) 2µ + λ (25) (26) denote the bare values of the Poisson's ratio and Young modulus for the two-dimensional crys- talline membrane, respectively. In order to clarify the meaning of Πxx and Πxy, it is useful to consider a general form of the polarization operator at finite momentum q: (cid:90) d2x(cid:48) e−iq(x−x(cid:48))(cid:104)(cid:104)(cid:0)∇αh(x)∇βh(x)(cid:1) ·(cid:0)∇γh(x(cid:48))∇δh(x(cid:48))(cid:1)(cid:105)(cid:105). Παβ,γδ(q) = 1 2 Due to the rotation symmetry and the symmetry under permutation of the indices α and β (as well as γ and δ), the polarization operator at zero momentum is expressed as follows [16] Παβ,γδ(0) = Πxyδαβδγδ + 1 2 (cid:0)Πxx − Πxy (cid:1)(cid:0)δαγδβδ + δαδδβγ (cid:1). We emphasize that, in general, there are no reasons for Παβ,γδ to be fully symmetric with respect to permutations of all its indices as it is assumed in the self-consistent screening approximation [6, 25]. Therefore, Eq. (25) yields the most general expression for the differential Poisson's ratio. We also note that Eq. (25) can be written as (see Appendix A) where λ(cid:48) and µ(cid:48) are the screened Lam´e coefficients: λ(cid:48) 2µ(cid:48) + λ(cid:48) , ν = (cid:48) = µ µ 1 + (Πxx − Πxy)µ , B(cid:48) = B 1 + (Πxx + Πxy)B . (29) (30) Here we have introduced bare and screened bulk moduli: B = µ +λ and B(cid:48) = µ(cid:48) +λ(cid:48), respectively. In order to find how ν depends on parameters of the problem, e.g., on the number of flexural phonon modes dc, one needs to compute Πxx and Πxy. In the next section we remind the reader 6 (27) (28) on the results of the self-consistent screening approximation and then compute corrections in 1/dc. Using the equation of state (20), we can express the stretching factors ξα via tensions σα. Then, with the help of Eq. (9), we find the following representation for the differential Poisson's ratio σy . (31) (cid:32) ∂(cid:104) Ky(cid:105) (cid:33) (cid:32) (cid:33) ∂σx ∂(cid:104) Kx(cid:105) ∂σx σy ν = ν0 + Y0 2 1 − Y0 2 (cid:90) d2k αG2 k (2π)2 k2 Here (cid:104) Kα(cid:105) is expressed in terms of σx and σy. After taking derivatives in Eq. (31), one sets σx = σy = σ. Below we demonstrate how the two representations of the differential Possion ratio, (25) and (31), are related. The irreducible polarization operator at the zero momentum can be exactly expressed via the full triangular vertex Γβ(k, k): Παβ = Γβ(k, k). (32) Here Gk = T κk4 + (ξ2 α − 1)Mαβk2 β/2 − Σk denotes the propagator for the flexural phonons, with Σk being the exact self-energy. The bare value of the triangular vertex Γβ(k, k) is equal to k2 β/T. The full triangle vertex satisfies the following identity: ∂G−1 k ∂σβ Γβ(k, k) = (33) . As a consequence of this identity, we obtain (cid:90) d2k Παβ = − ∂ ∂σβ αGk = − ∂(cid:104) Kα(cid:105) ∂σβ (2π)2 k2 . (34) Therefore, expressions (25) and (31) are identical. 3. Evaluation of the differential Poisson's ratio 3.1. Self-consistent screening approximation ization of the bending rigidity at k (cid:28) q∗ ≡ L−1∗ [6, 10, 16], κ → κ(k) = κ(q∗/k)η f (k/qσ). The interaction between flexural phonons with finite momentum transfer results in renormal- Here, qσ = q∗(σ/σ∗)1/(2−η) and the function f (x) has the following asymptotic behavior: 1, x−η, 7 f (x) = x (cid:29) 1, x (cid:28) 1. (35) (36) (cid:90) The simplest approach for computing the irreducible polarization operator is to neglect the vertex corrections. As we shall see below, this can be justified for dc (cid:29) 1. Then, we find (37) Independently of the form of the exact propagator Gk, we find the irreducible polarization oper- ator as = dc Π(0) αβ d2k (2π)2T β G2 αk2 k2 k. Π(0) αβ = dcγ(cid:104)n2 β(cid:105)n, αn2 (38) Here n stands for the two-dimensional unit vector and (cid:104). . .(cid:105)n denotes the averaging over direc- tions of n. Thus, neglecting the vertex corrections yields the following relation: γ = d2k (2π)2T k4 G2 k. (cid:90) Π(0) xx = 3Π(0) xy . (39) Relation (39) implies that Παβ,γδ is fully symmetric with respect to permutation of indices. This assumption is used in the self-consistent screening approximation. Motivated by the renormalization of bending rigidity (35) and the Ward identity (see Ap- pendix B), we use the following ansatz for the exact propagator: The integral over k in Eq. (37) is then dominated by k ∼ qσ and we obtain Then, from Eq. (25) we find at σ (cid:28) σ∗ that the differential Poisson's ratio becomes Gk = T κ(k)k4 + σk2 . (cid:19)η/(2−η) . (cid:18) σ∗ σ γ ∼ T κσ ν ≈ − Π(0) xy Π(0) xx = −1 3 . (40) (41) (42) It is exactly the result that was obtained within the self-consistent screening approximation [6]. 3.2. Vertex corrections to the polarization operator Corrections to the result (42) stem from the violation of the relation (39). In order to refine the differential Poisson's ratio, we expand the right-hand side of Eq. (25) in the difference 3Πxy−Πxx: ν ≈ −1 3 + 3Πxy − Πxx 9Π(0) xy (43) As we shall see below, the correction to the value −1/3 will be of the order of 1/dc. There are three diagrams with non-trivial vertex corrections (see Fig. 1) that contribute to Παβ at order d0 c. They yield the following corrections: [k × q]4 q4 α(kβ − qβ)2, N(cid:48) qk2 (44) = −2dc Π(a) αβ (cid:90) d2kd2q kG2 (2π)4T 2G2 k−q 8 and Π(b+c) αβ = 4d2 c (cid:90) d2kd2k(cid:48)d2q (2π)6T 3 G2 kGk−qG2 k(cid:48)Gk(cid:48)−q [k × q]4 [k(cid:48) × q]4 q4 q4 N(cid:48)2 q k2 αk(cid:48)2 β . Here N(cid:48) q denotes the screened interaction between flexural phonons (see Appendix A), N(cid:48) q = Y0/2 1 + 3Y0Π(0) q /2 , (45) (46) where Π(0) q correction. We note that Π(0) following expression: stands for the polarization operator at finite momentum calculated without vertex is given by q=0. The polarization operator Π(0) q xx = 3Π(0) xy = 3Π(0) (cid:90) Π(0) q = dc 3 d2k (2π)2T [k × q]4 q4 GkGk−q. 3Πxy − Πxx Since we are interested in the regime q (cid:28) q∗, we can approximate N(cid:48) (cid:90) (cid:90) d2kd2k(cid:48) combining both contributions together, we find kG2 (2π)4T 2 G2 k(cid:48) [k(cid:48) × q]4 ×[k × q]4 = − 2dc 27Π(0) xy (cid:40)[k × k(cid:48)]4  1 2(cid:41)(cid:104) 1 Π(0) k−k(cid:48) y − k2 xk(cid:48)2 3k2 k − k(cid:48)4 − 2dc 3 9Π(0) xy xk(cid:48)2 x (cid:105) . q4 q4 Π(0) q q by 1/[3Π(0) q ]. Then, d2q (2π)2T Gk−qGk(cid:48)−q (48) We note that this expression can be written in a rotationaly invariant way. Indeed, in the first term, the expression under the integral sign depends on the angle θ between k and k(cid:48) only. Averaging over directions of k, we find (47) (cid:90) 2π dφ 2π (cid:104) (cid:105) 3 cos2(φ + θ) − sin2(φ + θ) (49) In the second term, the expression under the integral sign depends on the angles θ and θ(cid:48) between k and q, and between k(cid:48) and q, respectively. Averaging over directions of q, we find 0 = sin2 θ. cos2 φ (cid:105) (cid:90) 2π dφ 2π (cid:104) cos2(φ + θ) 3 cos2(φ + θ (cid:48)) − sin2(φ + θ (cid:48)) = sin2(θ − θ (cid:48)). 0 Therefore, we obtain where 3Πxy − Πxx 9Π(0) xy = I(a) + I(b+c), (cid:90) d2kd2k(cid:48) (cid:90) d2kd2k(cid:48)d2q (2π)4T 2 (2π)6T 3 , kG2 G2 k(cid:48) Π(0) k−k(cid:48) [k × k(cid:48)]6 k − k(cid:48)4 [k × k(cid:48)]2 [k × q]4 q4Π(0) q I(a) = − 2dc 27Π(0) xy 4d2 c 81Π(0) xy I(b+c) = (50) (51) (52) [k(cid:48) × q]4 q4Π(0) q Gk−qGk(cid:48)−qG2 kG2 k(cid:48) . 9 (a) (b) (c) Figure 1: The diagrams of the first order in 1/dc for the polarization operators Πxx and Πxy at zero momentum transfer. The solid line denotes the propagator Gk. The wavy line depicts the screened interaction between flexural phonons, which is equal to 1/[3Π(0) q ] in the universal regime, q < q∗. 3.3. Correction to the self-energy The results (51) and (52) can be derived in a different way using the relation (33) between the triangular vertex at zero momentum and the inverse Green's function. In view of Eq. (34), in order to find the differential Poisson's ratio one needs to compute the change of the Green's function upon applying an infinitesimally small tension δσ along the x direction. In the presence of δσ, the Green's function can be written in terms of the self-energy Σk: Gk = T (53) We mention that the ansatz (40) used above for δσ = 0 corresponds to Σk = [κ − κ(k)]k4. We also note that the trivial term δσk2 x is included into Σk for the sake of convenience. κk4 + σk2 − Σk In order to find the change of Σk induced by the infinitesimally small tension δσ, we use the . lowest-order diagram for the self-energy (see Fig. 2): (cid:90) d2q Σk = 2 3 [k × q]4 Gk−q Π(0) q , (54) (2π)2 q4 We note that, as above, the dominant contribution comes from momenta q (cid:28) q∗ such that the interaction line is determined by the inverse polarization operator. As one can see from the diagram in Fig. 2, the variation of the self-energy in the presence of δσ arises from the variation of the Green's function: δGk−q = G2 k−qδΣk−q, as well as from the the change of the polarization operator (see Eq. (47)) [k × q]4 [k × q]4 (cid:90) d2k δΠ(0) q = 2dc 3 (2π)2 q4 (2π)2 (cid:90) d2k  δGk−q Π(0) q q4 G2 (cid:3)2 (cid:2)Π(0) − Gk−qδΠ(0)  . q q δGkGk−q = 2dc 3 (cid:90) d2q [k × q]4 (2π)2 q4 Now the correction δΣk can be found from the variation of Eq. (54): δΣk = −δσk2 x + 2 3 Since the right-hand side of this equation is linear in δΣk, it can be rewritten as (1 + α)δΣ = −δσk2 x, 10 kGk−qδΣk. (55) (56) (57) (58) Figure 2: The diagram for the self energy (see text). where we formally introduce the linear integral operator α as: α δΣk = −2 3 (k × k(cid:48))4 k − k(cid:48)4Π(0) k−k(cid:48) − 2dc 3 (cid:90) d2k(cid:48) (2π)2G2 (cid:90) d2q (2π)2 k(cid:48)  [k × q]4 q4Π(0) q [k(cid:48) × q]4 q4Π(0) q  δΣk(cid:48) . Gk(cid:48)−qGk−q (59) It is worthwhile to mention that the linear operator α conserves the angular momentum, as fol- lows from the rotational invariance of Eq. (59). Therefore, it is convenient to split α into the zeroth and second harmonics: x − k2 k2 2k2 The formal solution of Eq. (58) can be then written as α+k2 + αk2 x = 1 2 y α−k2.  δΣk = − δσ 2 1 1 + α+ x − k2 k2 k2 y + 1 1 + α−  k2. (60) (61) Although Eq. (61) yields a formal solution for δΣk, it is not justified to keep α± beyond the lowest order: not all the terms of the order 1/d2 c can be generated from the diagram in Fig. 2. After a straightforward calculation, we obtain ν ≈ −1 3 (cid:10) α+ − α−(cid:11) (62) k, + where (cid:82) 4 9 (cid:82) d2k k2G2 k α±k2 (cid:104) α±(cid:105)k = (63) Expressing the difference (cid:104) α+ − α−(cid:105)k in the rotationally invariant way, we obtain from Eq. (62) exactly the same expression as in Eqs. (51) and (52). d2k k4G2 k . 3.4. Evaluation of the vertex corrections As we shall see below, all the integrals determining the 1/dc correction to the differential Poisson's ratio are dominated by the momenta of the order of qσ. Since the dependence of the bending rigidity on q is controlled by η (cid:39) 2/dc, we can neglect this dependence in the calculation of the correction (51). Therefore, in what follows, we approximate the propagator of the flexural phonons by Eq. (40) with the bare bending rigidity. Then, we find 16πκ2q2P where the dimensionless function P(Q) is given as q = dcT Π(0) (cid:90) d2K P(Q) = 8 3 Q2 [K × Q]4 1 (2π)2 Q4 K2(K2 + 1) 11 1 Q − K2(Q − K2 + 1) . (64) (65) (cid:33) (cid:32) q √κ√ , σ The function P(Q) can be evaluated exactly with the help of the following set of transformations: P(Q) = 8 3 Q2 dt1dt2 ∞(cid:90) (cid:104) 1 − e−t1(cid:105)(cid:104) ×(cid:104) 1 − e−t1(cid:105)(cid:104) 1 − e−t2(cid:105) (cid:40)(cid:32) ∞(cid:90) −Q2 t1t2 e (cid:33) 0 4 cosh2 z dz 1 + Q2 1 − e−t2(cid:105)(cid:90) d2K ∞(cid:90) (2π)2 dz t1 +t2 = Q2 4 (cid:32) ln 1 + cosh4 z −∞ 4 cosh2 z (cid:33) Q2 = Q4 8 cosh4 z −∞ (cid:40) dt1dt2 (t1 + t2)3 Q4 [K × Q]4 (cid:90) ∞ (cid:32) dτ 0 − 2 1 + 0 e−t1K2−t2K−Q2 = Q2 ∞(cid:90) τ2 e−Q2τ/2(cid:89) (cid:104) 1 − e−τeσz cosh z(cid:105) (cid:32) (cid:33) (cid:112) 2ez cosh z σ=± 1 + Q2 Q2 ln (cid:41) 2ez cosh z (cid:33)(cid:41) = 1 3 1 + Q4 ln Q − (1 + Q2)3 ln(1 + Q2) (66) Here we used the parameterization t1,2 = τe±z cosh z. We note that the function P(Q) has the following asymptotic behavior: + Q(4 + Q2)3/2 ln 4 + Q2 + Q Q2 2 . P(Q) = Q2 − Q4 (1 − 2 ln Q), 2 6 1 − 1 2Q2 (1 + 4 ln Q), Q (cid:28) 1 Q (cid:29) 1. (67)  Now we compute the contribution I(a) in Eq. (51) from the diagram in Fig. 1a. This contri- In particular, we find that Π(0) xy = dcT/(32πκσ). q4 1 Π(0) q [k × q]6 bution can be written as I(a) = − 2dc 27Π(0) xy (cid:90) d2kd2q (2π)4T 2G2 kG2 k−q t j − 2 + (2 + t j)e−t j(cid:105)(cid:90) d2K (cid:89) where the function Y1(Q) is given by ∞(cid:90) (cid:104) t j − 2 + (2 + t j)e−t j(cid:105) e (cid:89) ∞(cid:90) (cid:104) = −(32π)2 27dc Y1(Q) = dt1dt2 (t1 + t2)4 15 32π dt1dt2 (2π)2 j=1,2 = 0 0 j=1,2 (cid:90) d2Q (2π)2 [K × Q]6 Q4 −Q2 t1t2 t1 +t2 . Q2 P(Q) Y1(Q), (68) e−t1K2−t2K−Q2 (69) Y1(Q) = Performing the transformation t1,2 = τe±z cosh z and integrating over τ, we find dz (cid:40)(cid:16) ∞(cid:90) Q4(5 + 10Q2 + 2Q4) ln Q + (1 + Q2)2(cid:104) (1 + 2Q2) cosh2 z + cosh6 z Q4 2 −∞ Q4 + 4(1 + Q2) cosh2 z (cid:17) 2 +(cid:0)2 − 9Q2 + 6Q4 + 2Q6(cid:1)ln(1 + Q2) (cid:41) − 2 cosh2 z (cid:105) Q4 + 4Q2 cosh2 z ln 15Q2 256π (cid:40) = − 1 32πQ2 (cid:112) (cid:41) Q2 . (70) (cid:112) +Q 4 + Q2(−10 − 3Q2 + 6Q4 + 2Q6) ln 12 4 + Q2 + Q 2 I(b+c) = (32π)3 81dc The contribution I(b+c) in Eq. (51) from the diagrams in Fig. 1b and Fig. 1c can be computed in a similar way. We rewrite I(b+c) as follows: (cid:104) (cid:89) j=1,2 ∞(cid:90) (cid:90) d2K1d2K2 1dt2dt(cid:48) dt1dt(cid:48) 0 2 × (2π)4 [K1 × K2]2 t j − 2 + (2 + t j)e−t j(cid:105) (cid:89) [K j × Q]4 Q4 j=1,2 [1 − e−t(cid:48) j]e −Q2 t jt(cid:48) t j +t(cid:48) j j −(t j+t(cid:48) j)(K j−Q e t(cid:48) j t j +t(cid:48) j )2 (cid:90) d2Q  . (2π)2 Q4 P2(Q) Then, integrating over K1 and K2, we get (cid:90) d2Q (cid:104) (2π)2 (cid:104)Y2(Q) + Y3(Q) (cid:105) (cid:105) 1 − e−t(cid:48) e 1 , Q4 P2(Q) Y2(Q) t1 − 2 + (2 + t1)e−t1(cid:105)(cid:104) −Q2 t1t(cid:48) 1 t1 +t(cid:48) 1 , I(b+c) = 2(32π)3 81dc ∞(cid:90) 0 15 32π dt1dt(cid:48) 1 (t1 + t(cid:48) 1)4 dt1dt(cid:48) 1 (t1 + t(cid:48) 1)4 1 t1 − 2 + (2 + t1)e−t1(cid:105)(cid:104) (cid:104) (cid:40) − cosh2 z − Q2 (cid:17)(cid:104) 4 (cid:90) ∞ (cid:16) Q2 (cid:0)6 + 7Q2 + 6Q4(cid:1) − 4 + cosh2 z cosh6 z dz (cid:40) (cid:105)(cid:104)−2 + Q2 1 − e−t(cid:48) 1 (cid:105) −Q2 t1t(cid:48) 1 t1 +t(cid:48) e 1 . t(cid:48)2 t1 + t(cid:48) 1 1 (cid:32) (cid:33) 1 + e−2z 1 + (Q2 + 1 + e−2z) ln Q2 ln(Q2 + 4 cosh2 z) − ln(Q2 + 1 + e−2z) (cid:105)(cid:41) (cid:0)1 + Q2(cid:1)3(cid:0)2Q4 + 4Q2 − 3(cid:1) ln(cid:0)1 + Q2(cid:1) . where and Y3(Q) = Y2(Q) = ∞(cid:90) 3 16π 0 Y2(Q) = 15 256π −∞ +(Q2 + 1 + e−2z) 4 Integrating over z, we arrive at Using the parametrization t1 = τez cosh z and t(cid:48) = τe−z cosh z, and integrating over τ, we obtain (71) (72) (73) (74) (75) Y2(Q) = 1 − 2 Q4 256π +4Q2(cid:0)2Q2 + 5(cid:1) ln Q + Q6 (cid:0)2Q2 + 3(cid:1)(cid:0)4 + Q2(cid:1)3/2 ln 4 Q (cid:112) The function Y3(Q) can be conveniently expressed as (cid:41) 4 + Q2 + Q 2 . (76) Y3(Q) = −4 5 Y2(Q) + Y3(Q), (77) where the function Y3(Q) after the integration over τ acquires the following form: Y3(Q) = (2Q2 + 1 + e2z) ln 1 + e2z ∞(cid:90) e−2z 1 + dz (cid:40) (cid:33) (cid:32) − (2Q2 + 1 + e−2z + 4 cosh2 z) 3Q2 512π cosh6 z −∞ ×(cid:104) Q2 (cid:105)(cid:41) 13 ln(Q2 + 4 cosh2 z) − ln(Q2 + 1 + e−2z) . (78) (cid:40) 2Q4 (cid:40) Integration over z yields Y3(Q) = 1 160π −Q2(5 + 6Q2) ln Q + (cid:112) (1 + Q2)2(6Q6 + 8Q4 + 8Q2 − 9) Q6 (cid:112) ln(1 + Q2) (cid:41) 18 + 11Q2 + 18Q4 + − 4 + Q2 Q (26 + 23Q2 + 6Q4) ln 4 + Q2 + Q 2 . (79) Then, we obtain the following expression Y3(Q) = − Q−4 128π ×(1 + Q2)2 Q2 −3(2 + Q2 + 2Q4) + 2Q6(1 + 2Q2) ln Q − 2(−3 + 3Q2 + 2Q4 + 2Q6) (cid:112) (cid:112) (cid:41) 4 + Q2 + Q 2 . (80) ln(1 + Q2) + 2Q3(8 + 7Q2 + 2Q4) 4 + Q2 ln 3.5. Final result for the differential Poisson's ratio Combining together the results for the contributions I(a) and I(b+c), we express the difference of the polarization operators responsible to the 1/dc correction to ν through a single integral: ∞(cid:90) 3Πxy − Πxx 9Π(0) xy = 16 81dc dQ H(Q) P2(Q) , where H(Q) = 96πQ3(cid:110)−Y1(Q)P(Q) + (cid:40) 0 (cid:111) Q2Y2(Q)[Y2(Q) + Y3(Q)] . 64π 3 Using Eqs. (66), (70), (76), and (80), we obtain the following lengthy explicit expression for H(Q): H(Q) = − 1 8Q3 ln2(1 + Q2) (cid:104) −4Q12(5 + 8Q2) ln2 Q + 4(1 + Q2)6(9 − 30Q2 + 26Q4) + Q2 (cid:112) 4 + Q2 + Q 4 + Q2 + Q 4 + Q2(cid:0)60 + 96Q2 + 143Q4 2 Q4 18 + 3Q2 − 26Q4 − 34Q6 − 58Q8 − 20Q10 − 4(1 + Q2)2 ln(1 + Q2) + 2Q3(cid:112) (cid:105) 4 + Q2(cid:0)−30 − 3Q2 + 97Q4 + 38Q6 + 4Q8(cid:1) ln 36 + 60Q2 + 77Q4 + 28Q6 + 20Q8 − 4Q3(cid:112) (cid:16) (cid:112) (cid:112) + 100Q6 + 20Q8(cid:1) ln + 4Q4(cid:16) 2(1 + Q2)2(9 + 5Q2 − 6Q4 + 4Q6) ln(1 + Q2) − Q2(cid:0)18 + 37Q2 (cid:41) + 56Q4 + 20Q6 − 2Q3(cid:112) 5Q5/8, (cid:32)485 + 4Q6(4 + Q2)3(11 + 8Q2) ln2 Q (cid:28) 1, Q (cid:29) 1. 4 + Q2(16 + Q2) ln ln Q + 10 ln2 Q /Q3, H(Q) = 4 + Q2 + Q ln Q . (cid:1)(cid:17) − 65 3 72 (cid:112) 2 2 2 The function H(Q) has the following asymptotic behavior: (cid:33) 14 4 + Q2 + Q (81) (82) (cid:17) (83) (84) Figure 3: The plot of the function H(Q)/P2(Q) (see text). The function H(Q)/P2(Q) is shown in Fig. 3. As one can see, it changes sign twice which leads to a partial compensation of the corrections from diagrams on Fig. 1a-c. Numerically evaluating the integral in Eq. (81) and substituting it into Eq. (43), we find the result (2). 4. Conclusions To summarize, we have computed the differential Poisson's ratio of a suspended two- dimensional crystalline membrane embedded into a space of large dimensionality d (cid:29) 1. Our result (2) demonstrates that, for σL (cid:28) σ (cid:28) σ∗, the differential Poisson's ratio of a crystalline membrane is a universal but non-trivial function of dc. This results invalidates a common belief (based on results of the self-consistent screening approximation) that the Poisson's ratio is equal to −1/3 independently of dc. In the physical case of a two-dimensional membrane (dc = 1), one may speculate that the differential Poisson's ratio is not too far from the value −1/3 since the correction of the order 1/dc in Eq. (2) is numerically small. Clearly, a comparison with computational results would be of great interest. Unfortunately, the existing numerical results the Poisson's ratio of two- dimensional membranes (including graphene) are, however, quite controversial. This may be partly related with a very delicate character of the problem, see a detailed analysis in Ref. [22]. As has been mentioned in Sec. 1, the Poisson ratio in the linear-response regime σ (cid:28) σL de- pends on boundary conditions. In order to get rid of such finite-size effects but still to be in the regime of universal elasticity, the stress should be in the intermediate range σL (cid:28) σ (cid:28) σ∗. To resolve well this regime in numerical simulations, sufficiently large systems should be consid- ered. Furthermore, in this regime, a care should be exerted in order to distinguish between the differential and the absolute Poisson ratio [22]. Finally, we mention that it would be interesting to extend our analytical result for the 1/dc- expansion of the differential Poisson's ratio of a two-dimensional membrane in two directions. First, one can address in a similar way the absolute Poisson ratio. (In this case, the zeroth-order term corresponding to the limit dc = ∞ is equal to −1, see Ref. [22].) Second, the case of a disordered membrane [21] is of interest. 15 012345678(cid:45)0.3(cid:45)0.2(cid:45)0.100.10.20.30.4Q(cid:72)
Q

(cid:80)2
Q
5. Acknowledgements We are grateful to E. Kats, M. Katsnelson, I. Kolokolov, V. Lebedev, and J. Los for use- ful discussions. The work was funded in part by Deutsche Forschungsgemeinschaft, by the Alexander von Humboldt Foundation, and by Russian Science Foundation under the grant No. 14-42-00044. Appendix A. Screening of the elastic modulus µ and λ In this Appendix, we present technical details of the calculation of screening of elastic modu- lus. We start from rewriting the term in Eq. (19) which describes the interaction between flexural phonons in a symmetric form [6]: Here we consider a membrane of dimensionality D. The interaction kernel reads (cid:90)  4(cid:89) j=1 1 4  δ  4(cid:88) j=1 k j dDk j (2π)2  Rαβ,γδ(k1 + k2)(cid:0)hk1 hk2 (cid:32) PαγPβδ + PαδPβγ (cid:1)(cid:0)hk3 hk4 (cid:1). (cid:33) − PαβPγδ D − 1 , Rαβ,γδ(q) = N D − 1 PαβPγδ + µ 2 where N = µ(2µ + Dλ)/(2µ + λ). The projection operator is given as Pαβ = δαβ − qαqβ q2 . The screened interaction kernel obeys [6]: (A.1) (A.2) (A.3) (A.4) δγ(cid:48)δ(cid:48)qα(cid:48)qβ(cid:48) The polarization operator at finite momenta can be written as [16] Rαβ,γδ(q) = Rαβ,γδ(q) − Rαβ,γ(cid:48)δ(cid:48)(q) Πγ(cid:48)δ(cid:48),α(cid:48)β(cid:48)(q) Rα(cid:48)β(cid:48),γδ(q). (cid:17) (cid:17) (cid:16) (cid:0)Πxx(q) − Πxy(q)(cid:1)(cid:16) δγ(cid:48)α(cid:48) δδ(cid:48)β(cid:48) + δγ(cid:48)β(cid:48) δδ(cid:48)α(cid:48) δγ(cid:48)β(cid:48)qδ(cid:48)qα(cid:48) + δγ(cid:48)α(cid:48)qδ(cid:48)qβ(cid:48) + δδ(cid:48)α(cid:48)qγ(cid:48)qβ(cid:48) + δδ(cid:48)β(cid:48)qγ(cid:48)qα(cid:48) 1 D (cid:16) + Π1(q) (cid:17) Πγ(cid:48)δ(cid:48),α(cid:48)β(cid:48)(q) = Πxy(q)δγ(cid:48)δ(cid:48) δα(cid:48)β(cid:48) + +δα(cid:48)β(cid:48)qγ(cid:48)qδ(cid:48) + Π2(q) + Π3(q)qα(cid:48)qβ(cid:48)qγ(cid:48)qδ(cid:48) . (A.5) Because of the projection operators entering Rαβ,γ(cid:48)δ(cid:48), the components Π1(q), Π2(q), and Π3(q) of the polarization operator drop from Eq. (A.4). This equation can be solved by Rαβ,γδ which has exactly the same structure as Rαβ,γδ, Eq. (A.2), but with the screened coefficients N(cid:48) and µ(cid:48) instead of N and µ, respectively: (cid:48)(q) = µ 1 + (cid:16) Πxx(q) − Πxy(q) µ (cid:17) (cid:16) , N(cid:48)(q) = µ 1 + N 2Πxx(q) + (D − 2)(D + 1)Πxy(q) . (A.6) N/D Within the self-consistent screening approximation the following relation holds: Πxx(q) = (D + 1)Πxy(q) ≡ (D + 1)Π(0) q , and we reproduce the results of Ref. [6]. For D = 2, we can rewrite these equations in the following way: (cid:48)(q) = µ 1 + (cid:16) Πxx(q) − Πxy(q) µ (cid:17) , µ B(cid:48)(q) = 1 + B Πxx(q) + Πxy(q) . B (A.7) (cid:16) (cid:17) (cid:17) The result (A.7) generalizes Eq. (30) to a finite momentum transfer. 16 vector r: Appendix B. Ward identity In this Appendix we discuss the Ward identity for the elastic action and its consequences for small-momentum behaviour of exact propagators of flexural phonons. While the main text focuses on the high-temperature regime, here we discuss a more general case of arbitrary tem- peratures. For the sake of simplicity, we consider the case d = 3. Appendix B.1. Basic equations We start from the following imaginary-time Lagrangian written in terms of the 3-dimensional (cid:16) ∂αr∂βr − δαβ (cid:17)2 (cid:16) ∂αr∂αr − 2 (cid:17)2 λ 8 κ 2 L[r] = ρ(∂τr)2 + ((cid:52)r)2 + (B.1) Here Greek indices correspond to the 2D coordinates (x, y) ≡ x parameterizing the membrane. We note that substituting r = {ξxx + ux, ξyy + uy, h} into Eq. (B.1) yields the membrane action (4). The Lagrangian (B.1) is manifestly invariant under O(3) rotations of the vector r. These + . µ 4 rotations can be parameterized as r j → r j + εata (B.2) where εa → 0 are some constants and ta =
a jk are generators of O(3) group. In order to explore implications of this symmetry, we shall follow a standard approach [26, 27]. Let us consider the functional Φ[ Σ] defined as follows jkrk, jk (cid:16)−Φ[ Σ] (cid:17) exp (cid:90) β(cid:90) (cid:40) − (cid:90) (cid:16)L[r] − Σ jα∂αr j (cid:17)(cid:41) = D[r] exp dτ d2x . (B.3) At this stage, Σ jα are arbitrary functions of x and y; as will become clear soon, they have a meaning of components of the stress tensor σ jα [9, 16]. The average deformation 0 can be found as ∂αR j = (cid:104)∂αr j(cid:105) ∂αR j = − δΦ[ Σ] δΣ jα . Evidently, R j transform according to Eq. (B.2) under rotation. Let us now consider the Legendre transform of Φ[ Σ]: β(cid:90) (cid:90) F [R] = Φ[ Σ] + dτ d2x Σ jα∂αR j. (B.4) (B.5) (B.6) Here Σ jα should be found from the solution of Eq. (B.5). There is also the reciprocal relation between Σα j and ∂αR j: 0 Σ jα = δF [R] δ∂αR j 17 . (B.7) We note that F [R] coincides with the free energy evaluated under the constraint (cid:104)∂αr j(cid:105) = ∂αR j, where R is a given function of x and y. the functional F [R]: Now let us introduce the two-point correlation function Sαβ jk (q, ω) as the second variation of Sαβ jk (xτ, x(cid:48) (cid:48)) = τ δ2F [R] δ∂αR j(xτ)δ∂βRk(x(cid:48)τ(cid:48)) . (cid:48)) = −(cid:10)Tτr j(xτ)rk(x(cid:48) (cid:48))(cid:11) We note that the propagator of displacements, G jk(xτ, x(cid:48) (B.9) where (cid:104)···(cid:105) is defined with respect to the Lagrangian L[r]− Σ jα∂αr j and Tτ denotes the ordering along the imaginary time contour, is related with the two-point correlation function Sαβ jk (q, ω) in the following way: Σ, τ τ (B.8) G−1 jk (xτ, x(cid:48) (cid:48)) = τ ∂2 ∂xα∂x(cid:48) β Sαβ jk (xτ, x(cid:48) The rotation symmetry (B.2) implies that Φ[ Σ] = Φ[ Σ(cid:48)], (cid:48)). τ (B.10) (B.11) where Σ(cid:48) identity: jα = Σ jα − εata (cid:90) jk β(cid:90) 0 0 = εata jk dτ Σkα. Expanding this equation to the lowest order in εa, we find the Ward d2x Σkα δΦ[ Σ] δΣ jα = −εata jk d2x ∂αR j δF [R] δ∂αRk . (B.12) (cid:90) dτ β(cid:90) 0 In order to use the Ward identity for analysis of the two-point correlation function, it it convenient to perform a variation of the last part of Eq. (B.12) with respect to ∂γRl(x(cid:48), τ(cid:48)). This yields εata lk Σkγ(x(cid:48) (cid:48)) + εata , τ jk dτ d2x ∂αR j(x, τ)Sαγ kl (xτ, x(cid:48) (cid:48)) = 0. τ (B.13) (cid:90) β(cid:90) 0 β(cid:90) (cid:90) Appendix B.2. The propagator of flexural phonons With the choice ε = {ε, 0, 0}, Eq. (B.13) reduces to zyΣyγ(x(cid:48) tx (cid:48)) + tx , τ yz dτ d2x ∂αRy(x, τ) Sαγ zz (xτ, x(cid:48) (cid:48)) = 0. τ (B.14) 0 Now we consider the function R(x, τ) which has the following form: (B.15) where ξx and ξy are arbitrary constants. The functional F [R(ξ)] corresponds to the free energy evaluated under the constraint (cid:104)∂αr j(cid:105) = ξαδα j, where the average is taken with respect to the R(x, τ) = R(ξ) = {ξxx, ξyy, 0}, 18 Lagrangian L[r]. This is exactly the action S (see Eq. (4)) discussed in the main text. Using Eq. (B.14), we find Syy zz (q, ω) = Σyy = Sxx zz (q, ω) = Σxx = ξy lim ω,q→0 ξx lim ω,q→0 ∂ f ∂ξy ∂ f ∂ξx , , Syx zz (q, ω) = 0, Sxy zz (q, ω) = 0. ξy lim ω,q→0 ξx lim ω,q→0 (B.16) We recall that the physical stress is defined by Eq. (7). Therefore, we obtain (B.17) By virtue of Eq. (B.10), this implies that the inverse propagator of the flexural phonons for q → 0 has the following exact form: lim ω,q→0 Sαγ zz (q, ω) = σαδαγ. G−1 zz (q, ω) = σxq2 x + σyq2 y + . . . lim ω→0 (B.18) We note that Eq. (B.18) extends the statement of Refs. [9, 16] to the case of σx (cid:44) σy. References [1] L. D. Landau and E. M. Lifshitz, Course of Theoretical Physics, vol.7: Theory of Elasticity, (Butterworth Heine- [2] K. E. Evans, M. A. Nkansah, I. J. Hutchinson, and S. C. Rogers, Molecular network design, Nature 353, 124 mann, 1986) (1991). [3] A. E. H. Love, A Treatise on the Mathematical Theory of Elasicity (Dover, New York, 4th ed., p.163, 1944). [4] R. Lakes, Foam Structures with a Negative Poisson's Ratio, Science 235, 1038 (1987). [5] J. W. Jiang, S. Y. Kim, and H. S. Park, Auxetic nanomaterials: Recent progress and future development, Appl. [6] P. Le Doussal and L. Radzihovsky, Self-consistent theory of polymerized membranes, Phys. Rev. Lett 69, 1209 [7] D. Nelson, T. Piran, and S. Weinberg (Eds.) Statistical Mechanics of Membranes and Surfaces (World Scientific, [8] F. David and E. Guitter, Crumpling transition in elastic membranes: Renormalization group treatment, Europhys. Phys. Rev. 3, 041101 (2016). (1992). Singapore, 1989) Lett. 5, 709 (1988). Physique 50, 1787 (1989). Rev. E 79, 040101(R) (2009). [9] E. Guitter, F. David, S. Leibler, and L. Peliti, Thermodynamical behavior of polymerized membranes, Journal de [10] J. A. Aronovitz and T. C. Lubensky, Fluctuations of Solid Membranes, Phys. Rev. Lett. 60, 2634 (1988). [11] J.-P. Kownacki, and D. Mouhanna, Crumpling transition and flat phase of polymerized phantom membranes, Phys. [12] G. Gompper and D. M. Kroll, A polymerized membrane in confined geometry, Europhys. Lett. 15, 783 (1991). [13] M. J. Bowick, S. M. Catterall, M. Falcioni, G. Thorleifsson, and K. N. Anagnostopoulos, The flat phase of crys- talline membranes, J. Phys. I France 6, 1321 (1996). [14] J. H. Los, M. I. Katsnelson, O. V. Yazyev, K. V. Zakharchenko, and A. Fasolino, Scaling properties of flexible membranes from atomistic simulations: Application to graphene, Phys. Rev. B 80, 121405(R) (2009). [15] E. Guitter, F. David, S. Leibler, and L. Peliti, Crumpling and Buckling Transitions in Polymerized Membranes, [16] J. Aronovitz, L. Golubovic, T. C. Lubensky, Fluctuations and lower critical dimensions of crystalline membranes, Phys. Rev. Lett. 61, 2949 (1988). J. de Physique, 50, 609 (1989). [17] R. J. T. Nicholl, H. J. Conley, N. V. Lavrik, I. Vlassiouk, Y. S. Puzyrev, V. P. Sreenivas, S. T. Pantelides, and K. I. Bolotin, The effect of intrinsic crumpling on the mechanics of free-standing graphene, Nat. Comm. 6, 8789 (2015). 19 [18] A. Kosmrlj and D. R. Nelson, Response of thermalized ribbons to pulling and bending, Phys. Rev. B 93, 125431 (2016). [19] J. H. Los, A. Fasolino, and M. I. Katsnelson, Scaling behavior and strain dependence of in-plane elastic properties of graphene, Phys. Rev. Lett. 116, 015901 (2016). [20] I. V. Gornyi, V. Yu. Kachorovskii, and A. D. Mirlin, Anomalous Hooke's law in disordered graphene, 2D Materials [21] I.V. Gornyi, V. Yu. Kachorovskii, and A. D. Mirlin, Rippling and crumpling in disordered free-standing graphene, 4, 011003 (2017). Phys. Rev. B 92, 155428 (2015). [22] I. S. Burmistrov, I. V. Gornyi, V. Yu. Kachorovskii, M. I. Katsnelson, J. H. Los, and A. D. Mirlin, Stress-controlled Poisson ratio of a crystalline membrane: Application to graphene, arxiv: [23] I. S. Burmistrov, I. V. Gornyi, V. Yu. Kachorovskii, M. I. Katsnelson, and A. D. Mirlin, Quantum elasticity of graphene: Thermal expansion coefficient and specific heat, Phys. Rev. B 94, 195430 (2016). [24] M. Falcioni, M. J. Bowick, E. Guitter, and G. Thorleifsson, The Poisson ratio of crystalline surfaces, Europhys. [25] D. Gazit, Structure of physical crystalline membranes within the self-consistent screening approximation, Phys. Lett. 38, 67 (1997). Rev. E 80, 041117 (2009). [26] D. J. Amit, Field Theory, the Renormalization Group, and Critical Phenomena (World Scientific, Singapore, 1993). [27] J. Zinn-Justin, Quantum Field Theory and Critical Phenomena, (Clarendon Press, Oxford, 1996) 20 

1406.6135

2

1406

Ultrafast terahertz probes of interacting dark excitons in chirality-specific semiconducting single-walled carbon nanotubes

Ultrafast terahertz spectroscopy accesses the {\em dark} excitonic ground state in resonantly-excited (6,5) SWNTs via internal, direct dipole-allowed transitions between lowest lying dark-bright pair state $\sim$6 meV. An analytical model reproduces the response which enables quantitative analysis of transient densities of dark excitons and {\em e-h} plasma, oscillator strength, transition energy renormalization and dynamics. %excitation-induced renormalization. Non-equilibrium, yet stable, quasi-1D quantum states with dark excitonic correlations rapidly emerge even with increasing off-resonance photoexcitation and experience a unique crossover to complex phase-space filling of %a complex distribution between both dark and bright pair states, different from dense 2D/3D excitons influenced by the thermalization, cooling and ionization to free carriers.

cond-mat.mes-hall

cond-mat

Ultrafast terahertz probes of interacting dark excitons in chirality-specific semiconducting single-walled carbon nanotubes Liang Luo, Ioannis Chatzakis†, Aaron Patz, Jigang Wang∗ Department of Physics and Astronomy and Ames Laboratory-U.S. DOE, Iowa State University, Ames, Iowa 50011, USA. (Dated: April 17, 2021) Ultrafast terahertz spectroscopy accesses the dark excitonic ground state in resonantly-excited (6,5) SWNTs via internal, direct dipole-allowed transitions between lowest lying dark-bright pair state ∼6 meV. An analytical model reproduces the response which enables quantitative analysis of transient densities of dark excitons and e-h plasma, oscillator strength, transition energy renormal- ization and dynamics. Non-equilibrium, yet stable, quasi-1D quantum states with dark excitonic correlations rapidly emerge even with increasing off-resonance photoexcitation and experience a unique crossover to complex phase-space filling of both dark and bright pair states, different from dense 2D/3D excitons influenced by the thermalization, cooling and ionization to free carriers. Quasi-one-dimension (quasi-1D) excitons in single- walled carbon nanotubes (SWNTs), with large binding energies of 100s of meV, naturally arise from strong quan- tum confinement and reduced screening of electron-hole (e-h) pairs [1, 2]. Their internal structure is charac- terized, in a 1D hydrogen atom-like description, by a center-of-mass momentum K and by internal quantum numbers (designated here as 1s, 2s, 2p...). These strong excitonic behaviors manifest themselves in extensive static/ultrafast optical interband absorption/emission spectra in individually separated SWNTs, which are stable even in the presence of unbound e-h carriers from residual tube aggregation and/or metallic tubes [3 -- 6]. However, unlike the hydrogen atom, the corre- lated e-h pairs in SWNTs evolve in a "modified vacuum" with exotic, chiral symmetry arising from the underlying graphene lattice that gives two-fold degeneracy at K and K(cid:48) points. Coulomb interaction splits such "doubling" into odd (u) and even (g) symmetry states entailing a se- ries of bright and optically-forbidden, dark exciton pairs, including the lowest 1s(u) and 1s(g) [7, 8] (Fig. 1(a)). Thus far, the dark ground state 1s(g), hidden from both single- and two-photon optical interband transitions, is still largely unexplored. While magneto-optics and light scattering measure- ments established the existence of dark excitons in SWNTs [9 -- 11], the search for possible dipole-allowed, lowest transitions 1s(g)→1s(u) remains open. Partic- ularly, as these linear probes do not depend critically on quasi-particle/exciton interaction and the associated fast dynamics, they provide little insight into interacting states that are characterized by the co-existence of both the dark & bright excitons and their complex interplay with e-h plasma. Exactly these aspects govern the ra- diative lifetime and photoluminesce (PL) efficiency, key for SWNT-based optoelectronic applications [7, 12]. The lack of quantitative probes and scarce ultrafast measure- ments for dark states seriously limit their thorough un- derstandings and perspectives of exploring related novel quantum phenomena, e.g., the excitonic Bose condensate [13] and other exotic ground states. Selective optical pump and THz probe technique rep- resents a versatile spectroscopy tool that is extremely relevant for quantitative study of dark excitons. THz pulses directly couple the 1s(g)→1s(u) transitions [ar- rows, Fig. 1(a)], which represents a direct consequence and measure of excitonic pair correlations and resulting dark states in SWNTs. Being independent of momen- tum K, THz pulses measure genuine dark exciton pop- ulation across entire K space, while the available inter- band optical probes (PL, absorption), due to symmetry and momentum conservation, only detect a subset of ex- citons near K=0. Additionally, the capability of resonant and off-resonant excitations by tuning the pump photon energy enables, among others, the study of excitons in single chirality tube and of exciton-plasma interaction. These critical aspects for dark exciton studies are ab- sent in prior THz experiments of SWNTs. They reveal some other interesting low-energy electrodynamics, e.g., the absorption band centered ∼4 THz [14], and evidence for the internal transitions and non-Drude conductivity [15]. In this Letter, we reveal lowest lying dark excitons and their interacting states in semiconducting SWNTs. With resonant excitation to the (6,5) E22 interband transition at low temperature, transient THz spectra evidence strong photoinduced absorption centered ∼6 meV, whose transition energy, pump photon and tem- perature dependence, and dynamics manifest the ob- servation of the 1s(g)→1s(u) transition. Most intrigu- ingly, the 1s(g)→1s(u) oscillator signals emerge quasi- instantaneously even with increasing off-resonance pho- toexcitation and drop significantly above ∼130 µJ cm−2, at which the intra-excitonic resonances are still stable with little shift or broadening. This appears to be dis- tinctly different from dense 2D and 3D excitons that are influenced by the thermalization, cooling and ionization to free carriers [16]. The robust non-equilibrium quasi- 1D many-exciton states, instead, uniquely evolve from a predominant dark exciton population to complex phase- 5 1 0 2 n a J 4 2 ] l l a h - s e m . t a m - d n o c [ 2 v 5 3 1 6 . 6 0 4 1 : v i X r a 2 1 1 (ω) (a) and ∆εNT FIG. 2. Ultrafast THz spectra of ∆σNT (ω) (b) (red dots), measured at pump-probe delay ∆τ =0.5 ps after 1.55 eV pumping. Thick line is the fit using the analyt- ical model, Eq.(1), which is sum of the 1s(g)→1s(u) excitonic (dashed green) and unbound e-h responses (dash-dotted blue) (main text). Inset: THz fields transmitted (raw data). Com- parison of photoinduced ∆σNT (ω) at ∆τ =0.5 ps and under excitation density 1.4×1014 cm−2: (c) Off-resonant pumping at 1.55 eV and T =5K; (d) on-resonant pumping at 2.17 eV and T =5K; (e) on-resonant pumping at 2.17 eV and T =300K. 1 εNT (ω), which measures the absorbed power and the out- 1 of-phase, inductive response, respectively [18]. Effective medium theory is applied to obtain the SWNT dielectric function from the experimental data (supplementary). 1 1 1 (ω) are shown in Figs. Representative ultrafast THz responses ∆σNT (ω) and ∆εNT 2(a)-2(b) (red dots) at ∆τ =0.5 ps and T =5 K, after 1.55 eV pumping with pho- ton density 3.9×1014 cm−2. A strong photo-induced ab- sorption appears in ∆σNT (ω) within the time resolution with a broad, resonant spectral shape centered at ∼6 meV and a dispersive zero crossing occurs in ∆εNT (ω) at slightly lower photon energy. These two features are characteristic of a well-defined driven THz oscillator while the shift between their spectral positions and the non-vanishing conductivity, albeit small, at the lowest probe energy ∼2 meV indicate additional low-frequency spectral weight, in the form of a coexisting unbound e-h plasma as discussed later. We emphasize three key facts to assign this resonance mainly from the 1s(g)→1s(u) transition of (6,5) tubes [19]. First, resonant photoex- citation of the (6,5) interband E22 excitonic transitions 1 FIG. 1. (a) Schematic of two-particle e-h pair dispersion, il- lustrating the lowest lying 1s(g)→1s(u) intra-excitonic transi- tions (arrows) resonantly probed by THz pulses. Absorption spectra (b) and time/spectral-resolved differential transmis- sion ∆T /T after 1.55 eV photoexcitation (c) of the SWNT sample at T =300 K. space filling of both dark and bright pair states. We study Co-Mo-catalyst grown SWNTs of mainly (6,5) and (7,5) chiralities embedded in a freestand- ing 50 µm Sodium Dodecylbenzenesulfonate (SDBS) film through drying a D2O solution of SDBS-dispersed SWNTs (supplementary) [17]. The absorption spectrum in Fig. 1(b) exhibits the distinct E11 and E22 quan- tized interband absorption peaks of dominant (6,5) and (7,5) chiralities, each associated with substantially pro- longed relaxation times and significantly enhanced photo- bleaching, as shown in the differential transmission spec- tra (Fig. 1(c)). These observations are consistent with the extensive interband studies of high-quality chirality- enriched SWNT samples at visible and near-infrared spectral regions, which are explained by the joint effects of resonantly-enhanced radiative lifetime and exciton- exciton annihilation (EEA) [3, 4]. Our optical pump, THz probe spectroscopy setup is driven by a 1 kHz Ti:Sapphire regenerative amplifier (40 fs, 800 nm center wavelength), which pumps an optical parametric amplifier and 1 mm thick (cid:104)110(cid:105) ZnTe crys- tals (supplementary). The THz fields in time domain are measured, shown in the inset of Fig. 2(a), for trans- mission through reference (a clear aperture in our case) Eref(t) (gray), unexcited sample ETHz(t) (blue), and its pump induced change ∆ETHz(t) after a pump-probe de- lay ∆τ =0.5 ps (red). Through the fast Fourier trans- formation and Fresnel equation, the full THz dielectric response is obtained and expressed as the real part of the conductivity, σNT (ω), and of the dielectric function, 1 at 2.17 eV, shown in Fig. 2(d), leads to significant en- hancement of the transient THz resonance ∼6 meV, e.g., roughly three times as large as the 1.55 eV off-resonant excitation (Fig. 2(c)), for the same excitation photon density n=1.4×1014 cm−2 per pulse. This clearly shows the excitonic origins of the THz resonance in (6,5) tubes. Second, raising the initial lattice temperature to T =300 K (Fig. 2(e)) diminishes the resonance as compared to T =5 K (Fig. 2(d)), for the same resonant excitation. Third, the resonance occurs in the transparent region of the unexcited sample and is close to the 1s(g)→1s(u) transition energy of (6,5) tubes indirectly extrapolated in high field magneto-optical studies [9, 10]. For a quantitative analysis, a theoretical model is used to reproduce the experimentally determined com- plex THz dielectric functions. The model consists of two components: the THz dielectric function of the intra- excitonic transitions ε1s g→u(ω) plus a Drude term of e-h plasma εeh (ω) εNT(ω) = ε1s g→u(ω) + εeh (ω) = ] − g→u)2 − ω2 − iωΓ g→u F 1s (ω1s [ε∞ + ω2 p ω2 + iωγ . (1) p)1s g − n1s g→u = f 1s g→u · ((cid:52)ω2 For the first component (dashed green), F 1s g→u denotes effective transition strength of the intra-excitonic ω1s g→u resonance as F 1s g→u, where f is p = ne2/ε0m. The (cid:52)ω2 the oscillator strength and ω2 p term, thereby, measures the population difference be- tween the lowest dark and bright pair states (Fig. 1(a)), i.e., ∆nX = n1s u . The second, Drude component (dash-dotted blue) is added with ω2 p proportional to the density of unbound e-h density neh . ε0 and ε∞ are the vacuum and background permittivity, and e and m are the electron charge and the effective exciton (or electron) mass obtained from [20]. Such composite THz response model (solid black lines) provides an excellent agreement with the low temperature data (red dots) in Figs. 2(a)- 2(b) by varying the ω1s g→u, carrier density neh , and the exciton (carrier) broadening Γ (γ). The best fit are strongly restrained by the requirement of simul- taneously satisfying both dielectric responses ∆σNT (ω) and ∆εNT (ω), over a broad spectral range, and by the distinctly different spectral shapes of the excitonic oscil- lator and Drude carriers. g→u and F 1s 1 1 1 Fig. 3(a) presents in detail of ultrafast THz conduc- tivity spectra ∆σNT (ω) for various time delays ∆τ af- ter the resonant E22 photoexcitation at 2.17 eV. Pump fluence is 48 µJ cm−2 and T =5 K. The transient spec- tra are characterized by the ω1s g→u resonance that re- tains its shape as the amplitude decays with time. The resonance peaks (dash line) exhibit no noticeable shift. The results can be fitted very well by the model de- scribed above (red lines). To further determine the ex- g→u)·(ε0m/e2), the os- citon density ∆nX = (F 1s cillator strength f is needed for the 1s(g)→1s(u) tran- g→u/f 1s 3 1 (a) photo-induced conductivity changes ∆σNT FIG. 3. (ω) (black dots) at several pump-probe delays after resonant (6,5) E22 excitation at 2.17 eV and T =5 K. Shown together are the model calcualtion (red lines) and peak positions (dash line). g→u ·(ε0m/e2)/(nph − neh) as a function (b) The ratio R = F 1s of fluence. (c) The reciprocal of exciton (∆nX) and unbound e-h (neh) densities as a function of time delay together with fitting. The traces are vertically offset for clarity. Inset: tem- poral decay of ∆nX and neh. u /n1s sition which has never been determined before. This can be directly obtained here via fluence dependence g→u ·(ε0m/e2)/(nph − neh) down to of the ratio R = F 1s I =3µJ cm−2 in Fig. 3(b), where nph is the actual ab- sorbed photon density after taking into account the re- flection and transmission of the pump beam in the op- tical path. This ratio, with sufficiently weak photoex- g→u · [1 − α(I)]/[1 + α(I)] with citation, follows R = f 1s α = n1s g , considering strongly limited EEA in di- lute exciton gas and fast inter-subband E22-to-E11 relax- ation ∼40 fs [21, 22]. The measurement clearly shows R becomes mostly pump-power-independent below ∼10µJ cm−2 which indicates α (cid:28) 1 [23] and the ratio converges g→u ≈0.79. This is consistent with the to a constant f 1s fact that the optically generated exciton density is ≈1 per tube at ∼10µJ cm−2 estimated from the tube den- sity and linear absorption. Increasing pump fluence to 10s of µJ cm−2 leads to a significant drop of the ratio (Fig. 3(b)), which can be attributed to efficient EEA that lowers the photon-exciton conversion efficiency. The obtained density ∆nX and its relaxation dynam- ics (black dots) are well described by a bimolecular de- cay shown in Fig. 3(c). Although it exhibits a highly non-exponential profile over ps time scales (the inset), a reciprocal plot of 1/∆nX(t), instead, yields a sim- i.e., 1/∆nX(t) ∼ βt (black dots), ple straight line, which is the hallmark of the bimolecular decay given by d X(t) with decay rate of sheet dt ∆nX(t) = −(1/2)β∆n2 4 1 (ω) shifts to lower frequency (Fig. 4(d)), produces very well the experimental results, which are divided into individual components in the same manner as Fig. 2(a): the intraexcitonic 1s(g)→1s(u) (green lines) and unbound e-h carriers (blue lines). Note the bleach- ing component of the 4 THz band (pink lines) [14, 25] is added and only contributes at high pumping above 512 µJ cm−2 (supplementary). Two salient features are visi- ble with increasing pump fluence: (1) The first feature is that the ω1s g→u resonance in ∆σNT illus- trated by two dash lines in Fig. 4(a), ∼20 % change from 15 to 814 µJ cm−2. We attribute this to the critical role of plasma-exciton interaction in the renormalization of the excitonic levels, considering the unbound e-h den- sity significantly increases in the off-resonance pumping, e.g., neh can reach ∼2.5×1013cm−2, comparable to ex- citon density, shown in Figs. 4(c) and 4(e). Note that the ω1s g→u shift is not seen in the system of a predomi- nant exciton population, shown in Fig. 3 for the resonant pumping, which has larger exciton density but one order of magnitude smaller neh . This reconciles some contra- dictory results regarding the exciton stability in the liter- ature, e.g., photoluminescence spectra of dense excitons have shown no excitonic shift up to complete absorption saturation [4, 22], while ultra-violet pumping results in some effects [6]. This can be underpinned unambiguously here to the e-h plasma-exciton interaction. (2) The second feature is that the exciton density ∆nX exhibits a distinct non-monotonic variation with increas- ing photoexcitation (Fig. 4(e)), which peaks at IC ∼130 µJ cm−2 followed by a significant decrease with further increasing pump fluence. However, most interestingly, the intra-excitonic resonance ω1s g→u exhibits little shift or broadening at IC (<5%, dash arrow in Fig. 4(d)), indicat- ing that there is no obvious exciton ionization to unbound e-h plasma. This appears to be fundamentally different from dense 2D and 3D excitons where the strong decrease of excitonic signals occurs only upon the ionization of ex- citons manifested by a significant shift, broadening, and ultimately the disappearance of the THz resonances. The quasi-1D many-body state in SWNTs reveal, instead, the evolution from a predominant dark exciton population in the 1s(g) to phase space filling of both the lowest dark and bright exciton pair states [26]. Such crossover at IC is illustrated by two shaded colors in Fig. 4(e): below IC, photoexcited excitons primarily populate in the low- est lying dark state 1s(g) that leads to a rapid rise of ∆nX with increasing pump fluence; above IC, photoexcited ex- citons have larger probability to populate the 1s(u) state rather than the 1s(g) ground state, due to electronic heat- ing of the many-body systems, which is responsible for the reduction of the ω1s g→u resonance via Pauli blocking of the transition. The 1s(g)/1s(u) exciton pairs, being well-isolated from higher lying exciton levels and contin- uum (>200 meV) [1, 2, 5], is stable against high density ionization. 1 1 (ω) and ∆εNT (a)-(b) photo-induced ∆σNT FIG. 4. (ω) changes (red dots) at various pump fluences after 1.55 eV excitation. ∆τ =0.5 ps and T =5K. The model fitting (black lines) is divided (see text) and the dashed lines indicate the shift of the 1s(g)→1s(u) peak positions. (c) The unbound e-h density as a function of pump fluence. (d) The resonance peak as a function of unbound e-h density. (e) Fluence dependence of the exciton signal ∆nX, which is divided into two regimes at IC ∼130 µJ cm−2 (dash arrow). density β=2.3×10−14 cm2 ps−1 [3]. The 1/neh (t) decay (red dots), on the contrary, show large variation from the bimolecular behavior, which further underscores the excitonic origin of the ∆nX response. Furthermore, the strong similarity between the ∆nX(t) decay and firmly established exciton bimolecular kinetics in SWNTs also indicates ∆nX(t)≈ n1s g , i.e., it quantitatively follows the 1s(g) exciton density. This shows that, under current pumping conditions, most excitons populate in the 1s(g) ground state and n1s u . Therefore, the resonant g ∼90% of the photoexcitation generates dark excitons n1s total density, much larger than neh (inset, Fig. 3(c)). g (cid:29) n1s 1 1 (ω) and ∆εNT Increasing off-resonant excitation reveals the complex many-particle interacting state. Figs. 4(a)-(b) shows ∆σNT (ω) for various pump fluences af- ter 1.55 eV photoexcitation at T =5 K. The spectra at ∆τ =0.5 ps clearly exhibit the ω1s g→u resonance, despite 100s of meV detuning away from the excitonic reso- nances, which dominates e-h plasma especially at low pumping. This reveals an ultrafast sub-ps formation of non-equilibrium 1D dark excitonic correlations, much faster than the 100s of ps thermalization of bright ex- citons to the dark ground states seen in time-resolved PL [24]. The difference occurs since the initial electron- ically correlated states are mostly hidden from those in- terband probes. The formation also appears to be much more robust than 2D/3D excitons that are strongly in- fluenced by the co-existing e-h plasma with 100s of ps cooling time under the off-resonant pumping [16]. The composite THz model (black lines) again consistently re- 5 L. Luo, et al., Nat. Commun. 5, 3055 (2014). [19] A phonon resonance scenario can be excluded since the resonances (or periodical oscillations) are absent in static THz conductivity spectra (or the photoinduced temporal traces). [20] A. Jorio et al., Phys. Rev. B 71, 075401 (2005). [21] C. Manzoni et al., Phys. Rev. Lett. 94, 207401 (2005). [22] Y. Murakami, and J. Kono, Phys. Rev. Lett. 102, 037401 [23] The small n1s (2009). g is consistent with the estimation from time-averaged photoluminance ≈0.06−0.08 below 10K [24]. u /n1s [24] R. Matsunaga, Y. Miyauchi, K. Matsuda and Y. Kane- mitsu, Phys. Rev. B. 80, 115436 (2008). [25] The same 4 THz bleaching component also contributes at the high lattice temperature, e.g., the 300K spectral shape in Fig. 2(e). [26] Note that the photoexcited exciton density is still below the expected Mott density in SWNTs [22]. [27] T. Li, et. al., Nature 496, 69 (2013); A. Patz, et. al., Nat. Commun. 5, 3229 (2014); T. Li, et. al., Phys. Rev. Lett. 108, 167401 (2012) In conclusion, we provide the first insights into the chirality-specific THz response of non-equilibrium exci- tonic correlations and dynamics from the dark ground states in SWNTs. The THz 1s(g)→1s(u) resonant probes revealed, being independent of ground state symmetry and momentum conservation restrictions, identify that strong electronic correlation regulates the sub-ps forma- tion of 1D dark excitons beyond the previously-held slow thermalization/cooling scenario. This may also evolve into a benchmark approach for quantitative exciton man- agement in SWNT-based device development, and moti- vates for fundamental quantum phase discovery of exci- tons [13] and other strongly correlated excitations [27]. This work was supported by the US Department of En- ergy, Office of Basic Energy Science, Division of Materials Sciences and Engineering (Ames Laboratory is operated for the US Department of Energy by Iowa State Univer- sity under Contract No. DE-AC02-07CH11358). ∗Corresponding author. jgwang@iastate.edu; jwang@ameslab.gov †Present address: Ming Hsieh Department of Electri- cal Engineering, University of Southern California, 3737 Watt Way, Los Angeles, CA 90089-0271 [1] F. Wang, G. Dukovic, L. E. Brus, and T. F. Heinz, Sci- ence 308, 838 (2005). [2] J. Maultzsch, et al., Phys. Rev. B 72, 241402(R) (2005). [3] Y. Z. Ma, L. Valkunas, S. L. Dexheimer, S. M. Bachilo, and G. R. Fleming, Phys. Rev. Lett. 94, 157402 (2005). [4] G. N. Ostojic et al., Phys. Rev. Lett. 94, 097401 (2005). [5] J. Wang, M. W. Graham, Y. Ma, G. R. Fleming, and R. A. Kaindl, Phys. Rev. Lett. 104, 177401 (2010). [6] J. J. Crochet et al., Phy. Rev. Lett. 107, 257402 (2011). [7] H. Zhao, and S. Mazumdar, Phys. Rev. Lett. 93, 157402 (2004). [8] V. Perebeinos, J. Tersoff, and P. Avouris, Nano Lett. 5, 2495 (2005). [9] See, for example, J. Kono, R. J. Nicholas and S. Roche, Topics in Applied Physics 111, 393 (2008). [10] R. Matsunaga, K. Matsuda, and Y. Kanemitsu, Phys. Rev. Lett. 101, 147404 (2008). [11] O. N. Torrens, M. Zheng, and J. M. Kikkawa, Phys. Rev. Lett. 101, 157401 (2008). [12] F. Xia, M. Steiner, Y. Lin, and P. Avouris, Nat. Nan- otech. 3, 609 (2008). [13] L. V. Butov, A. C. Gossard, and D. S. Chemla, Nature 418, 751 (2002). [14] T. Kampfrath et al., Phys. Rev. Lett. 101, 267403 (2008). [15] X. Xu, et al., J. Phys. Chem. C 113, 18106 (2009). [16] R. Huber, R. A. Kaindl, B. A. Schmid, and D. S. Chemla, Phys. Rev. B 72, 161314(R) (2005); R. A. Kaindl, D. Hagele, M. A. Carnahan, R. Lovenich, and D. S. Chemla, Nature 423, 734 (2003). [17] T. Ogawa, S. Watanabe, N. Minami, and R. Shimano, Appl. Phys. Lett. 97, 041111 (2010). [18] I. Chatzakis, et al., Phys. Rev. B 86, 125110 (2012); I. Chatzakis, et al., Appl. Phys. Lett. 103, 043101 (2013); 

1501.02313

1

1501

Titanium trisulfide monolayer: A new direct-gap semiconductor with high and anisotropic carrier mobility

A new two-dimensional (2D) layered material, namely, titanium trisulfide (TiS$_3$) monolayer sheet, is predicted to possess desired electronic properties for nanoelectronic applications. On basis of the first-principles calculations within the framework of density functional theory and deformation theory, we show that the TiS$_3$ 2D crystal is a direct gap semiconductor with a band gap of 1.06 eV and high carrier mobility. More remarkably, the in-plane electron mobility of the 2D TiS$_3$ is highly anisotropic, amounting to $\sim$10,000 cm$^2$V$^{-1}$s$^{-1}$ in the \emph{b} direction, which is higher than that of the MoS$_2$ monolayer. Meanwhile, the hole mobility is about two orders of magnitude lower. We also find that bulk TiS$_3$ possesses lower cleavage energy than graphite, indicating high possibility of exfoliation for TiS$_3$ monolayers or multilayers. Both dynamical and thermal stability of the TiS$_3$ monolayer is examined via phonon-spectrum calculation and Born-Oppenheimer molecular dynamics simulation in \emph{NPT} ensemble. The predicted novel electronic properties render the TiS$_3$ monolayer an attractive 2D material for applications in future nanoelectronics.

cond-mat.mes-hall

cond-mat

a Titanium trisulfide monolayer: A new direct-gap semiconductor with high and anisotropic carrier mobility Jun Dai1 and Xiao Cheng Zeng1, ∗ 1Department of Chemistry, University of Nebraska-Lincoln, Lincoln, NE 68588, USA (Dated: January 13, 2015) A new two-dimensional (2D) layered material, namely, titanium trisulfide (TiS3) monolayer sheet, is pre- dicted to possess desired electronic properties for nanoelectronic applications. On basis of the first-principles calculations within the framework of density functional theory and deformation theory, we show that the TiS3 2D crystal is a direct gap semiconductor with a band gap of 1.06 eV and high carrier mobility. More remarkably, the in-plane electron mobility of the 2D TiS3 is highly anisotropic, amounting to ∼10,000 cm2V−1s−1 in the b direction, which is higher than that of the MoS2 monolayer. Meanwhile, the hole mobility is about two orders of magnitude lower. We also find that bulk TiS3 possesses lower cleavage energy than graphite, indicating high possibility of exfoliation for TiS3 monolayers or multilayers. Both dynamical and thermal stability of the TiS3 monolayer is examined via phonon-spectrum calculation and Born-Oppenheimer molecular dynamics simula- tion in NPT ensemble. The predicted novel electronic properties render the TiS3 monolayer an attractive 2D material for applications in future nanoelectronics. The successful isolation of two-dimensional (2D) graphene in 2004 [1] has captivated great research interests in 2D mate- rials, particularly atomic-layered materials with weak inter- layer van der Waals bonding[2]. Besides the graphene[3– 6], the family of 2D layered materials also include transition metal dichalcogenides (TMDCs)[2, 7, 8], hexagonal boron nitride (h-BN)[9, 10], silicene[11–13], germanene[14], and phosphorene[15, 16], among others. These 2D atomic-layered materials exhibit not only novel 2D geometries as they repre- sent the thinnest crystalline solids that can be formed, but also new and exotic condensed matter phenomena that are absent in their bulk counterparts[2, 8]. For example, single graphene sheet is a zero-gap semiconductor with a linear Dirac-like dis- persion near the Fermi level, while bulk graphite is known to exhibit semimetallic behavior with bandgap overlap of ∼41 meV.[17] A MoS2 monolayer sheet possesses a direct bandgap of ∼1.8 eV, while bulk MoS2 possesses an indirect bandgap of 1.29 eV[18]. The bandgap of a few layer phosphorene (a 2D form of black phosphorus) is highly layer-dependent. For the phosphorene monolayer, the bandgap is ∼1.5 eV while for the bulk phosphorus, the bandgap is merely ∼0.3 eV[16, 19–22]. 2D layered materials offer opportunities for a variety of ap- plications, particularly in next-generation electronic devices such as field-effect transistors (FET) and logic circuits. For high-performance FET applications, a 2D material should possess a moderate bandgap and reasonably high in-plane car- rier mobility. Graphene is a highly promising 2D material for high-speed nanotransistors due to its massless charge carriers. However, it lacks a bandgap for controllable operations.[23– 25] The molybdenum disulfide (MoS2) monolayer sheets are more promising for FET applications since not only they pos- sesses a direct bandgap of ∼1.8 eV,[18] but also the 2D MoS2- based FET devices show good performance with a high on/off ratio of ∼108 as well as a carrier mobility of ∼200 cm2V−1s−1. The latter can be enhanced even up to 500 cm2V−1s−1 with improvement.[26, 27] Also, recent experiments demonstrated that FET devices built upon few-layer phosphorene exhibit reasonably high on/off ratio (up to 104) and appreciably high hole mobility of ∼55 cm2V−1s−1 (at a thickness of ∼5 nm) to ∼1000 cm2V−1s−1 (at a thickness of ∼10 nm).[15, 16] Nevertheless, new 2D layered materials with moderate direct bandgap and high carrier mobility are still highly sought. In this work, we show an ab initio calculation evidence of a new 2D layered material – the TiS3 monolayer sheet – with the desired electronic properties. Historically, bulk materials such as graphite, TMDCs and black phosphorous were studied well ahead of their 2D layered-material counterparts. Likewise, properties of bulk TiS3 are known much earlier than those of 2D form. Bulk TiS3 has a monoclinic crystalline structure (with the space group of p21/m), and the TiS3 crystal can be viewed as stacked parallel sheets with each sheet being composed of 1D chains of triangular TiS3 unit. These sheets interact with one an- other via the van der Waals (vdW) forces.[28, 29] It is also known that materials with stacking-layer structures can be a good precursor for contriving 2D atomic layers via either exfoliation[30, 31] or mechanical cleavage[32]. Several elec- trical and transport measurements have been reported,[33, 34] showing that the bulk TiS3 is an n-type semiconductor with carrier mobility of ∼30 cm2V−1s−1 at room temperature. The mobility can be further enhanced up to ∼100 cm2V−1s−1 at the low temperature 100 K.[34] Moreover, optical absorption measurements indicate that the bulk TiS3 exhibits an opti- cal gap about 1 eV.[35] More importantly, several recent ex- periments demonstrate that thin films of TiS3 with thickness of 102 nanometers possess a direct bandgap of ∼1.1 eV and exhibit good photo response.[36–38] The moderate bandgap of bulk TiS3 coupled with relatively high carrier mobility renders the bulk TiS3 a highly promising precursor for iso- lating 2D TiS3 sheets with desired properties for nanoelec- tronic applications. For the 2D TiS3 monolayer sheet, ge- ometrical optimization and electronic structure calculations are carried out using density-functional theory (DFT) meth- ods within the generalized gradient approximation (GGA) and with the Perdew-Burke-Ernzerhof (PBE) exchange correla- tion functional, as implemented in the Vienna ab initio sim- low dimensional systems.[19, 44–48] On the basis of effective mass approximation, the charge mobility in 2D materials can be expressed as: 2 µ = 2e3C 3kBTm∗2E2 1 (1) Here, C is the elastic modulus defined as C = [∂2E/∂δ2]/S 0, where E is the total energy of the system (per supercell), and δ is the applied uniaxial strain, and S 0 is the area of the op- timized 2D structure. m∗ is the effective mass, which can be given as m∗ = 2(∂2E/∂k2)−1 (where  is the Plancks constant and k is magnitude of the wave-vector in momentum space), T is the temperature, and E1 is the deformation potential con- stant, which is proportional to the band edge shift induced by the strain. E1 is defined as δE = E1 × (δl/l0), where δE is the energy shift of the band edge position with respect to the lattice dilation δl/l0 along either direction a or b, the energies of the band edges are calculated with respect to the vacuum level. The PBE-D2 optimized structure of bulk TiS3 is shown in Fig.1(a), and the associated lattice constants are a = 4.982 Å, b = 3.392 Å and c = 8.887 Å, and lattice angle β=97.24◦, all in very good agreement with the experimental results, a = 4.958 Å, b = 3.401 Åand c = 8.778 Å, and β=97.32◦.[28] Further- more, the computed band structures of the bulk TiS3 from both PBE-D2 and HSE06 are shown in Supplemental Material[49]. PBE-D2 and HSE06 give qualitatively the same results except for the band gap. Both PBE-D2 and HSE06 calculations indi- cate that the bulk TiS3 is an indirect gap semiconductor from Γ (0, 0, 0) to Z (0, 0, 0.5). The PBE-D2 computation gives a band gap of 0.21 eV, while HSE06 gives 1.02 eV. The latter agrees well with the measured optical gap which is around 1 eV.[35] The good agreement between the benchmark calcula- tions and experiments for the bulk TiS3 show that the theoreti- cal methods chosen for this system is reliable. In addition, the band structures near the conduction band minimum (CBM) or the valence band maximum (VBM) of the bulk TiS3 exhibit notable in-plane dispersion behavior (from Γ to Y (0, 0.5, 0)) or Γ to B (0.5, 0, 0)), indicating that the 2D TiS3 monolayer sheet may have relatively high carrier mobility. The computed HSE06 band structures and density of states (DOS) of the TiS3 monolayer are shown in Fig.2(a). Since the original Z point of the bulk TiS3 folds back to the Γ point for the TiS3 monolayer, the TiS3 undergoes an indirect-direct transforma- tion from an indirect band gap semiconductor for the bulk to a direct band gap semiconductor for the 2D monolayer coun- terpart, akin to the case of MoS2.[18] Both VBM and CBM are located at the Γ point, yielding a direct band gap of 1.06 eV. Moreover, from the orbital and atom projected DOS, we can see that the valance bands exhibit strong hybridization be- tween the S p states and Ti d states from -2 eV to the top of valence band, while the conduction bands are mainly con- tributed from the d states of Ti (see Fig.2(a)). The isosurface plots of the VBM and CBM are shown in Fig.2 (b), which show that the holes (from VBM) favor the a direction, while FIG. 1. (Color online)(a)A 2×2×1 supercell of the bulk TiS3 struc- ture, (b) top view of a 2×2 TiS3 monolayer sheet (left), and the first Brillouin zone and the high symmetry points associated with the monolayer (right). The grey and yellow spheres refer to Ti and S atoms, respectively. ulation package (VASP).[39] The Grimme’s D2 dispersion correction[40] is adopted to account for the long-range vdW interactions. The ion-electron interaction is treated using the projector-augment-wave (PAW) technique and a kinetic en- ergy cutoff of 500 eV is chosen. A vacuum space of ∼20 Å along the direction normal to the monolayer plane is un- dertaken so that the interlayer interaction due to the periodic boundary condition can be neglected. For the geometric op- timization, a 7×10×1 Monkhorst-Pack[41] grid is used and all structures are relaxed until the forces on the atoms are less than 0.01 eV/Å, and the total energy change becomes less than 1.0 × 10−5 eV. For total energy calculations, a fine 35×50×1 grid is adopted. Since the PBE functional tends to underes- timate the band gap of semiconductors, the hybrid HSE06 functional[42] is also used to compute the band gap of op- timized TiS3 monolayer sheet. The carrier mobility (µ) is calculated based on the defor- mation theory proposed by Bardeen and Shockley.[43] Due to the fact that for inorganic semiconductors, the coherent wave- length of thermally activated electrons or holes is close to the acoustic phonon wavelength and is much longer than typi- cal bond length, the scattering of a thermal electron or hole is dominated by the electron-acoustic phonon coupling.[43] The deformation theory has been widely used to evaluate µ of 3 FIG. 3. (Color online)(a) Strain-total energy relations and (b) shifts of VBM and CBM under uniaxial strain along a and b directions for TiS3 monolayer sheet, δl refers to the dilation along a or b, while l0 refers to the lattice constant of a or b at equilibrium geometry. In (b), the vacuum level is set at zero for reference. TABLE I. Calculated deformation-potential constant (E1), 2D mod- ulus (C), effective mass (m∗), relaxation time (τ), and electron and hole mobility (µ) in a and b directions of TiS3 monolayer sheet at 300 K. E1 (eV) C (N/m) m∗ (me) τ (ps) µ (103 cm2V−1s−1) 0.73 3.05 0.94 -3.76 81.29 81.29 145.05 145.05 1.01 1.21 13.87 0.15 electron (a) hole (a) electron (b) hole (b) 1.47 0.32 0.41 0.98 0.84 0.22 3.23 0.085 As shown in TableI, the 2D modulus along b is nearly two times higher than that along a direction. This is because the Ti-S bond strength along a is weaker than that along b direc- tion as the Ti-S bond length is 2.65 Å along a while 2.45 Å along b. The difference between the Ti-S bond strength along a and b also makes the deformation-potential constant along b larger than that along a, as the band energies are more sensi- tive to dilations along b than along a. The effective mass also shows an anisotropic feature: in a direction, the effective mass of hole is much smaller than that of electron, while in b direc- tion, the effective mass of electron is about half of that of the hole. These results can be well explained by a charge-density FIG. 2. (Color online)(a) Computed HSE06 band structures of TiS3 monolayer sheet; Γ (0.0, 0.0, 0.0), Y (0.0, 0.5, 0.0), A (0.5, -0.5, 0.0), B (0.5, 0.0, 0.0) refer to the special points in the first Brillouin zone; red circles, green rectangles refer to the contributions of Ti d states and S p states; and the Fermi level is set to zero; (b) iso-surface plots of the charge density of VBM (left) and CBM (right) of the TiS3 monolayer sheet, with an iso-value of 0.003 e/Bohr3. the electrons (from CBM) favor the b direction. To compute the 2D elastic modulus (C) and the deformation-potential con- stant (E1), we dilate the lattice of the cell up to 1.5% along both a and b directions, and then calculate the total energy and the positions of CBM and VBM with respect to the dila- tion. The atomic positions are relaxed at the dilation, and the electronic energies are calculated at the PBE-D2 level with the ultra-fine k-meshes (35×50×1). We note that although the PBE functional underestimates the band gap, it can give quite good carrier mobility data for MoS2,[44] graphene,[46] graphyne[45] and graphdiyne[47]. The total energy-strain re- lation and the positions of CBM and VBM with respect to the strain are plotted in Fig.3. As shown in Fig.3(b), the re- sponse of CBM and VBM to the applied strain appears to be highly anisotropic. The CBM increases monotonously with the strain either along a or b direction, while the VBM de- creases monotonously with the strain along the b direction but increases along a, resulting in band gap increase due to the strain along b but band gap decrease along a. The 2D modu- lus (C) is attained by the quadratic fitting of the total energy versus strain, and the deformation potential constant (E1) is calculated by the linear fitting of the CBM (VBM)-strain re- lation. With C, E1 and the effective mass known, the carrier mobilities are calculated via equation (1). These data and the relaxation time (τ = µm∗/e) are summarized in Table I. plot of CBM and VBM as shown in Figure 2(b) where one can see that VBM electrons are quite localized along b while the CBM ones are delocalized along b but localized along a. 4 The predicted carrier mobilities of TiS3 monolayer are highly anisotropic. The computed electron mobility along b direction is 13.87×103 cm2V−1s−1, about 14 times higher than that along a direction (1.01×103 cm2V−1s−1), while the hole mobility along a direction is 1.21×103 cm2V−1s−1, about 8 times higher than that along b direction (0.15×103 cm2V−1s−1). the pre- dicted carrier mobilities are notably higher than those of the MoS2 monolayer sheet (which are in the range of 60-200 cm2V−1s−1)[44]. Especially, along the b direction, the elec- tron mobility is about 100 times higher than the hole mobility, making b direction more favorable for the electron conduc- tion. The large difference in electron/hole mobility can be exploited for electron/hole separation. is worthy of mentioning that It Although TiS3 monolayer exhibits some novel properties for potential nanoelectronic applications, feasibility of isola- tion of TiS3 monolayer sheet via either exfoliation or mechan- ical cleavage technique has yet to be confirmed. To example this feasibility, we calculate the cleavage energy by introduc- ing a fracture in the bulk TiS3 (see Fig.4(b)). To this end, the total energies under variation of the separation d between the fractured parts are computed to simulate the exfoliation process.[50, 51] The resulting cleavage energy is plotted in Fig.4 (a). It can be seen that the total energy increases with the separation d and gradually converges to the ideal cleav- age cohesion energy of about 0.20 J/m2. The latter value is less than the experimentally estimated cleavage energy for the graphite ( ∼0.36 J/m2)[52], indicating that the exfoliation of bulk TiS3 should be highly feasible experimentally. The sta- bility of TiS3 monolayer is another issue that should be ex- amined. First, we compute the phonon spectrum of the TiS3 monolayer, based on density functional perturbation theory with the linear response as implemented in the QUANTUM- ESPRESSO package.[53] As shown in Fig.4 (c), the TiS3 shows no imaginary phonon mode, indicating its dynamical stability. Next, we perform BOMD simulations. The constant- temperature (300 K) and -pressure (0 GPa) (NPT) ensemble is adopted. Here, the time step is 2 fs and the total simulation time is 8 ps. As shown in Fig.4 (d), the in-plane structure in- tegrity of TiS3 monolayer is well kept during the BOMD run, suggesting good thermal stability of the TiS3 monolayer. In conclusion, we predict a new 2D material, TiS3 mono- layer sheet, which is a semiconductor with a desired direct band gap of ∼1 eV. The electron mobilities of the TiS3 mono- layer are dominant and highly anisotropic. More specifically, the electron mobility along b direction exhibits a very high value of 13.87×103 cm2V−1s−1, rendering the TiS3 monolayer particularly attractive for future applications in nanoelectron- ics as its mobility is even notably higher than that of the MoS2 monolayer. The computed ideal cleavage cohesion energy for TiS3 is about 0.20 J/m2, less than that of graphite (∼0.36 J/m2), indicating the isolation of a 2D TiS3 monolayer can be technically attainable via either liquid exfoliation or mechan- FIG. 4. (Color online)(a) Cleavage energy Ecl as a function of the separation between two TiS3 monolayers. (b) Schematic view of the exfoliation of a TiS3 monolayer from bulk, (c) Phonon band structure and density of states of TiS3 monolayer and (d) Snapshot of TiS3 monolayer at 8 ps of the BOMD simulation in the NPT ensemble at 300 K and 0 GPa. ical cleavage as done for isolation of 2D graphene or MoS2 sheet.[30–32] Lastly, dynamic and thermal stability of TiS3 is confirmed by both phonon spectrum and BOMD simulations. Thus, we expect that fabrication of 2D TiS3 monolayer and measurement of its electronic properties will be likely accom- plished in the near future. This work was supported by the National Science Foun- dation (NSF) through the Nebraska Materials Research Sci- ence and Engineering Center (MRSEC) (grant No. DMR- 1420645)NSF, UNL Nebraska Center for Energy Sciences Research, and UNL Holland Computing Center. ∗ xzeng1@unl.edu [1] K. S. Novoselov, A. K. Geim, S. Morozov, D. Jiang, Y. Zhang, S. Dubonos, I. Grigorieva, and A. Firsov, Science 306, 666 (2004). [2] S. Z. Butler et al., ACS Nano 7, 2898 (2013). [3] Y. Zhang, Y. W. Tan, H. L. Stormer, and P. Kim, Nature 438, 201 (2005). [4] K. Novoselov, A. K. Geim, S. Morozov, D. Jiang, M. K. I. Grig- orieva, S. Dubonos, and A. Firsov, Nature 438, 197 (2005). [5] A. K. Geim and K. S. Novoselov, Nat. Mater. 6, 183 (2007). [6] A. C. Neto, F. Guinea, N. Peres, K. S. Novoselov, and A. K. Geim, Rev. Mod. Phys. 81, 109 (2009). [7] Q. H. Wang, K. Kalantar-Zadeh, A. Kis, J. N. Coleman, and M. S. Strano, Nat. Nanotech. 7, 699 (2012). [8] M. Chhowalla, H. S. Shin, G. Eda, L. J. Li, K. P. Loh, and H. Zhang, Nat. Chem. 5, 263 (2013). [9] H. Zeng, C. Zhi, Z. Zhang, X. Wei, X. Wang, W. Guo, Y. Bando, and D. Golberg, Nano. Lett. 10, 5049 (2010). [10] L. Song et al., Nano. Lett. 10, 3209 (2010). [11] L. Chen, C. C. Liu, B. Feng, X. He, P. Cheng, Z. Ding, S. Meng, Y. Yao, and K. Wu, Phys. Rev. Lett. 109, 056804 (2012). (2005). [12] B. Feng, Z. Ding, S. Meng, Y. Yao, X. He, P. Cheng, L. Chen, [33] P. L. Hsieh, C. Jackson, and G. Gruner, Solid State Commun. 5 46, 505 (1983). [34] E. Finkman and B. Fisher, Solid State Commun. 50, 25 (1984). [35] H. G. Grimmeiss, A. Rabenau, H. Hahn, and P. Ness, Zeitschrift fur Elektrochemie, 65, 776 (1961). [36] J. O. Island et al., Adv. Opt. Mater. 2, 641 (2014). [37] I. Ferrer, M. Maci´a, V. Carcel´en, J. Ares, and C. S´anchez, En- ergy Procedia 22, 48 (2012). [38] I. Ferrer, J. Ares, J. Clamagirand, M. Barawi, and C. S´anchez, Thin Solid Films 535, 398 (2013). [39] G. Kresse and J. Furthmuller, Phys. Rev. B 54, 11169 (1996). [40] S. Grimme, J. Antony, S. Ehrlich, and H. Krieg, J. Chem. Phys. [41] H. .J. Monkhorst and J. D. Pack, Phys. Rev. B 13, 5188 (1976). [42] J. Heyd, G. E. Scuseria, and M. Ernzerhof, J. Chem. Phys. 124, 132, 154104 (2010). 219906 (2006). [43] J. Bardeen and W. Shockley, Phys. Rev. 80, 72 (1950). [44] Y. Cai, G. Zhang, and Y. W. Zhang, J. Am. Chem. Soc. 136, [45] J. Chen, J. Xi, D. Wang, and Z. Shuai, J. Phys. Chem. Lett. 4, [46] J. Xi, M. Long, L. Tang, D. Wang, and Z. Shuai, Nanoscale 4, [47] M. Long, L. Tang, D. Wang, Y. Li, and Z. Shuai, ACS Nano 5, 6269 (2014). 1443 (2013). 4348 (2012). 2539 (2011). [48] M. Long, L. Tang, D. Wang, L. Wang, and Z. Shuai, J. Am. Chem. Soc. 131, 17728 (2009). [49] See Supplemental Material at [URL will be inserted by pub- lisher] for the PBE-D2 and HSE06 band structures of bulk TiS3. [50] X. Li, X. Wu, and J. Yang, J. Am. Chem. Soc. 136, 11065 and K. Wu, Phys. Rev. Lett. 12, 3507 (2012). [13] P. Vogt, P. De Padova, C. Quaresima, J. Avila, E. Frantzeskakis, M. C. Asensio, A. Resta, B. Ealet, and G. Le Lay, Phys. Rev. Lett. 108, 155501 (2012). [14] M. D´avila, L. Xian, S. Cahangirov, A. Rubio, and G. Le Lay, New J. Phys. 16, 095002 (2014). [15] L. Li, Y. Yu, G. J. Ye, Q. Ge, X. Ou, H. Wu, D. Feng, X. H. Chen, and Y. Zhang, Nat. Nanotech. 9, 372 (2014). [16] H. Liu, A. T. Neal, Z. Zhu, Z. Luo, X. Xu, D. Tom´anek, and P. D. Ye, ACS Nano 8, 4033 (2014). [17] B. Partoens and F. Peeters, Phys. Rev. B 74, 075404 (2006). [18] K. F. Mak, C. Lee, J. Hone, J. Shan, and T. F. Heinz, Phys. Rev. Lett. 105, 136805 (2010). [19] J. Qiao, X. Kong, Z. X. Hu, F. Yang, and W. Ji, Nat. Commun. [20] V. Tran, R. Soklaski, Y. Liang, and L. Yang, Phys. Rev. B 89, 5, 4475 (2014). 235319 (2014). [21] J. Dai and X. C. Zeng, J. Phys. Chem. Lett. 89, 1289 (2014). [22] H. Guo, N. Lu, J. Dai, X. Wu, and X. C. Zeng, J. Phys. Chem. C 118, 14051 (2014). [23] L. Liao et al. Nature 467, 305 (2010). [24] F. Schwierz, Nat. Nanotech. 5, 487 (2010). [25] Y. Wu, Y. M. Lin, A. A. Bol, K. A. Jenkins, F. Xia, D. B. Farmer, Y. Zhu, and P. Avouris, Nature 472, 74 (2011). [26] B. Radisavljevic, A. Radenovic, J. Brivio, V. Giacometti, and A. Kis, Nat. Nanotech. 6, 147 (2011). [27] Y. Yoon, K. Ganapathi, and S. Salahuddin, Nano. Lett. 11, 3768 [28] S. Furuseth, L. Brattas, and A. Kjekshus, Acta Chem. Scand. [29] L. Brattas and A. Kjekshus, Acta. Chem. Scand. 26, 3441 29, 623 (1975). (2011). (1972). [30] J. N. Coleman et al., Science 331, 568 (2011). [31] V. Nicolosi, M. Chhowalla, M. G. Kanatzidis, M. S. Strano, and J. N. Coleman, Science 340, 1226419 (2013). [32] K. Novoselov, D. Jiang, F. Schedin, T. Booth, V. Khotkevich, S. Morozov, and A. Geim, Proc. Natl. Acad. Sci. 102, 10451 [51] B. Sachs, T. Wehling, K. Novoselov, A. Lichtenstein, and M. Katsnelson, Phys. Rev. B 88, 201402 (2013). [52] R. Zacharia, H. Ulbricht, and T. Hertel, Phys. Rev. B 69, 155406 [53] P. Giannozzi et al., J. Phys.: Condens. Matter. 21, 395502 (2014). (2004). (2009). 

1902.09530

2

1902

Quantum Electronic Circuit Simulation of Generalized sine-Gordon Models

Investigation of strongly interacting, nonlinear quantum field theories (QFT-s) remains one of the outstanding challenges of modern physics. Here, we describe analog quantum simulators for nonlinear QFT-s using mesoscopic superconducting circuit lattices. Using the Josephson effect as the source of nonlinear interaction, we investigate generalizations of the quantum sine-Gordon model. In particular, we consider a two-field generalization, the double sine-Gordon model. In contrast to the sine-Gordon model, this model can be purely quantum integrable, when it does not admit a semi-classical description - a property that is generic to many multi-field QFT-s. The primary goal of this work is to investigate different thermodynamic properties of the double sine-Gordon model and propose experiments that can capture its subtle quantum integrability. First, we analytically compute the mass-spectrum and the ground state energy in the presence of an external `magnetic' field using Bethe ansatz and conformal perturbation theory. Second, we calculate the thermodynamic Bethe ansatz equations for the model and analyze its finite temperature properties. Third, we propose experiments to verify the theoretical predictions.

cond-mat.mes-hall

cond-mat

Quantum Electronic Circuit Simulation of Generalized sine-Gordon Models Ananda Roy1, 2, ∗ and Hubert Saleur1, 3 1Institut de Physique Th´eorique, Paris Saclay University, CEA, CNRS, F-91191 Gif-sur-Yvette, France 2Department of Physics, T42, Technische Universitat Munchen, 85748 Garching, Germany 3Department of Physics and Astronomy, University of Southern California, Los Angeles, CA, USA In particular, we consider a two-field generalization, the double sine-Gordon model. Investigation of strongly interacting, nonlinear quantum field theories (QFT-s) remains one of the outstanding challenges of modern physics. Here, we describe analog quantum simulators for nonlinear QFT-s using mesoscopic superconducting circuit lattices. Using the Josephson effect as the source of nonlinear interaction, we investigate generalizations of the quantum sine-Gordon model. In contrast to the sine-Gordon model, this model can be purely quantum integrable, when it does not admit a semi-classical description - a property that is generic to many multi-field QFT-s. The primary goal of this work is to investigate different thermodynamic properties of the double sine- Gordon model and propose experiments that can capture its subtle quantum integrability. First, we analytically compute the mass-spectrum and the ground state energy in the presence of an external 'magnetic' field using Bethe ansatz and conformal perturbation theory. Second, we calculate the thermodynamic Bethe ansatz equations for the model and analyze its finite temperature properties. Third, we propose experiments to verify the theoretical predictions. I. INTRODUCTION The longstanding goal of quantum field theory (QFT) is to predict the masses of the excitations and their scat- tering cross-sections in terms of the parameters of the theory and to characterize the different phases and the phase-transition points. While remarkable progress has been achieved in analyzing strongly coupled QFT-s using numerical or effective field theory methods, many quan- tities of interest remain elusive, either due to the com- putational limitations or due to the lack of an effective field theory. Quantum simulation, both analog and digi- tal, takes a different approach to solving these aforemen- tioned problems, where one quantum system is tailored to simulate another in a controlled manner1 -- 9. Here, we describe analog quantum simulators for some QFT-s with superconducting quantum electronic circuit (QEC) lattices10,11. Specifically, we are interested in ob- taining bosonic QFT-s directly, and not as consequences of mathematical manipulations (bosonization) of an un- derlying fermionic or spin system12 -- 17. Even in 1 + 1 space-time dimensions, this is crucial to distill the true behavior of a bosonic system from a bosonized fermionic system18,19, despite the well-known fermion-boson corre- spondences20,21. Of course, this distinction is even more important in higher dimensions. Attempts in this direction have been considered pre- viously in the well-established Bose-Hubbard model paradigm22,23. Another scheme is to use the Josephson effect to give rise naturally to the cosine potential24 of the sine-Gordon (SG) model (see more below). Our con- struction generalizes the latter strategy. The role of the bosonic field at a point in space-time is played by the time-integral of the voltage at that point25 (so the under- lying degrees of freedom of the QEC lattices are faithful directly to the bosonic description). The building blocks of the QEC lattices are superconducting self and mutual inductances, capacitances and Josephson junctions. The current-voltage constitutive relations of these elements, together with Kirchhoff's laws for the circuits, give rise to the nonlinear field equations of the QFT-s. The playground for bosonic nonlinear QFT-s is wide, even in 1 + 1 dimensions, and includes typically cousins of the SG theory involving multiple component fields, together with some cosine interaction. The case of two bosons already escapes our complete understanding. In contrast with the SG case, the physics of the various "double sine-Gordon models" (see definitions below) can- not always be inferred from some classical limit. For in- stance, fixed points are known to exist, which are truly quantum in nature and do not admit a semi-classical description26. In fact, one of the simplest new aspects to formulate and study in multi-component bosonic theories is integrability. Integrability leads to a factorized multi- particle scattering matrix27,28. This, in turn, allows ana- lytic computation of thermodynamic and transport prop- erties of these QFT-s29,30. The ability to do such compu- tations is crucial to understand non-perturbative aspects of strongly interacting systems - for instance, charge frac- tionalization31 -- 33 and its effect on full-counting statis- tics34, both topics of high current interest. It is well known that the bulk and boundary SG models are classically integrable35,36. This was largely used his- torically to establish their quantum integrablity as well 28,37,38. While the naive intuition that quantum fluctu- ations might well destroy classical integrability - as is known to be the case, for instance, for many sigma mod- els39- is usually correct, it is now understood that the SG model remains integrable in the quantum regime due to the existence of subtle quantum group symmetries40. The presence of these symmetries in 1 + 1 dimensions re- lies heavily on the existence of conformally invariant UV fixed points, of which integrable QFTs can be considered as deformations41 -- 45. The simplest generalization of the SG model in this regard is probably the double sine-Gordon (dSG) model, 9 1 0 2 t c O 4 1 ] l l a h - s e m . t a m - d n o c [ 2 v 0 3 5 9 0 . 2 0 9 1 : v i X r a (cid:90) whose euclidean action is SdSG = d2x (∂µφi)2 + (cid:88) i=1,2 1 2 (cid:89) i=1,2 2M0 π cos(αiφi), (1) where φ1, φ2 are two bosonic fields, M0 the interaction strength and α1, α2 the coupling constants. This model and its close cousins involving a boundary interaction ap- pear in a variety of guises in condensed matter physics - for instance, in transport experiments involving one- dimensional particles with charge and spin, or in quan- tum Brownian motion on two dimensional lattices46. Un- fortunately, little remains known about the possibility to solve the dSG model exactly, and to infer from such a solution properties of physical interest. This is because, remarkably, as soon as more than one bosonic degree of freedom is involved, quantum and classical integrability often part ways. In fact, it is known that the dSG model is classically integrable only for α1 = α2, when it reduces to two decoupled SG models47. However, the same fluc- tuations that can destroy classical integrability in some cases can also give rise to quantum integrability. Purely quantum integrable manifolds, if they exist, should thus arise for values of the coupling constants that cannot ap- proach the origin in the (α1, α2) plane. The dSG model is probably only quantum integrable on several (α1, α2) manifolds15,18,48 -- 51. While some doubts remain on the exact nature of this statement, it is strongly believed in 2 = 4π/ indeed the community that the manifold: α2 is quantum integrable (in the following we will set  = 1). This is, of course, a remarkable statement: we are facing a situation where a classical soliton wave-packet, which solves the classical field equations, gets scrambled, but its quantized counterpart, when the couplings reach some magical values, propagates undistorted and scatters only with phase-shifts! One of our goals is to propose a set- up where such behavior - which, we believe, is the norm rather than the exception in many multi-field theories - might be observed experimentally. 1 +α2 It is important to stress that the calculations presented in this paper are done exactly for the model in Eq. (1). It should be contrasted with those done by extrapolating the results of the more general Fateev model48, of which the dSG is a special case. However, the Fateev model involves, on top of the two (compact) bosonic fields φi, an extra non-compact bosonic field. The presence of this other field makes many calculations of the Fateev model - in particular, the thermodynamic Bethe ansatz (TBA) - considerably easier. It is expected that the final results can then be continued to the limit where this extra boson decouples, but this assumption is dangerous in view of the singularity of this limit. II. THE DOUBLE SINE-GORDON MODEL AT ZERO TEMPERATURE The solitons of the quantum dSG model cannot be ap- proached with semiclassical methods, and are not fully 2 understood. It is known that they are topological ex- citations of the fields ϕ1, ϕ2, where ϕ1,2 = (α1φ1 ± √ α2φ2)/2 π. They carry a pair of quantum numbers cor- responding to the fields φ1, φ2, just as an electron carries electric charge and spin quantum numbers. These soli- tons scatter with a factorized scattering matrix. More remarkably, because of the underlying (pair of) quan- tum group symmetries49, the two quantum numbers of each soliton scatter independently of each other. The factorized scattering matrix is expected to be given by48 S = Sp1 ⊗ Sp2, where Spi, i = 1, 2 is the SG scatter- ing matrix. The corresponding SG couplings βi-s are re- lated to pi-s by β2 i /8π = pi/(pi + 1), i = 1, 228,45, where p1,2 = α2 1,2/2π48. √ (cid:82) dx∂xϕ1,2/ The masses of the physical excitations in terms of the parameters of the action, together with a verifica- tion of the scattering matrix, are obtained by calculat- ing the ground state energy in the presence of 'mag- netic fields' h1,2 coupling to the conserved charges Q1,2 = π. Consider the limit h1,2 (cid:29) M0 and tune h2 such that (cid:104)Q2(cid:105) = 0. Standard Bethe ansatz calcu- lations using Wiener-Hopf technique48,52 show that, on the integrable manifold, the ground state energy gets a logarithmic correction in h1 (to be contrasted with the purely polynomial corrections of the Fateev model48), in addition to its usual quadratic dependence: E(h1) = − h2 1p1(2 − p1) 2π + m2 s 2π sin2(cid:16) πp1 (cid:17) 2 ln h1. (2) Here ms is the mass of the ϕ1 soliton53. It is obtained in terms of the action parameters using conformal perturba- tion theory (CPT)54. We find that ms = M0/ sin(πp1/2), which matches the result of Ref. 48. The calculation is is described in detail in Appendix A. III. THE DOUBLE SINE-GORDON MODEL AT NONZERO TEMPERATURE The thermodynamics of QFT-s in an infinite volume can be inferred from its scattering matrix55 using the TBA technique29. Using the factorized scattering ma- trix given above, we now compute the free energy of the dSG model at a finite temperature 1/R. The TBA given below is essential to correctly predict the dSG free en- ergy and cannot be directly inferred from the TBA of the Fateev model (the free energy extrapolated from the Fateev model will appear to be infinite). Without loss of generality, we choose p1 < 1, i.e., the first SG sec- tor to be attractive, and p2 > 1, i.e., the second, re- pulsive. In addition to the solitons, the spectrum of the theory also includes n < 1/p1 particles which are neu- tral with respect to first quantum number, but still carry the second quantum number. They are the bound states of the fundamental 'double' solitons and correspond to the ordinary breather poles of the attractive SG scatter- ing amplitude of the first sector28. The TBA analysis is, in general, highly non-trivial, the scattering in each 3 FIG. 2. (a) - (d): Essential primitives of QEC lattices [(a) an inductance, L, (b) a capacitance, C, (c) a mutual induc- tance, M , and (d) a Josephson junction (junction energy EJ and junction capacitance CJ )]. The magnetic fluxes going through the inductance (L), the mutual inductance (M ) and the Josephson junction are denoted by φ, φ1, φ2 and φ respec- tively. The charges on the capacitor plates of the capacitances C and CJ are denoted by Q. (e) One unit cell of the QEC for the dSG model. The top and bottom horizontal lines denote the electrical ground. The Josephson junctions (junction en- ergy EJ,0 and junction capacitance C0) on the vertical links give rise to the nonlinear interaction. The Josephson junc- tions (junction energy EJ and junction capacitance CJ ) give rise to a large impedance of the array, without any nonlinear effects. Finally, the mutual inductance M and the capacitance C provide the necessary inductive and capacitive coupling be- tween the upper and lower parts of the array. The bosonic fields, ϕ1, ϕ2, are the node-fluxes at the points indicated. viations from the conformal limit, in addition to the poly- nomial corrections in powers of (msR)254. We calculate these corrections by numerically solving the TBA equa- tions. The logarithmic correction for the different values of n is then verified using CPT. We find the logarithmic correction to ceff to be sin2(cid:104) (cid:105) (∆ceff )log = 6(msR)2 π2 π 2(n + 1) ln(msR). (4) More details on the TBA and the explicit forms of the dif- ferent polynomial corrections to ceff for n = 2, 3 are given in Appendix B. Note that this logarithmic correction is very similar to those obtained for the sine-Gordon model for suitably chosen coupling constants 61,62 and should be contrasted with a very different type of logarithmic correction that arises in the Fateev or sinh-Gordon mod- els48,63. IV. THE DOUBLE SINE-GORDON MODEL WITH QUANTUM CIRCUITS The predictions for the various thermodynamic quan- tities computed above can, in principle, be measured in an experimental setup. In particular, experiments should be able to capture the subtle quantum vs classical inte- grability of the dSG model. To that end, we provide a QEC for the dSG model. In terms of the rotated fields FIG. 1. TBA diagram for the dSG model. The solid circles de- note the physical massive particles: soliton (s), antisoliton(a) and breathers (bj, j = 1, . . . , n). The empty circles denote pseudoparticles (j = 1, . . . , n + 2) needed to diagonalize the dSG scattering matrix. The pseudoparticle 1 has a cross on it, which indicates that this particle has a mass term in its TBA equation. The connectivity of the diagram encodes which par- ticles show up in the TBA equation of a given particle. The arrows on the links encode the sign of the term on the right hand side (RHS) of the TBA equation. For instance, for a link connecting particles p, q, if there is an arrow incident on q and none on p, then the RHS of the TBA equation for p has a minus sign in front of the term involving q, while the term on the RHS for q involving p has a plus sign. sector being non-diagonal. We consider the simpler case p1 = 1/(n + 1) when the first sector scattering is diago- nal. The diagonalization problem of the second sector is done by the Takahashi-Suzuki classification of the solu- tion of the Bethe equations56 -- 58 and the algebraic Bethe ansatz59,60. The detailed derivation of the TBA equa- tions is given in Appendix B. The resulting TBA equa- tions have an universal form for all n:
γ(θ) = ϕn (cid:63) Lbn , γ = s, a,
bn (θ) = ϕn (cid:63) (Ls + La + Lbn−1 ),
bj (θ) = ϕn (cid:63) (Lbj+1 + Lbj−1), j = n − 1, . . . , 3,
b2(θ) = ϕn (cid:63) (Lb3 + Lb1 − L1),
b1(θ) = ϕn (cid:63) (Lb2 − L2),
1(θ) = −mb1 R cosh θ + ϕn (cid:63) (Lb2 − L2),
2(θ) = − ϕn (cid:63) (Lb1 − L1 + L3),
j(θ) = − ϕn (cid:63) (Lj+1 + Lj−1), j = n − 1, . . . , 3,
n(θ) = − ϕn (cid:63) (Ln−1 + Ln+1 + Ln+2),
k(θ) = − ϕn (cid:63) Ln, k = n + 1, n + 2. (3) The TBA kernel is ϕn(θ) = (n + 1)/ cosh[(n + 1)θ], while L(θ) = ln[1 + e
(θ)] and L(θ) = ln[1 + e−
(θ)]. Finally, mb1 = 2ms sin πp1/2 is the mass of the first breather28 and a (cid:63) b =(cid:82) a(θ − θ(cid:48))b(θ(cid:48))dθ(cid:48)/2π. The TBA diagram is given in Fig. 3. We check our TBA equations by analytic computation of the central charge in the conformal (UV) limit. This yields the expected result of cUV = 2 for the two bosonic fields (see Appendix B for details of the computation). Next, we compute the free energy of the system upon de- viation from the conformal limit. This is given in terms of the effective central charge, ceff = −6R2f /π, where f is the free energy per unit length. On the integrable man- ifold, the perturbation has dimension (α2 2)/(4π) = 1. 61,62 upon de- This leads to logarithmic corrections to ceff 1 +α2 ϕ1,2, the dSG action can be viewed as two SG actions 1 − π/α2 2. Recall that in QEC-s, the bosonic field is time-integral of the voltage at a point. Thus, the first coupling term, ∂tϕ1∂tϕ2, is a voltage-voltage coupling, realized in QEC- s with a capacitance (C). The second coupling term, ∂xϕ1∂xϕ2, is a current-current coupling, realized by a mutual inductance (M ). All that is left is to realize two separate identical SG models. The QEC-s for the two SG models are two one-dimensional arrays [see panel (e) of Fig. 2] with Josephson junctions on horizon- tal links (junction energy EJ and junction capacitance CJ ) together with Josephson junctions on vertical links (junction energy EJ,0 and junction capacitance C0). We work in the regime when EJ (cid:29) ECJ and EC0 (cid:29) ECJ , where ECJ = (2e)2/2CJ and EC0 = (2e)2/2C0. The array impedance is Z = RQ RQ((cid:39) 6.5 kΩ) is the resistance quantum (in units of Cooper pairs). Only the kinetic inductance of the Joseph- son junctions on the horizontal links are used to increase the impedance of the array, with the phase-slip ampli- EJ /ECJ 68. The Lut- tinger parameter is K = 2Z/RQ. Capacitively and in- ductively coupling these two SG models results in the QEC for the dSG model. Choosing M = C, the cor- respondences between the circuit components and the dSG action are: α2 2 = π/(2/K + 4πC) and M0 = πEJ,0. A more detailed analysis of the circuit is given in Appendix C. (cid:112)2EC0/EJ /2π64 -- 67, where tudes are exponentially small ∼ e−√ coupled by B(cid:82) d2x∂µϕ1∂µϕ2, where B = π/α2 1 = πK/2, α2 Next, we describe various experimental measurements of the dSG model possible in a QEC setup. Estima- tion of α1, α2 can be done by biasing with two current sources (the 'magnetic fields' h1, h2 considered earlier) and measuring the transmission of an input field through the dSG lattice. This provides an estimate of the average current, (cid:104)∂xϕ1,2(cid:105), flowing through the horizontal links in the circuit. Comparison to CPT calculations, which give (cid:104)∂xϕ1,2(cid:105) = [(α1 ± α2)h1 + (α1 ∓ α2)h2]/4π3/2 to leading order, yields estimates of α1, α2. Classical integrability can be captured by scattering measurements of a clas- sical sine-Gordon soliton wavepacket36 for the either ϕ1 or ϕ2 propagating through the array. Due to the pres- ence of the mutual inductance and capacitance coupling to the upper and lower part of the array (Fig. 2), the wavepacket will distort. However, on the quantum in- tegrable manifold, a quantum soliton would propagate undistorted. A signature of the factorized scattering of the quantum soliton, can be obtained by measuring the specific heat69, which is predicted from the free energy computed above (recall that quantum integrability and the factorized scattering of the quantum solitons is cen- tral to the entire TBA computation). This measurement also provides the mass of the quantum soliton. More de- tails on the specific heat measurement proposal is given in Appendix C. Additional transport signatures of quan- tum integrability can be obtained from the boundary dSG model, which is realized by terminating the lattice shown in panel (e) of Fig. 2 with two impurity junctions, 4 one each for the upper and lower parts of the array. The quantum transport properties can be calculated along the lines of Refs. 15 and 49 starting with our TBA. However, inclusion of the boundary in the TBA involves additional technical complications. We intend to report on this in a future publication. V. SUMMARY To summarize, we have provided analytical predic- tions of the thermodynamic properties of the dSG model and a QEC for simulating it. Faithful analog simula- tion of QFT-s using QEC-s open new possibilities in in- vestigation of their non-perturbative properties. Radio- frequency measurements possible with QEC-s allow a more detailed verification of the QFT-s' properties. At the same time, QEC simulation of integrable QFT-s may be viewed as a method to benchmark these analog sim- ulators by comparing experiments to analytical theoret- ical predictions. This leads to a systematic investigation of more general QFT-s, by systematically including per- turbations which break integrability. Using the robust controllability and customizability of QEC-s, this can be achieved as described below. A massless free-boson QFT can be included by taking the SG circuit after remov- ing the Josephson junctions on the vertical links [upper or lower half of panel (e) of Fig. 2]. A massive free- boson QFT is obtained from the latter by adding linear inductances to the ground at each vertical link. Inter- actions can be realized by either inductive coupling or capacitive coupling or through Josephson junctions. For instance, interacting QFT-s may be obtained by specific arrangements of Josephson junctions in tailored geome- tries70 -- 72. Fermionic modes may potentially also be in- cluded using the topological Josephson effect73. Finally, the QEC-s proposed in this work can be generalized to simulate higher dimensional QFT-s (integrable and non- integrable)74. ACKNOWLEDGMENTS A.R. acknowledges fruitful discussions with Jonathan Belletete, Michel Devoret, Hermann Grabert, Michal Pawelkiewicz, and A. Douglas Stone. A.R. acknowl- edges the support of the Alexander von Humboldt foun- dation. H.S. was supported by the ERC Advanced Grant NuQFT. H.S. thanks F. Lesage and P. Simonetti for (early) collaboration on this project, and T. Albash for discussions. Appendix A: The double sine-Gordon model at zero temperature In this section we consider the dSG model in the pres- ence of external magnetic fields hi-s, which couple to the (cid:90) ∞ (cid:90) conserved charges Qi-s of the solitons. First, we calcu- late the ground state energy in the limit hi → ∞ and then consider deviations from this limit using CPT. Sub- sequently, the results are compared with Bethe ansatz calculations. The charges, Qi-s, of the solitons are given by (cid:90) dx Qi = 1√ π where ϕ1,2 = (α1φ1 ± α2φ2)/2 magnetic field is given by the following action ∂1ϕi, i = 1, 2, √ −∞ π. The coupling to the (A1) Smag = − h1√ π d2x∂1ϕ1 − h2√ π d2x∂1ϕ2. (A2) In limit h1/M0, h2/M0 → ∞, the interaction term Sint can be neglected. The ground state energy can be com- puted by completing squares in a Gaussian integral and is given by E = − (h2 1 + h2 2)A − 2h1h2B 2π(A2 − B2) . (A3) The equilibrium values of the charges Q1,2 = ∂E/∂h1,2 are given by Q1 = h1A − h2B π(A2 − B2) , Q2 = h2A − h1B π(A2 − B2) . (A4) We will be considering the situation when the fields h1, h2 are so chosen that Q2 = 0 which implies, from the above equation, h2 = Bh1/A. Then, the charge Q1 is given by Q1 = h1 πA = h1α2 π2(α2 1α2 2 1 + α2 2) (A5) and the ground state energy is 1α2 Ec = − h2 2π2(α2 1α2 2 1 + α2 2) . 5 Next, we look at finite h1/M0, h2/M0 and consider fluc- tuations of the fields ϕ1,2 around their asymptotic values given by Eq. [A4]. This is where our results differ from those obtained in Ref. 48. As will be shown below, the corrections to the limit h1/M0, h2/M0 → ∞ occur at the second order and lead to logarithmic corrections, while the same for the Fateev model occur at fourth order and give rise to polynomial corrections. We define √ ϕi(x) = πQix1 + ϕi(x), i = 1, 2, (A7) where (cid:104) ϕi(cid:105) = 0. The original fields φ1,2 can be written has φ1,2 = π α1 (Q1 ± Q2)x1 + φ1,2, (A8) π( ϕ1± ϕ2)/α1. Rewriting where we have defined φ1,2 = the action in terms of φ1,2 and considering the case when Q2 = 0, we have S0 + Smag = S0 + LREc, where √ d2x(cid:8)(∂µ (cid:90) S0 = 1α2 Ec = − h2 2π2(α2 1α2 2 1 + α2 2) 1 2 φ2)2(cid:9), φ1)2 + (∂µ . (A9) Here, L is the spatial extent of the 1D system and R is the inverse temperature. In the following, we will keep the fluctuating fields φ1,2 and evaluate the corrections due to the interaction term Sint perturbatively. We consider the case when Q2 = 0. The interaction term can be written as Sint = φ1) cos(πQ1x1+α2 d2x cos(πQ1x1+α1 φ2). (cid:90) 2M0 π This leads to ln Z = −EcLR + ln Z0 + (A10) where we have used the fact that (cid:104)Sint(cid:105)0 = 0, Z0 = (cid:82) D φ1D φ2exp(− S0) and the averages (cid:104) (cid:105)0 are with re- int(cid:105)0 (cid:104)S 2 1 2 spect to Z0. In order to calculate the correction term, we need the following formula for averages of the vertex operators75 (A6) (cid:81)n (cid:81)n i>j(xi − xjy1 − yj) α2 i,j=1 xi − yj α2 2π 2π , (A11) (cid:104)eiαφ(x1) . . . eiαφ(xn)e−iαφ(y1) . . . e−iαφ(yn)(cid:105)0 = where the average is taken with respect to the free-field action. Performing the integral, on the integrable mani- fold: α2 int(cid:105)0 = (cid:104)S 2 where γE is the Euler constant and a is a lattice cut- (cid:110) − 2γE − 2 ln(2πQ1) + 2 ln 1 + α2 2 = 4π, we get M 2 0 2π (cid:16) R (cid:17)(cid:111) LR 2a , off. Using this result, the specific ground state energy, up to second order in M0, is given by E = − 1 L ∂ ln Z ∂R = Ec + M 2 0 2π ln(2πQ1) + E0, (A12) where E0 = − 1 L ∂ ln Z0 ∂R + M 2 0 2π (cid:8)γE + 1 + ln(R/2a)(cid:9) where (A13) Λ = − ln 2 + 2 − p1 2 ln(2 − p1) + p1 2 ln p1. (A21) 6 Next, we calculate the ground state energy using Bethe ansatz. Consider the case when the fields are chosen so that Q2 = 0. Due to the presence of the magnetic field, the solitons of the ϕ1 fields acquire an additional energy −h1Q1 and as long as h1 is larger than the mass of the ϕ1 solitons, these solitons fill up all possible states within a 'Fermi interval' −B < θ < B, where θ is the rapidity parameter and B is the 'Fermi wavevector'. The specific ground state energy is then given by59,76,77 (cid:90) B −B Ec = − ms 2π dθ cosh(θ)
(θ), (A14) (cid:90) B where ms is the mass of the ϕ1 soliton and the quasipar- ticle energies
(θ) satisfy dθ Kc(θ − θ(cid:48))
(θ(cid:48)) = h1 − ms cosh(θ), with the boundary condition
(±B) = 0. Here, −B (A15) Then, the Bethe ansatz integral equation is reduced to the following linear integral equation for the v(ω), given by v(k) = − ih1N (0) k + imseB 2 N (−i) k − i (cid:90) dω 2πi e2iωB w + k α(ω)v(ω), (A22) + C+ where ms is the mass of physical excitations of solitons of ϕ1, α(ω) = N (ω)/N (−ω) and (cid:90) ih1N (0) = − imseB N (−i) dω 2πi + 2 C+ e2iωBα(ω)v(ω). (A23) Here, C+ includes all the poles on the positive imaginary axis, but not the one at ω = −k. Then, the ground state energy is given by Kc(θ) = δ(θ) − 1 2π d dθ [δp1 (θ) + δp2 (θ)], (A16) E(h1) − E(0) = − h2 1N (0)2 (cid:16) πω (cid:17) . 2 where δpi-s are soliton-soliton scattering phase-shifts (for instance, see Ref. 48 for explicit forms). The Fourier transform of Kc is given by Kc(ω) = sinh[πω(p1 + p2)/2] 2 sinh(πωp1/2) sinh(πωp2/2) tanh (A17) The leading order energy contribution in the limit of h1/M0, h2/M0 → ∞ can be obtained by the standard manipulations48,52 and the result is identical to that of the Fateev model. This leads to Ec = h2 1 2πKc(0) = − h2 1p1p2 π(p1 + p2) . (A18) Comparing Eqs. [A6, A18], we get pi = α2 i 2π , i = 1, 2. (A19) In terms of pi-s, the integration manifold is then given by p1 + p2 = 2. Next, we consider deviations from the h1/M0, h2/M0 → ∞ limit. This is done by a Wiener- Hopf calculation. Since we are interested in the dSG model on the integration manifold, we set p2 = 2 − p1 at the outset. Then, the scattering kernel reduces to Kc(ω) = 2 sinh( πωp1 2 sinh(πω) ) sinh(πω − πωp1 2 tanh ) (A20) The kernel factorizes as Kc(ω) = 1/N (ω)N (−ω)48, where (cid:16) πω (cid:17) . 2 N (ω) = 4π p1(2 − p1) p(iω)p( iω 2 )p(iω − iωp1 2 )eiωΛ 2 )p( 1 p( iωp1 2 + iω 2 ) , (cid:115) (cid:90) 2π − h1N (0) π dω 2πi e2iωB ω − i α(ω)v(ω), C(cid:48) + where C(cid:48) + is the contour that includes the poles included by C+ plus the one at ω = i. Our goal is to calculate this ground state energy, which gets contribution from the poles of α(ω), v(ω) and ω = i. We include these contributions iteratively. Keeping to the lowest order contribution, we evaluate the contribution due to the pole at ω = ω1 = i. This leads to 1N (0)2 E(h1) − E(0) = − h2 h2 1N (0)2 2π + (cid:90) dω 2πi e2iωB ω(ω − i)3 α(ω), π C(cid:48) + where α(ω) = (ω − i)α(ω). Application of the residue theorem leads to E(h1) − E(0) = − h2 − h2 (cid:110) − 1 1N (0)2 1N (0)2 2π s sin2(πp1/2) m2 1(2 − p1)p1 2h2 p1 − 2 + 2 ln sin − 2 ln πp1 2 4 2π +2γE + ln 4 + (cid:16) p1 πp1 2 − 1 (cid:17)(cid:111) , +π cot +2ψ h1(2 − p1)p1 ms (cid:112)(2 − p1)p1 Then, the ground state energy is given by the digamma function and N (0) = where ψ is 2 (A24) E(h1) − E(0) = − h2 1p1(2 − p1) − m2 s sin2(πp1/2) 4π πp1 2 + 2ψ 2 2π πp1 2 +π cot + 2 ln sin (cid:110) − 1 + 2γE + ln 4 + (cid:16) p1 (cid:17)(cid:111) − 1 . 4 p1 − 2 − 2 ln h1(2 − p1)p1 ms 7 (A25) The contributions to the ground state energy can be iden- tified as follows. The first term on the right hand side is Ec, the leading order contribution to the ground state energy when h1/M0, h2/M0 → ∞. In the second term, the h1-independent term can be identified as the ground state energy E(0), while the logarithmic term dependent on h1 is the perturbative correction as we move away from the infinitely large magnetic fields limit. Thus, we get E(h1) = Ec + δE, (A26) where δE = m2 s 2π sin2(πp1/2) ln h1 (A27) and the ground state energy in the absence of magnetic field as 4 p1 − 2 (A28) E(0) = m2 s sin2(πp1/2) 4π (2 − p1)p1 −2 ln +2 ln sin + 2ψ ms πp1 2 (cid:110) − 1 + 2γE + ln 4 + (cid:16) p1 (cid:17)(cid:111) πp1 2 − 1 + π cot 2 Furthermore, comparing the coefficient of ln h1 term in Eqs. (A12, A27) leads to an exact expression connect- ing the interaction parameter M0 of the action with the masses of the ϕ1 solitons: M0 = ms sin πp1 2 . (A29) Here, we have used the definition of Q1 in Eq. [A5] and the relation Eq. [A19]. We note that the masses of the ϕ2 solitons are the same as that of the ϕ1 solitons on the integrability manifold. (cid:89) (cid:89) (cid:89) (cid:89) δ=s,a pδ (θ − θkδ ) Sp1 kδ biδ(θ − θkδ ) Sp1 (cid:89) (cid:89) (cid:89) (cid:89) kbi i=1,...,n Sp1 pbi Sp1 bibj Appendix B: The double sine-Gordon model at nonzero-temperature In this section, we analyze the dSG model at finite temperature. The solitons of this model scatter in a fac- torized manner. In particular, the scattering of the two quantum numbers of each soliton also occurs indepen- dent of each other: S = Sp1 ⊗ Sp2, where on the in- tegrable manifold p1 + p2 = 2. This relation between p1, p2 forces one of the amplitude to be in the attrac- tive regime and the other to be in the repulsive regime. This is precisely why the analysis is different from the Fateev model, for which the TBA analysis was done only for the case p1, p2 ≥ 148. Without loss of generality, we choose p1 < 1, i.e. attractive, and p2 > 1 repulsive. In addition to the solitons, the spectrum of the theory also includes n < 1/p1 particles which are neutral with respect to first quantum number, but still carry the sec- ond quantum number. They are the bound states of the fundamental 'double' solitons and correspond to the ordi- nary breather poles of the attractive sine-Gordon ampli- tude of the first sector. The masses of these breathers are given by the standard formula mi = 2ms sin[πi/2(n+1)], where i = 1, . . . , n28. The TBA analysis of this problem is in general highly non-trivial, being each scattering sec- tor source of non diagonally. We will restrict therefore ourselves to the simpler case p1 = 1/(n + 1) for which the sector one scattering is diagonal. Our TBA analysis will therefore consists in the diagonalization problem of the second sector only, which can be done by means of the Takahashi-Suzuki classification of the solution of the Bethe equations56 -- 58. The periodicity condition for the wave function of this given number of solitons, antisolitons and breathers on a circle of length L becomes the following set of equations: (θ − θkbi (θ − θkbj )λp2 p (θ) = e−impL sinh θ, p = s, a, (θ) = e−imbi L sinh θ, )λp2 bi (B1) δ=s,a kδ j=1,...,n kbj i = 1, . . . , n. In this equation, θ is the rapid- where ity of the particle going around the circle, θks , θka , θkbi are the rapidities of the incoming soliton, antisoliton and the breathers respectively. The scattering coefficients for the sine-Gordon model, Sp1 -s, are well-known28, while λp2 is the contribution of the second phase-shift due to the scattering of the second quantum number. This con- tribution is computed by diagonalization of the transfer matrix of a second sector soliton going around the world repulsively interacting with solitons and antisolitons with amplitudes Sp2. The diagonalization procedure for repul- sive sine-Gordon produces in the thermodynamic limit equations for the densities of states in terms of massless pseudoparticles and a physical particle carrying the mass of the soliton. When 'gluing' this result with the above equations, the only care will be needed in the identifica- tion of the role of the massive physical particles. This is done using the algebraic Bethe ansatz technique57,59,60. The calculation is nontrivial. First, we do it for n = 2 and then generalize the results so obtained. The detailed derivation for n = 2 is given at the end of the manuscript (Appendix D) and here, we present directly the final re- sults:
p(θ) = ϕn (cid:63) Lbn, p = s, a,
bn (θ) = ϕn (cid:63) (Ls + La + Lbn−1),
bj (θ) = ϕn (cid:63) (Lbj+1 + Lbj−1), j = n − 1, . . . , 3,
b2 (θ) = ϕn (cid:63) (Lb3 + Lb1 − L1),
b1 (θ) = ϕn (cid:63) (Lb2 − L2),
1(θ) = −mb1R cosh θ + ϕn (cid:63) (Lb2 − L2),
2(θ) = − ϕn (cid:63) (Lb1 − L1 + L3),
j(θ) = − ϕn (cid:63) (Lj+1 + Lj−1), j = n − 1, . . . , 3,
n(θ) = − ϕn (cid:63) (Ln−1 + Ln+1 + Ln+2),
k(θ) = − ϕn (cid:63) Ln, k = n + 1, n + 2. (B2) In the above equation, the TBA kernel is given by ϕn(θ) = (n + 1)/ cosh[(n + 1)θ], L(θ) = ln[1 + e
(θ)], and L(θ) = ln[1 + e−
(θ)]. Finally, we use the following con- 2π a(θ − θ(cid:48))b(θ(cid:48)). vention for the convolution: a (cid:63) b =(cid:82) dθ(cid:48) The pictorial representation for the TBA equations is given Fig. 3. The solid circles denote the physical mas- sive particles, while the empty ones denote the pseu- doparticles. The cross on the empty circle encodes which of the pseudoparticles has a massive term in its TBA equation. The connectivity of the diagram encodes the structure of the TBA equations. Finally, the arrows de- note, for the TBA equation for a given particle, the sign of the contribution of the different particles. Note the somewhat special structure of the TBA equations. In contrast to the usual structure of the TBA equations, in Eq. [B2], only one pseudoparticle (1) has a (negative!) mass term, while the physical massive particles do not. This is because the equations are written in terms of both the functions L-s and L-s, which allows us to write the TBA equations concisely in terms of a single universal kernel ϕn. When written only in terms of the L-s, the 8 TBA equations are in the 'standard form' and do have the expected structure. Note the difference of the TBA equations from those obtained for the Fateev model. The latter model involves in addition to the two bosonic fields considered in this work, a noncompact bosonic field ϕ with coupling β. The TBA for the Fateev model has been done in the regime FIG. 3. TBA diagram for the dSG model. The solid circles de- note the physical massive particles: soliton (s), antisoliton(a) and breathers (bj, j = 1, . . . , n). The empty circles denote pseudoparticles (j = 1, . . . , n + 2) needed to diagonalize the dSG scattering matrix. The pseudoparticle 1 has a cross on it, which indicates that this particle has a mass term in its TBA equation. As usual, the connectivity of the diagram en- codes which particles show up in the TBA equation of a given particle. The arrows on the links encode the sign of the term on the right hand side of the TBA equation. For instance, for a link connecting particles p, q, if there is an arrow incident on q and none on p, then the right hand side of the TBA equation for p has a minus sign, while the same for q has a plus sign. √ 2π. In when both the couplings α1, α2 are larger than this regime, both the sine-Gordon models are in the re- pulsive regime (without any breathers in either sector) and the resultant TBA is obtained by 'gluing' two sine- Gordon TBA diagrams in the repulsive regime. In con- trast, in the dSG model, the pure quantum integrable 2 = 4π, with α1 (cid:54)= α2. manifold corresponds to α2 Thus, the breathers in one of the sectors have to be taken into account to correctly obtain the thermodynamics of the dSG model. This difference is visible by taking the nave limit of β = 0 in Eq. 33 of Ref. 48, which gives divergent results for the ground state energy. 1 + α2 To check the TBA equations, we first compute the cen- tral charge in the UV limit. The central charge is given by n(cid:88) j=1 (cid:105) mbi ms Lbi(θ) (cid:90) cUV = 6 π2 (cid:104) dθ cosh θ Ls(θ) + La(θ) + (B3) For TBA equations with off-diagonal scattering, the cen- tral charge can be computed in the usual manner using the solutions of the energies
s,
a,
bi,
i-s in the UV (T → ∞) and IR (T → 0) limits61,78. The resultant expression for the central charge can be written as (cid:110) 2Ldlog (cid:16) xs (cid:17) 1 + xs + (cid:104)Ldlog n(cid:88) j=2 (cid:16) xj (cid:17) 1 + xj (cid:16) xbj (cid:17)(cid:105) 1 + xbj + Ldlog cUV = 6 π2 (cid:16) xn+1 (cid:16) yj 1 + xn+1 1 + yj +2Ldlog − n(cid:88) Ldlog j=2 + Ldlog (cid:17) (cid:17) − Ldlog (cid:16) xb1 (cid:16) 1 1 + xb1 (cid:16) (cid:17) (cid:17) (cid:17) − 2Ldlog (cid:16) yn+1 + Ldlog 1 + x1 1 1 + y1 1 + yn+1 (cid:17)(cid:111) , 9 (B4) where Ldlog is the Rogers dilogarithm function and we have used
s =
a,
n+1 =
n+2. Furthermore, xp = e−
p , p = s, a, bi, i, i = 1, . . . , n in the UV limits, while the yp = e−
p in the IR limit. Note that in the IR limit, the energies of the massive physical particles diverge and so, they do not contribute to the expression of the central charge61. The solutions of the TBA equations in the UV limit are given by xn+1 = xj = 1 xs 1 xbj = n, x1 = xb1 = 1, = j2 − 1, j = 2, . . . , n, (B5) while for the IR limit, the non-zero y-s are given by (cid:16) yn+1 = n − 1 2 j − 1 2 yj = , y1 = 3, (cid:17)2 − 1, j = 2, . . . , n. (B6) Plugging these solutions in Eq. [B4] and using some re- markable dilogarithm identities79, we arrive the desired result of cUV = 2. (B7) 1 + α2 Next, we compute the effective central charge as we move away from the UV limit. The perturbation Sint has di- mension (α2 2)/(4π). Thus, on the integrable man- ifold, the perturbation has dimension 1. This leads to logarithmic corrections to the central charge61,62 as one moves away from the UV limit, in addition to the poly- nomial corrections in powers of (msR)254. We compute these corrections by solving the TBA equations numeri- cally and doing a numerical fit of the resulting expression for the central charge. The coefficient of the logarithmic correction for the different cases is then verified using CPT. These calculations are described below for n = 2 and n = 3. Consider the case n = 2, i.e., p1 = 1/3, p2 = 5/3. The first sector spectrum consists of soliton s, antisoliton a and two breathers b1 and b2 whose masses are mb1 = ms and mb2 = 3ms. For the numerical solutions of the TBA equations, it is convenient to use the equations in the standard form, given by √
p(θ) = mpR cosh θ + φ1 (cid:63) (Ls + La + Lb1 − L1) + φ3 (cid:63) Lb2 , p = s, a,
b2 (θ) = mb2 R cosh θ + φ3 (cid:63) (Ls + La + Lb1 − L1) + 2 φ1 (cid:63) Lb2
b1 (θ) = mb1 R cosh θ + φ1 (cid:63) (Ls + La + Lb1 − L1) + φ3 (cid:63) Lb2 − φ2 (cid:63) L2,
1(θ) = φ1 (cid:63) (Ls + La + Lb1 − L1) + φ3 (cid:63) Lb2 − φ2 (cid:63) L2,
2(θ) = − φ2 (cid:63) (Lb1 − L1 + L2 + L3),
k(θ) = − φ2 (cid:63) L2, k = 3, 4, (B8) where the kernels are given by 3 φ1(θ) = 2 cosh(3θ/2) √ φ3(θ) = 3 cosh(3θ/2) cosh(3θ) 2 , φ2(θ) = 3 cosh(3θ) , . (B9) [B8] numerically and plug it in Eq. We solve Eq. [B3] to obtain the effective central charge ceff . The result is shown in Fig. 4. The first six correction terms capturing the deviation from the UV limit is obtained by fitting the numerical data to the following expression: ceff = c0 + a(2) log (msR)2 (2π)2 ln(msR) 5(cid:88) j=1 + a(2) j (msR)2j, (B10) where the superscript 2 denotes the value of n. This leads to c0 = 2, a(2) log = 6, a(2) 3 = −0.01554, a(2) a(2) 1 = −0.39283, a(2) 4 = 0.25742, a(2) 2 = 0.01051, 5 = −1.54212. Identical calculations can be done for n = 3. We only provide the final fit to the effective central charge: log = 3.5149, a(3) c0 = 2, a(3) 3 = −0.00818, a(3) a(3) 4 = 0.1851, a(3) 5 = −1.2709. 1 = −0.29775, a(3) 2 = 0.00739, 10 FIG. 5. Unit cells for a free-boson quantum field theory [panel (a)], the quantum sine-Gordon model [panel (b)] and the double sine-Gordon model [panel (c)]. the role of the bosonic field. As shown in Refs. 67, 80 -- 82, a 1D array built out of this unit cell, in the con- tinuum limit, provides a circuit realization of a 1D Lut- tinger liquid. The action of the latter is given by SLL = (cid:82) d2x(∂µφ)2, where we have rescaled the space and 1 2πK time axis to set the plasmon velocity to be unity. It is es- sential that EJ , EC0 (cid:29) ECJ so that the phase-slip rates are exponentially suppressed. This ensures that despite the presence of the nonlinear Josephson element on the horizontal links, the resultant action is Gaussian. Addi- tion of a Josephson junction on vertical links, adds the Josephson cosine-potential to the action, leading to the sine-Gordon action, where SSG = SLL − EJ0 where EJ0 is the strength of the Josephson nonlinear- ity. It is this nonlinearity that gives rise to the cosine nonlinearity of the SG action. In the quantum field the- ory notation, the SG coupling constant β is given by 4πK. The free-fermion point of the SG model cor- β = responds to β = 4π corresponds to the attractive (repulsive) regime of the model. (cid:82) d2x cos φ, 4π, while β ≤ (≥) √ √ √ From the above circuits, it is straightforward to infer the circuit realizing the dSG action, presented in Fig. 5 (c). Consider the upper and lower halves of the array, without the mutual inductance M and the capacitance C, which couples the two halves. From the above expla- nation, it is clear that each of the two halves of the array realize the SG actions, given by SSG,i, i = 1, 2: SSG,i = 1 2πK −EJ0 d2x(∂µϕi)2 d2x cos ϕi, i = 1, 2. (C1) Now, consider the mutual inductance M and the capac- itance C. The first leads to a term in the action SM = −M d2x∂xϕ1∂xϕ2, (C2) (cid:90) (cid:90) (cid:90) (cid:90) while the second leads to SC = C 2 d2x(∂tϕ1 − ∂tϕ2)2. (C3) FIG. 4. Effective central charge for n = 2. The solid blue line is obtained by solving the TBA equations given by Eq. [B8]. The black dots are obtained by performing a fit to Eq. [B10]. The coefficient of the logarithmic term in the effective central charge calculation can be checked analytically us- ing CPT. Taking Eq. [A12] for h1 = 0 and using Eq. [A29], we arrive at the expression for the free energy: f = − 1 LR = − 1 LR ln Z = − πceff 6R2 ln Z0 − m2 s π sin2 π 2(n + 1) (cid:16) R (cid:17) 2a ln . (B11) Thus, we arrive at an analytical the logarithmic correction in ceff , proportional (msR)2 (2π)2 ln(msR), denoted by alog: expression for to alog = 24 sin2 π 2(n + 1) . (B12) For n = 2 and n = 3, this coefficient is given by 6 and 3.5147, which are in good agreement with what was ob- tained by numerically solving the TBA equations (Eqs. [B11, B11]). We have checked our results for n = 4 and n = 5 as well. Appendix C: The double sine-Gordon model with quantum circuits In this section of the appendix, we present some addi- tional details on the realization of the dSG with quantum electronic circuits. As mentioned in the main text, the dSG model can be viewed as two sine-Gordon (SG) mod- els coupled by a mutual inductance and a capacitance. Below, we sketch how the different terms of the circuit contribute to the different terms of the action presented in Eq. (1) of the main text. Consider the circuit in Fig. 5 (a). Define the node-flux at a point in space-time (x, t) as φ(x, t)25, which plays The last two equations follow directly from the current- voltage constitutive relations of the circuit elements un- der consideration. The total action of the dSG circuit is given by SSG,1 +SSG,2 +SM +SC. From this action, sim- ple algebraic manipulations give rise to the dSG action in terms of the fields ϕ1,2, given by (cid:90) d2x[(∂µϕ1)2 + (∂µϕ2)2] + B √ d2x{cos(2 πϕ1) + cos(2 √ d2x∂µϕ1∂µϕ2 πϕ2)}, (C4) SdSG = (cid:90) (cid:90) A 2 − M0 π where A = π(1/α2 terms of the circuit components, 1 + 1/α2 2) and B = π(1/α2 1 − 1/α2 2). In α2 1 = Kπ 2 , α2 2 = π 2/K + 4πC , (C5) which are the expressions provided in the main text of the article. In the above derivation of the dSG action from the circuit, we have implicitly assumed that the unit cell is repeated an infinite number of times with periodic/open boundary conditions. However, real experiments are done with finite number of such unit cells. There are two main difficulties in realizing arrays very large number of junctions. First, in the derivation of the dSG action, we have neglected the phase-slips in the Josephson junctions in the horizontal links. However, for a large enough num- ber of unit cells, this rate is no longer negligible. A crude estimate for the maximum number of unit cells is pro- √ vided by the inverse of the phase-slip rate and is given by EJ /ECJ 64,67. The same problem also provides a limi- e tations on the frequencies over width the phase-slip rates can be safely neglected and is given by u(cid:112)CJ /C0/a0, (cid:112)2EC0EJ , the where a0 is the lattice spacing and u = a0 plasmon velocity67. Second, the superinductance of the Josephson junction on the horizontal links should not be shunted by the capacitance to ground. This is ensured 64. Despite these difficulties, ar- by choosing N < CJ /C0 rays with up to O(105) Josephson junctions have been reliably built and experimentally analyzed83 and we are (cid:89) ks b1s(θ − θks ) Sp1 b2s(θ − θks) Sp1 (cid:89) ss (θ − θks ) (cid:89) Sp1 ka b1a(θ − θka ) (cid:89) Sp1 b2a(θ − θka ) Sp1 (cid:89) sa(θ − θka ) (cid:89) Sp1 (cid:89) kb1 kb1 ka Sp1 sb1 Sp1 b1b1 Sp1 b2b1 ) (cid:89) (cid:89) (cid:89) kb2 kb2 Sp1 sb2 Sp1 b1b2 Sp1 b2b2 (θ − θkb1 (θ − θkb1 (θ − θkb1 ) ) ka kb1 kb2 (cid:89) (cid:89) ks ks 11 optimistic that our analytical predictions can be experi- mentally verified. Next, we comment on some of the possible measure- ments in the circuit. Estimation of α1, α2 can be done by biasing with two current sources (the 'magnetic fields' h1, h2 considered earlier) and measuring the transmis- sion of an input field through the dSG lattice. This provides an estimate of the average current, (cid:104)∂xϕ1,2(cid:105), flowing through the horizontal links in the circuit. A different approach will be to locally probe the dSG cir- cuit by coupling resonators to a unit cell and measuring current/voltage correlations. While transport properties can be measured in the ways mentioned above, the spe- cific heat can be measured using AC calorimetry84. The latter method involves inducing temperature oscillations on the array patterned on a membrane. The amplitude of the oscillations, together with the applied amplitude and frequency of the applied AC power provides the specific heat84,85. A detailed theoretical analysis for the specific heat measurements for the sine-Gordon model has been made in Refs. 69 and 86. We note that opening the circuit to drive and dis- sipation leads to an interesting and physically relevant problem of dissipative quantum field theories, a largely open field. The calculations made for the sine-Gordon model87,88 indicate a finite width for the sine-Gordon minima which presumably leads to a finite width for the different particles in the spectrum. We speculate that a similar phenomenon may occur in the dSG model. How- ever, a full quantum analysis is still an open problem even for the sine-Gordon model and we hope to return to this problem in the future. Appendix D: Derivation of the TBA equations for n = 2 √ Consider again the case n = 2, i.e., p1 = 1/3, p2 = 5/3. The first sector spectrum consists of soliton s, antisoliton a and two breathers b1 and b2 whose masses are mb1 = ms and mb2 = 3ms. We refer to the two sine-Gordon sec- tors as T1, T2 which correspond to the couplings α1, α2. The periodicity condition for the wave function of a given number of solitons, antisolitons and breathers on a circle of length L becomes the following set of equations (θ − θkb2 (θ − θkb2 (θ − θkb2 s (θ{θks},{θka},{θkb1 )λp2 (θ{θks},{θka},{θkb1 (θ{θks},{θka},{θkb1 )λp2 b2 )λp2 b1 },{θkb2 },{θkb2 },{θkb2 }) = e−imsL sinh θ }) = e−im1L sinh θ }) = e−im2L sinh θ (D1) where θ is the rapidity of the particle going around the world, θks,a,b1 ,b2 the rapidities of the external "incoming" solitons, antisolitons and breathers and in which λp2 rep- i resents the contribution to the phase shift of the scatter- ing of the second quantum number. We compute these contributions by the diagonalizing the transfer matrix of a T2-soliton going around the world repulsively interact- ing with solitons and antisolitons with amplitudes Sp2 SG. The diagonalization procedure for repulsive sine-Gordon produces in the thermodynamic limit equations for the densities of states in terms of massless pseudoparticles and a physical particle carrying the mass of the soliton. When "gluing" this result with the equations (D1) the only care will be needed in the identification of the role of the massive physical particles. Through the algebraic Bethe ansatz technique57,61, the eigenvalues of the repul- sive sine-Gordon transfer matrix are 12 λp2(θ,{θi},{yr}) = (cid:89) i ss (θ − θi) Sp2 (cid:89) r sinh 1 p2 sinh 1 p2 (iπ − yr − θ) (yr − θ) . (D2) The rapidities yr label the Bethe eigenvectors of the transfer matrix and have to satisfy the Bethe equation (cid:89) sinh 1 p2 sinh 1 p2 i (iπ − yr − θ) (yr − θ) = −(cid:89) r(cid:48) sinh 1 p2 sinh 1 p2 (yr − yr(cid:48) − iπ) (yr − yr(cid:48) + iπ) . (D3) The presence of the external particles does not invali- date the string classification56,58 of the rapidities yr in the thermodynamic limit. This classification applies actually to the case p2 > 2. It is easy to convince yourself that the classification of the yr's of equation (D3) is equivalent to r = (1 − p2)yr of an equation the classification of the y(cid:48) 2 = (1 − 1/p2)−1. For the general like (D3) but with p(cid:48) case p2 = 2 − 1/(n + 1) we have the following pattern of n + 2 strings y(j) k = (cid:60)y(j) − iπ y(j) 2 n n + 1 (2j − 1 − 2k) − iπ 2 p2 1 − vj 2 , k = 1, 2, . . . Nj , j = 1, . . . , n + 2 . (D4) The lengths Nj and the parities vj of the string y(j) are N1 = 1 ; Nj = 2j − 3 , j = 2, 3, . . . , n + 1 ;Nn+2 = 2 v1 = 1 ; vj = (−1)j−1 , j = 2, 3, . . . , n + 1 ;vn+2 = 1 . (D5) Since there cannot exist different strings with the same length and parity it is useful to denote the strings with Nv. The center of a string, i.e. its real part, plays the role in the thermodynamical analysis of the rapidity of a massless pseudoparticle Nv to which we can associate a density PNv of states containing it and a density of occupied states P + . As usual the Yang-Yang relations Nv between the densities is obtained by plugging the classifi- cation (D4) in the Bethe equation (D3) and by taking the imaginary part of the logarithmic derivative with respect to the external rapidity. The result for the pseudoparti- cles is  φi0 (cid:63) P + (cid:88) j 0 + φij (cid:63) P + j (D6) Pi(y) = (−1)vi where   φij(y) = −id/dy log = (−1)vivj 2 p2 φi0(y) = −id/dy log  Ni(cid:89) Nj(cid:89) Ni(cid:88) Nj(cid:88) (cid:40) Ni(cid:89) kj =1 ki=1 ki=1 kj =1 ki=1 cosh 2y p2 − sinh 1 sinh 1 p2 p2 sinh 1 p2 sinh 1 p2 [y − iπ [y − iπ 2 2 n n n+1 (Ni − Nj − 2ki + 2kj) − iπ n+1 (Ni − Nj − 2ki + 2kj) iπ n+1 (Ni − Nj − 2ki + 2kj) + 2] [ n 2 p2 1−vivj 2 p2 − 1−vivj 2 − iπ] 2 + iπ] sin π p2 − (−1)vivj cos π [y − iπ n p2 [y − iπ 2 2 n n+1 (Ni + 2 − 2ki) iπ n+1 (Ni + 2 − 2ki) − iπ 1−vi 2 p2 2 ] 2 p2 − 1−vi 2 − iπ] n+1 (Ni − Nj − 2ki + 2kj) + 2] [ n , (D7) (cid:41) Ni(cid:88) cosh 2y p2 ki=1 = (−1)vi 2 p2 n+1 (Ni + 2 − 2ki)] [ n sin π p2 − (−1)vi cos π n+1 (Ni + 2 − 2ki)] [ n p2 13 (D8) are the kernels of the particle i with the physical particle denoted with 0 and with the other pseudoparticles j. The convolution normalization is a (cid:63) b =(cid:82) dθ(cid:48) 2π a(θ − θ(cid:48))b(θ(cid:48)). For sine-Gordon the method would go on with the equa- tion for the density of the physical particle state, obtain- able from the periodicity equation. But in the case we are examining now the role of the physical particle is played by the T1 soliton, antisoliton and breathers. Therefore in our case we have to modify the equation (D6) to the equation Pi(y) = (−1)vi where the breather kernels are φi0 (cid:63) (P + (cid:18) s + P + a ) +  (cid:88) j φibj (cid:63) P + bj + (cid:88) j φij (cid:63) P + j (cid:19) (cid:18) + φi0 (cid:19) . θ − iπ 2 n − j + 1 n + 1 φibj (θ) = φi0 θ + n − j + 1 n + 1 iπ 2 (D9) (D10) The equations for the densities of the physical particles states can be obtained by taking the logarithmic deriva- tive of equations (D1) with the explicit formulas for the eigenvalues (D2) and (D4). The result is Ps(θ) = Pbi(θ) = msL 2π miL 2π cosh θ + φss (cid:63) (P + s + P + a ) + cosh θ + φbis (cid:63) (P + s + P + a ) + (cid:88) (cid:88) j (cid:88) (cid:88) j φsbj (cid:63) P + bj + φsj (cid:63) P + j φbibj (cid:63) P + bj + φbij (cid:63) P + j . (D11) j j The new kernels are (cid:19) (cid:18) + φ00 n − j + 1 n + 1 iπ 2 θ − iπ 2 n − j + 1 n + 1 (cid:19)(cid:21) (cid:19) (cid:18) θ + (cid:19) i − j n + 1 iπ 2 φss(θ) = −i d dθ φsbj (θ) = −i d dθ φsi(θ) = φi0(θ) φbis(θ) = φsbi(θ) φbibj (θ) = −i d dθ log Sp1 ss (θ) + φ00(θ) log Sp1 sbj (θ) + φ00 θ + (cid:18) (cid:18) (cid:20) (cid:20) (cid:19) φ00 (cid:18) log Sp1 bibj θ − iπ 2 (θ) + i − j n + 1 θ + (cid:18) iπ 2 θ − iπ 2 + φ00 + φ00 2n − i − j + 2 + φ00 (cid:19)(cid:21) n + 1 2n − i − j + 2 φbij(θ) = φjbi(θ) where we denote with φ00(θ) = −i d dθ log Sp2 ss (θ) n + 1 (cid:90) = dkeikθ (D12) (D13) sinh(p2 − 1) πk 2 cosh πk 2 sinh p2 πk 2 2 the soliton-soliton kernel of the pure p2-sineGordon TBA. In order to be concrete, let's come back to the specific example p1 = 1/3 and p2 = 5/3. The equations for the density can be written in Fourier space with φ(θ) = (cid:82) dkeikθ φ(k) and assume the form 14 (cid:20) FT Ps − msL 2π cosh θ (cid:20) Pb1 − mb1 L 2π (cid:20) Pb2 − mb2 L 2π FT FT cosh θ cosh θ (cid:21) (cid:21) (cid:21) = − 2c1/6s1/3 a ) − s1/2 s5/6 s1/6 s5/6 s + P + ( P + s5/6 1 − s1/3 P + 1− + s5/6 a ) − s1/6 s5/6 ( P + P + P + b1 − 4c2 P + b1 2 − s1/6 P + s5/6 − 2c1/6s1/2 1/6s1/3 s5/6 P + 3 P + b2 P + b2 s + P + 1 − s2/3 P + s5/6 P + 1− + 2c1/3s1/6 s5/6 s5/6 2 − 2c1/3s1/6 P + P + 3 + s1/2 s5/6 = − s1/2 s5/6 s1/6 s5/6 = − 4c2 + 1/6s1/3 s5/6 + 2c1/6s1/2 s5/6 ( P + s + P + 1 − 2c1/6s1/3 P + s5/6 a ) − 2c1/6s1/2 s5/6 P + b1 + s5/6 c1/6(s1/6 + s1/3 − 3s5/6) P + b2 c1/2s5/6 P + 3 ( P + s + P + a ) + P + 1 + s2/3 s5/6 ( P + s + P + a ) + + P + b1 s1/6 s5/6 1− − 2c1/3s1/6 P + 2c1/3s1/6 s5/6 P + 1− + s1/3 s5/6 2c1/6s1/2 2 − s1/3 P + s5/6 P + b2 s5/6 P + 2 + 2c1/3s1/6 s5/6 P + 3 P + b1 + 2c1/6s1/3 s5/6 P + b2 s5/6 P + 1 + s1/6 s5/6 P + 1− + 2c1/2s1/6 s5/6 s + P + ( P + s1/6 s5/6 − 2c1/3s1/6 s5/6 2c1/3s1/6 a ) + s5/6 1 − 2c1/2s1/6 P + s5/6 P + b1 + P + 1− + P + 3 s5/6 2 − 2c1/2s1/6 P + s1/3 s5/6 c1/2s1/6 c1/6s5/6 P + b2 2 − c1/2s1/6 P + c1/6s5/6 s1/2 s5/6 − s1/6 s5/6 s1/3 s5/6 − s2/3 s5/6 P1 = P1− = P2 = P3 = P2 P + 3 (D14) where sm = sinh mπk and cm = cosh mπk and the positive parity of the strings 1+, 2+ and 3+ is understood in the notation. Furthermore we never write the corresponding equation for the antisoliton density being trivially Ps = Pa. Simple manipulations with trigonometric identities re- sults into a drastic simplification of the system (cid:20) FT (cid:20) Ps − msL 2π (cid:20) Pb2 − mb2 L 2π Pb1 − mb1 L 2π cosh θ cosh θ cosh θ FT FT (cid:21) (cid:21) (cid:21) = − 1 2c1/3 = − c1/6 c1/3 = − 1 2c1/3 1 ( P + s + P + ( P + s + P + ( P + s + P + a ) − 1 2c1/3 a ) − c1/6 c1/3 a ) − 1 2c1/3 1 + P + b1 − c1/6 c1/3 + P + b2 − 1 2c1/3 + P − 1 + P − 1 + P + b1 + P + b2 − 1 c1/3 − c1/6 c1/3 + P + b1 + P + b2 + P + b1 + c1/6 c1/3 + P + b2 + − c1/6 c1/3 − 1 2c1/3 1 2c1/3 + P − 1 s + P + a ) + 2c1/3 2 + P + 3 + P − 1 + P + (cid:105) + P − 1 − 1 2c1/6 + P − 1− P1 = P1− = P2 = ( P + (cid:104) P + b1 P − 1− 2c1/3 1 2c1/6 1 2c1/6 P3 = P2 . 15 (D15) The hole densities P − i are defined as P − i = Pi − P + i . We now have all the ingredients to make a thermody- namical analysis. The internal energy is (cid:90) U = dθ{ms cosh θ[P + s (θ) + P + a (θ)] +mb1 cosh θP + b1 (θ) + mb2 cosh θP + b2 (θ)} (D16) and the entropy, in the Stirling approximation, (cid:90) (cid:88) S = dθ (Pi log Pi + P + i log P + i + P − i log P − i ) . i (D17) Next, we minimizing the free energy F = U − T S (here T is the temperature) with respect to the densities given by i = {P + ρ+ s , P + a , P + bi , P + 2 , P + 3 , P − 1 , P − 1−}. (D18) Taking into account the constraints of Eq. (D15), we end up to the following TBA system for the pseudoenergies
i(θ) defined as ρ+ i /Pi = e−
i/(1 + e−
i)
s(θ) = msR cosh θ + φ1 (cid:63) [Ls + La + Lb1 − L1] + φ3 (cid:63) Lb2
b2(θ) = mb2R cosh θ + φ3 (cid:63) [Ls + La + Lb1 − L1] + 2 φ1 (cid:63) Lb2
b1(θ) = mb1R cosh θ + φ1 (cid:63) [Ls + La + Lb1 − L1] + φ3 (cid:63) Lb2 − φ2 (cid:63) L1−
1(θ) = φ1 (cid:63) [Ls + La + Lb1 − L1] + φ3 (cid:63) Lb2 − φ2 (cid:63) L1−
1− (θ) = − φ2 (cid:63) [Lb1 − L1 + L2 + L3]
2(θ) = − φ2 (cid:63) L1−
3(θ) = − φ2 (cid:63) L1−, (D19) where R is the inverse temperature. Here we have intro- duced the notation Li(θ) = log(1 + e−
i) and the new kernels φ1(k) = 1/2 cosh πk φ3(k) = cosh πk 6 / cosh πk 3 . 3 , φ2(k) = 1/2 cosh πk 6 and A further trivial analytic and algebraic manipulation gives the final form for the
-system
s(θ) = ϕ2 (cid:63) Lb2
b2(θ) = ϕ2 (cid:63) [Ls + La + Lb1 − L1]
b1(θ) = ϕ2 (cid:63) [Lb2 − L2]
1(θ) = −mb1 R cosh θ + ϕ2 (cid:63)(cid:2)Lb2 − L1−
1− (θ) = − ϕ2 (cid:63) [Lb1 − L1 + L2 + L3]
2(θ) = − ϕ2 (cid:63) L1−
3(θ) = − ϕ2 (cid:63) L1− . (cid:3) (D20) The functions Li are the positive part of
i, i.e. Li = log(1 + e
i). The universal kernel is ϕ2(k) = 1/2 cosh πk 6 . The above procedure can be generalized to other integer values of n, leading to the equations presented in Eq. (B2). ∗ ananda.roy@tum.de 1 R. P. Feynman, Inernational Journal of Theoretical Physics 21, 467 (1982). 2 S. Lloyd, Science 273, 1073 (1996). 3 D. S. Abrams and S. Lloyd, Phys. Rev. Lett. 79, 2586 (1997). 31 T. Senthil and M. P. A. Fisher, Phys. Rev. B 62, 7850 (2000). 32 C. Xu and L. Fu, Phys. Rev. B 81, 1 (2010). 33 A. Roy, B. M. Terhal, and F. Hassler, Phys. Rev. Lett. 4 B. Dou¸cot, L. B. Ioffe, and J. Vidal, Phys. Rev. B 69, 119, 180508 (2017). 16 214501 (2004). 5 H. P. Buchler, M. Hermele, S. D. Huber, M. P. A. Fisher, and P. Zoller, Phys. Rev. Lett. 95, 040402 (2005). 6 J. I. Cirac, P. Maraner, and J. K. Pachos, Phys. Rev. Lett. 105, 190403 (2010). 7 J. Casanova, L. Lamata, I. L. Egusquiza, R. Gerritsma, C. F. Roos, J. J. Garc´ıa-Ripoll, and E. Solano, Phys. Rev. Lett. 107, 260501 (2011). 8 S. P. Jordan, K. S. M. Lee, and J. Preskill, Quant. Inf. 34 Y. V. Nazarov and Y. M. Blanter, Quantum Transport: Introduction to Nanoscience (Cambridge University Press, 2009). 35 R. Rajaraman, Solitons and Instantons: An Introduc- tion to Solitons and Instantons in Quantum Field Theory, North-Holland personal library (North-Holland Publishing Company, 1982). 36 M. Ablowitz and H. Segur, Solitons and the Inverse Scat- tering Transform, SIAM Studies in Applied Mathematics (Society for Industrial and Applied Mathematics, 2006). 37 S. Mandelstam, Phys. Rev. D 11, 3026 (1975). 38 S. Ghoshal and A. Zamolodchikov, of Modern Physics A 09, Journal https://doi.org/10.1142/S0217751X94001552. International 3841 (1994), 39 K. Zarembo, in Les Houches Summer School: Integrability: From Statistical Systems to Gauge Theory Les Houches, France, June 6-July 1, 2016 (2017) arXiv:1712.07725 [hep- th]. 40 D. Bernard and A. LeClair, Comm. Math. Phys. 142, 99 Comput. 14 (2014). Jordan, 9 S. J. http://science.sciencemag.org/content/336/6085/1130.full.pdf. P. Preskill, K. Science and (2012), Lee, 1130 S. M. 336, 10 A. A. Houck, H. E. Tureci, and J. Koch, Nature Physics 8, 292 EP (2012). 11 M. H. Devoret and R. J. Schoelkopf, Science 339, 1169 (2013). 12 P. Fendley, A. W. W. Ludwig, and H. Saleur, Phys. Rev. B 52, 8934 (1995). 13 P. Fendley, A. W. W. Ludwig, and H. Saleur, Phys. Rev. (1991). Lett. 74, 3005 (1995). 41 A. Belavin, A. Polyakov, and A. Zamolodchikov, Nuclear 14 F. Milliken, C. Umbach, and R. Webb, Solid State Com- Physics B 241, 333 (1984). munications 97, 309 (1996). 42 A. B. Zamolodchikov, JETP Lett. 46, 160 (1987), [Pisma 15 F. Lesage, H. Saleur, and P. Simonetti, Phys. Rev. B 56, Zh. Eksp. Teor. Fiz.46,129(1987)]. 7598 (1997). 16 D. Giuliano and P. Sodano, Nuclear Physics B 711, 480 (2005). 17 V. Gritsev, A. Polkovnikov, and E. Demler, Phys. Rev. B 75, 174511 (2007). 18 A. P. Bukhvostov and L. N. Lipatov, Nucl. Phys. B180, 116 (1981), [Pisma Zh. Eksp. Teor. Fiz.31,138(1980)]. 19 H. Saleur, Journal of Physics A: Mathematical and General 32, L207 (1999). 20 T. Giamarchi, Quantum Physics in One Dimension, In- ternational Series of Monographs on Physics (Clarendon Press, 2003). 21 A. Gogolin, A. Nersesyan, and A. Tsvelik, Bosonization and Strongly Correlated Systems (Cambridge University Press, 2004). 22 M. P. A. Fisher, P. B. Weichman, G. Grinstein, and D. S. Fisher, Phys. Rev. B 40, 546 (1989). 23 S. Sachdev, Quantum Phase Transitions (Cambridge Uni- versity Press, 2011). 24 M. Tinkham, Introduction to Superconductivity: Second Edition, Dover Books on Physics (Dover Publications, 2004). 25 M. H. Devoret, Quantum Fluctuations in Electrical Cir- cuits (Les Houches Session LXIII) (Elsevier, 1997) pp. 351 -- 386. 26 This is true also if one tries to "backtrack" to the fermionic liquid" fixed points are formulation, where "non-Fermi then encountered. 27 D. Iagolnitzer, Phys. Rev. D 18, 1275 (1978). 28 A. B. Zamolodchikov and A. B. Zamolodchikov, Annals of Physics 120, 253 (1979). 29 A. Zamolodchikov, Nuclear Physics B 342, 695 (1990). 30 F. Smirnov, Form Factors in Completely Integrable Models of Quantum Field Theory, Advanced series in mathemati- cal physics (World Scientific, 1992). 43 A. Zamolodchikov, in Integrable Sys Quantum Field The- ory, edited by M. Jimbo, T. Miwa, and A. Tsuchiya (Aca- demic Press, San Diego, 1989) pp. 641 -- 674. 44 P. Francesco, P. Di Francesco, P. Mathieu, D. S´en´echal, and D. Senechal, Conformal Field Theory, Graduate Texts in Contemporary Physics (Springer, 1997). 45 G. Mussardo, Statistical Field Theory: An Introduction to Exactly Solved Models in Statistical Physics, Oxford Grad- uate Texts (OUP Oxford, 2010). 46 I. Affleck, M. Oshikawa, and H. Saleur, Nuclear Physics B 594, 535 (2001). 47 M. Ameduri and C. J. Efthimiou, Nonlinear Mathematical Physics 5, https://doi.org/10.2991/jnmp.1998.5.2.4. 132 Journal of (1998), 48 V. A. Fateev, Nucl. Phys. B473, 509 (1996). 49 F. Lesage, H. Saleur, and P. Simonetti, Phys. Rev. B 57, 4694 (1998). 50 M. Ameduri, C. J. Efthimiou, and B. Gerganov, Mod. Phys. Lett. A14, 2341 (1999), arXiv:hep-th/9810184 [hep- th]. 51 V. V. Bazhanov, S. L. Lukyanov, and B. A. Runov, Nu- clear Physics B 927, 468 (2018). 52 V. Fateev, E. Onofri, and A. Zamolodchikov, Nuclear Physics B 406, 521 (1993). 53 On the integrable manifold, the mass of the ϕ2 soliton is also given by ms. 54 T. R. Klassen and E. Melzer, Nuclear Physics B 350, 635 (1991). 55 R. Dashen, S.-k. Ma, and H. J. Bernstein, Phys. Rev. 187, 345 (1969). Takahashi 56 M. of (1972), pdf/48/6/2187/5255323/48-6-2187.pdf. Progress 2187 http://oup.prod.sis.lan/ptp/article- Suzuki, 48, Theoretical Physics and M. 57 P. Fendley and K. Intriligator, Nuclear Physics B 372, 533 (1992). 71 A. Roy, L. Jiang, A. D. Stone, and M. Devoret, Phys. Rev. 58 M. Takahashi, Thermodynamics of One-Dimensional Solv- Lett. 115, 150503 (2015). 17 able Models (Cambridge University Press, 2005). 59 V. E. Korepin, a. G. Izergin, and N. M. Bogoliubov, "Quantum Inverse Scattering Method and Correlation Functions," (1993). 60 L. D. Faddeev, in Relativistic gravitation and gravitational radiation. Proceedings, School of Physics, Les Houches, France, September 26-October 6, 1995 (1996) pp. pp. 149 -- 219, arXiv:hep-th/9605187 [hep-th]. 61 A. Zamolodchikov, Nuclear Physics B 358, 497 (1991). 62 P. Fendley, H. Saleur, and A. B. Zamolodchikov, Inter- national Journal of Modern Physics A 08, 5717 (1993), https://doi.org/10.1142/S0217751X93002265. 63 A. B. Zamolodchikov, Journal of Physics A: Mathematical and General 39, 12847 (2006). 64 V. E. Manucharyan, and M. H. Devoret, http://science.sciencemag.org/content/326/5949/113.full.pdf. J. Koch, L. Science 326, I. Glazman, 113 (2009), 65 J. Koch, V. Manucharyan, M. H. Devoret, and L. I. Glaz- man, Phys. Rev. Lett. 103, 217004 (2009). 66 K. Le Hur, Phys. Rev. B 85, 140506 (2012). 67 M. Goldstein, M. H. Devoret, M. Houzet, and L. I. Glaz- man, Phys. Rev. Lett. 110, 017002 (2013). 68 J. Koch, T. M. Yu, J. Gambetta, A. A. Houck, D. I. Schus- ter, J. Majer, A. Blais, M. H. Devoret, S. M. Girvin, and R. J. Schoelkopf, Phys. Rev. A 76, 1 (2007). 69 C. Guarcello, P. Solinas, A. Braggio, and F. Giazotto, Scientific Reports 8, 12287 (2018). 70 N. Bergeal, R. Vijay, V. E. Manucharyan, I. Siddiqi, R. J. Schoelkopf, S. M. Girvin, and M. H. Devoret, Nat Phys 6, 296 (2010). 72 A four-field integrable QFT with QEC-s will be described elsewhere. 73 L. Fu, Phys. Rev. Lett. 104, 1 (2010). 74 A. B. Zamolodchikov, Comm. Math. Phys. 79, 489 (1981). 75 A. Jour- nal (1995), https://doi.org/10.1142/S0217751X9500053X. of Modern Physics A 10, International 1125 Zamolodchikov, B. 76 A. Polyakov and P. Wiegmann, Physics Letters B 131, 121 (1983). 77 P. B. Wiegmann, Physics Letters B 152, 209 (1985). 78 A. Zamolodchikov, Nuclear Physics B 358, 524 (1991). 79 L. Lewin, Polylogarithms and associated functions (North Holland, 1981). 80 A. A. Clerk, M. H. Devoret, S. M. Girvin, F. Marquardt, and R. J. Schoelkopf, Rev. Mod. Phys. 82, 1155 (2010). 81 A. Roy and M. Devoret, Comptes Rendus Physique 17, 740 (2016). 82 A. Roy and M. Devoret, Phys. Rev. B 98, 045405 (2018). 83 R. Kuzmin, N. Grabon, N. Mehta, R. Mencia, N. Pankra- tova, M. Goldstein, and V. Manucharyan, in APS Meeting Abstracts (2018) p. H33.003. 84 F. Fominaya, T. Fournier, P. Gandit, and J. Chaussy, Instruments 68, 4191 (1997), Review of Scientific https://doi.org/10.1063/1.1148366. 85 H. Rabani, F. Taddei, O. Bourgeois, R. Fazio, and F. Gi- azotto, Phys. Rev. B 78, 012503 (2008). 86 C. Guarcello, P. Solinas, A. Braggio, and F. Giazotto, Phys. Rev. B 98, 104501 (2018). 87 K. Hida and U. Eckern, Phys. Rev. B 30, 4096 (1984). 88 K. Hida, Zeitschrift fur Physik B Condensed Matter 61, 223 (1985). 

1903.05731

1

1903

Hydrodynamic Phonon Transport: Past, Present, and Prospect

The hydrodynamic phonon transport was studied several decades ago for verifying the quantum theory of lattice thermal transport. Recent prediction of significant hydrodynamic phonon transport in graphitic materials shows its practical importance for high thermal conductivity materials and brought a renewed attention. As the study on this topic has been inactive to some extent for several decades, we aim at providing a brief overview of the past studies as well as very recent studies. The topics we discuss in this chapter include the collective motion of phonons, several approaches to solve the Peierls-Boltzmann transport equation for hydrodynamic phonon transport, the role of normal scattering for thermal resistance, and the propagation of second sound. Then, we close this chapter with our perspectives for the future studies and the practical implication of hydrodynamic phonon transport.

cond-mat.mes-hall

cond-mat

Hydrodynamic Phonon Transport: Past, Present, and Prospect 1,2Sangyeop Lee*, 1Xun Li 1Department of Mechanical Engineering and Materials Science, University of Pittsburgh, Pennsylvania, 15261, USA 2Department of Physics and Astronomy, University of Pittsburgh, Pennsylvania, 15261, USA *sylee@pitt.edu Abstract The hydrodynamic phonon transport was studied several decades ago for verifying the quantum theory of lattice thermal transport. Recent prediction of significant hydrodynamic phonon transport in graphitic materials shows its practical importance for high thermal conductivity materials and brought a renewed attention. As the study on this topic has been inactive to some extent for several decades, we aim at providing a brief overview of the past studies as well as very recent studies. The topics we discuss in this chapter include the collective motion of phonons, several approaches to solve the Peierls-Boltzmann transport equation for hydrodynamic phonon transport, the role of normal scattering for thermal resistance, and the propagation of second sound. Then, we close this chapter with our perspectives for the future studies and the practical implication of hydrodynamic phonon transport. I. Introduction The transport of phonons, a major heat carrier in non-metallic solids, has been usually described by the diffusive limit since the Fourier's law was suggested 200 years ago. The Fourier's law has a simple form that correlates thermodynamic driving force (i.e. temperature gradient, −∇T ) and the resulting heat flux ( ′′q ): 1 κ ′′q = −∇T (1) κ , involved in the transport This empirical law shows that there is always a damping coefficient, 1/ phenomena. The 1/ is thermal resistance which determines the extent of damping in heat flow and the resulting heat flux at a given temperature gradient. However, such a damping effect is not observed in fluid flow although both phonons and molecules are well-described by the same Boltzmann transport theory. Also, they have similar thermodynamic driving forces; molecules are driven by pressure gradient like phonons are driven by temperature gradient. Assuming an infinitely large domain to exclude any effect from boundary, the molecular flow at macroscale can be described by the Euler's equation: κ D(ρu) Dt = −∇p (2) ρ u −∇p and are the density and velocity of fluid element. With the Lagrangian coordinate, Eq. ) without any damping where (2) shows the acceleration of molecules under the pressure gradient ( effect. This is the thermodynamic limit where the entropy generation is zero. Now one may ask a question -- why does phonon flow described by the Fourier's law exhibit a damping effect while molecular flow does not? Interestingly, Nernst speculated a century ago that heat in high thermal conductivity materials may have inertia like fluid [1]. The different behaviors of damping in molecular and phonon flows can be associated with the difference in scattering processes of those two particles in terms of the momentum conservation. For molecular flow, total momentum of molecules is always conserved upon molecule-molecule scattering. Therefore, inter-molecular scattering itself cannot cease the given molecular flow. For phonon flow, however, the total momentum of phonons is not always conserved upon phonon-phonon scattering. There are two different scattering mechanisms regarding the momentum conservation: normal and umklapp scattering (hereafter N-scattering and U-scattering, respectively), suggested by Peierls [2]. As shown in Fig. 1(a), the N-scattering involves phonon states with small wavevectors and the total momentum of phonon particles is conserved ( ) like inter- molecular scattering case. However, for U-scattering, the total momentum of phonon particles is not conserved. As a result, the phonon propagation direction is reversed upon U-scattering, thus directly causing thermal resistance. Phonon scattering by impurities shown in Fig. 1c also directly causes thermal resistance as it does not conserve total momentum. Hereafter R-scattering refers to combined U- and impurity-scattering. In most materials at room temperature, N-scattering is weak compared to R-scattering, leading to the large damping effect of heat flow in solid materials. q1 + q2 = q3  (a)$Normal$Scattering !" !$ !# !′$ (b)$Umklapp Scattering !" !$ & !# (c)$Impurity$Scattering !" !# Fig. 1. Schematic of N-scattering, U-scattering, and impurity scattering in the reciprocal space. The hexagon represents the first Brillouin zone. We have assumed the infinitely large domain to compare the intrinsic damping of phonon flow and molecular flow. However, all solid materials have finite size, introducing phonon- boundary scattering. In most cases, the phonon-boundary scattering is diffuse boundary scattering rather than specular boundary scattering. The three types of phonon scattering (i.e., diffuse boundary scattering, N-scattering, and R-scattering) influence phonon transport in different ways and thus there exist three regimes of phonon transport -- ballistic, hydrodynamic, and diffusive regimes schematically shown in Fig. 2 -- depending on the dominant type of scattering mechanisms. Those three regimes occur in different ranges of temperature. The ballistic regime occurs at low temperature where internal phonon scattering is much weaker than phonon-boundary scattering. Therefore, the phonon transport is limited by the diffuse boundary scattering and the thermal resistance is determined by the size and shape of samples. As temperature increases, the internal phonon scattering starts to play a role in the transport process. At sufficiently low temperature, majority of internal phonon scattering is N-scattering as phonon states with large wavevectors cannot be occupied. Because of the momentum conserving nature of N-scattering, the resulting phonon transport is similar to fluid flow and thus called hydrodynamic phonon transport. Fig. 2(b) shows the schematic of heat flux profile, similar to the molecular Poiseuille flow. When temperature increases further, U-scattering becomes significant and the thermal resistance is due to the direct momentum destruction by U-scattering. As U-scattering occurs in any location, the heat flux has a spatially uniform profile as shown in Fig. 2(c). (a)$Ballistic (b)$Hydrodynamic (c)$Diffusive Fig. 2. Schematic of phonon flux profile in ballistic, hydrodynamic, and diffusive regimes [3] The N- and U-scattering for phonons, since suggested by Peierls around a century ago [2], have been a foundation for the quantum theory of thermal transport in solids. Although the concept of N- and U-scattering were well accepted, the direct confirmation of N-scattering was still lacking. This led to the theoretical [4-10] and experimental efforts [11-15] for the prediction and observation of hydrodynamic phonon transport, namely phonon Poiseuille flow and second sound which will be discussed later in more detail. The phonon Poiseuille flow was first measured in solid He at the temperature range of 0.6 to 1.0 K [11]. The second sound was measured in solid 3He at 0.5 K [12], in NaF at around 15 K [13, 14], and in Bi at 2 K [15]. Those experimental observations combined with the theoretical studies directly confirm the N-scattering for phonons and show remarkably different effects of N- and U-scattering on thermal transport. This was regarded as "one of the great triumphs of the theory of lattice vibrations" [16]. Despite of the confirmation of hydrodynamic phonon transport, the study on this topic has been inactive for the recent several decades. As can be seen from the previous measurements, the hydrodynamic phonon transport was observed at very low and narrow temperature ranges and thus considered not relevant to practical applications. The conditions for hydrodynamic phonon transport are stringent because it is rare to satisfy the weak U-scattering and strong N-scattering at the same time. The U-scattering can be easily suppressed if temperature is much lowered than the Debye temperature so as to limit the phonon population to small wavevector states. However, if the temperature is lowered, there is not enough N-scattering events and the transport easily becomes ballistic. Thus, for hydrodynamic phonon transport to be significant, a material should exhibit a high Debye temperature and large anharmonicity at the same time. This is not common; a material with a high Debye temperature like diamond usually exhibits small anharmonicity. The quality of sample is another issue as the impurity scattering is the momentum-destroying scattering and weakens the hydrodynamic features. It is interesting to note that NaF was chosen for the second sound experiments [13, 14, 17] because Na and F are naturally monoisotopic elements and thus at least isotope impurity does not exist. The hydrodynamic phonon transport has recently received a renewed attention after first- principles-based studies predicted the significant hydrodynamic phonon transport in graphitic materials including single-wall carbon nanotubes (SWCNTs) [18], graphene [3, 19], and graphite [20]. Interestingly, those graphitic materials exhibit a high Debye temperature and large anharmonicity at the same time, leading to the strong N-scattering shown in Fig. 3 and the significant hydrodynamic phonon transport [3]. The light atomic mass of carbon and strong sp2 bonding result in the high Debye temperature and weak U-scattering. Also, the flexural phonon modes from its layered atomistic structure are largely anharmonic for small wavevector states [21], leading to the strong N-scattering.  Fig. 3. The mean free paths of N- and R-scattering in suspended graphene at 100 and 300 K from first-principles calculation. The primary objective of this book chapter is to provide a brief overview of basic concepts and recent studies of hydrodynamic phonon transport for those who have previously worked on ballistic and diffusive phonon transport. Other comprehensive review articles are available for advanced theoretical aspects [22, 23] and macroscopic governing equations of heat wave which is related to second sound [24]. This book chapter is organized as follows. Section II discusses the displaced Bose-Einstein distribution as an equilibrium distribution under N-scattering and collective hydrodynamic phonon flow. Section III summarizes the methods to solve the Peierls- Boltzmann transport equation for hydrodynamic and quasi-hydrodynamic phonon transport. Section IV provides our current understanding on the role of N-scattering for thermal resistance for various cases. Section V will review the theoretical and experimental studies of second sound. We then briefly discuss the future perspectives of phonon hydrodynamics in Section VI. We also would like to mention that the term 'hydrodynamic phonon transport' has been used in a different context in recent publications [25-31]. Those studies used phonon hydrodynamic equations that were derived assuming strong N-scattering compared to U-scattering and thus have a term similar to the viscous term of the Navier-Stokes equation [7, 8, 32]. However, to avoid any confusion, the phenomena studied in those studies are quasi-ballistic phonon transport and do not require the strong N-scattering; the hydrodynamic equations were used to phenomenologically describe the quasi-ballistic transport. In this book chapter, we focus on the hydrodynamic phenomena of phonon transport due to strong N-scattering and do not discuss the phenomenological hydrodynamic description of quasi-ballistic phonon transport. For readers who are interested in the latter topic, a recent review article can provide a comprehensive summary [33]. II. Collective Phonon Flow One unique feature of hydrodynamic transport that can be distinguished from ballistic and diffusive regimes is the collective motion of particles. The term 'collective' is often used to describe different phenomena in solid-state physics. Here, we call the transport of particles is collective when the flux of particles can be represented by a single value of velocity regardless of their quantum states. As an example, let us assume that we are able to track the movements of all molecules in a small fluid element. Assuming strong molecule-molecule scattering and small pressure gradient for the well-defined local equilibrium condition, the molecules then follow the displaced Boltzmann distribution: fB disp = ⎛ ⎝⎜ m 2πkBT 3/2 ⎞ ⎠⎟ exp − ⎛ ⎜ ⎜ ⎝ m v − u 2 2kBT ⎞ ⎟ ⎟ ⎠ (3) u , m kB , and v T is the actual velocity of a molecule and represent mass of a molecule, the Boltzmann constant and temperature, where respectively. The is the drift velocity. Note that the drift velocity is the same for all molecules regardless of their quantum states. Usually the actual velocity is much larger than the drift velocity, making the movement of each molecule looks like random. However, the small drift velocity causes a net flow of molecules. As a result, the fluid element containing many molecules that seemingly move along random direction can move with the drift velocity as a whole. Thus, we call the molecular transport collective in this case. Likewise, phonon particles show the collective motion when the transport is hydrodynamic. The equilibrium distribution of phonons with N-scattering is the displaced Bose-Einstein distribution: f disp = exp ! ω− q⋅u ( ⎡ ⎢ ⎢ ⎣ ⎛ ⎜ ⎝ kBT −1 ) ⎤ ⎞ ⎟ −1 ⎥ ⎠ ⎥ ⎦ (4) where q and u are the phonon wavevector and drift velocity (or displacement), respectively. In most cases where the transport is non-hydrodynamic, u differs for each phonon mode. However, in hydrodynamic regime u is a constant for all phonon modes. The displaced distribution function q⋅u ≪ω can be linearized assuming a small displacement, i.e., , ( )q⋅u f 0 +1 (5) f 0 f disp ≈ f 0 + ! kBT The fact that the displaced Bose-Einstein distribution function is the equilibrium distribution upon N-scattering can be shown with the Boltzmann's H-theorem [34]. For example, the rate of entropy generation upon coalescence three-phonon scattering is !Sscatt ∼ φi +φj −φk ( ∑ ijk )2 Pi, j k (6) k Pi, j is the equilibrium transition rate of the coalescence process where the phonon particles where at the state i and j are merged to the state k. A similar expression can be written for the decay 0 process. The φi fi distribution φi = fi − fi as distribution, ( 0 represents the deviation of distribution function from the stationary Bose-Einstein (i.e., displaced Bose-Einstein distribution with zero displacement), and is defined ( ) . If the three phonon states exhibit the displaced Bose-Einstein ) 0 +1 ) ( fi fi 0 φi +φj −φk = qi + q j − qk ( )⋅u qi + q j = qk (7) Considering the momentum conservation of N-scattering, , the entropy generation in this case is zero, verifying that the displaced Bose-Einstein distribution is an equilibrium distribution under N-scattering. From Eq. (7), even U-scattering ( ) does not generate any entropy if the reciprocal lattice vector . This was also shown through the simulation of second sound in a recent study [35]. qi + q j = qk ± Gm is orthogonal to Gm u Whether a certain scattering process is N- or U-scattering depends on the choice of the Brillouin zone, which may lead to confusion or misunderstanding about the role of N- and U- scattering on phonon transport. We would like to emphasize that the concept of momentum conservation for understanding the phonon transport is valid only when the crystal momentum is defined with the first Brillouin zone which is the Wigner-Seitz unit cell in reciprocal space. Otherwise, the displaced distribution function in Eq. (4) is incorrect and distinguishing N- and U- scattering based on the non-Wigner-Seitz unit cell is not meaningful. With the linearized form of the displaced distribution function in Eq. (5), it is straight forward to show that the phonon particle flux ′′nx = 1 NV ∑ i vx f disp = ⎛ ⎝⎜ ′′nx 1 NV can be described by the single value of u: vx ∑ i ! kBT f 0 ( ) q f 0 +1 ⎞ ⎠⎟ ⋅u (8) where N and V are the number of atoms and the volume of unit cell. Similarly, heat flux ′′qx is ′′qx = 1 NV ∑ i !ωvx f disp = ⎛ ⎝⎜ 1 NV ∑ i !ωvx ! kBT f 0 ( ) q f 0 +1 ⎞ ⎠⎟ ⋅u (9) u It is noteworthy that both particle flux and heat flux are linearly proportional to the local drift velocity , representing the collective motion of phonon particles. The coefficients in the parenthesis are constants determined by phonon dispersion and temperature. The fact that single value can describe the transport of all phonon particles is the basis for the macroscopic transport equation about from the displaced Bose-Einstein distribution to some extent: As U-scattering cannot be completely avoided, the actual phonon distribution deviates which will be discussed in the section III. u u fi ≈ fi 0 + ! kBT 0 fi ( fi 0 +1 ) qi ⋅u +δi ( ) (10) δ ( ) is where represents the deviation from the displaced Bose-Einstein distribution. It would be interesting to see how close the actual phonon distribution is to the displaced Bose-Einstein distribution in real materials in which hydrodynamic phonon transport is expected to be significant. In Fig. 4, we show the distribution function of phonon particles along the armchair direction in graphene at 100 K from the Peierls-Boltzmann transport equation (PBE) in an infinitely large sample case which will be discussed in the section III. In most cases where the transport is not and thus in Eq. (10) is larger compared to the collective part hydrodynamic, δi ) ) ( ( ) fi fi fi − fi 0 +1 qi,x with a constant slope, representing the collective motion of phonon particles nearly linear to with the same displacement regardless of the phonon mode. Fig. 5 shows the contribution of the collective motion of phonon particles to total heat flux in (20,20) SWCNTs. At low temperature below 100 K, most of heat is carried by the collective motion of phonon particles and the contribution of collective motion gradually decreases with temperature due to U-scattering. . However, in graphene, is not linear to qi ⋅u ( ( fi 0 ) 0 +1 0 fi − fi 0 ) ( 0 fi qi,x Fig. 4. Normalized deviational distribution, graphene at 100 K [3]. ( 0 fi − fi ) ( 0 fi ( ) 0 +1 fi ) , in an infinitely large Fig. 5. Contribution of collective motion of phonon particles to total heat flux in (20,20) SWCNT with naturally occurring 13C isotope content (1.1%) [18]. III. Peierls-Boltzmann Transport Equation The phonon distribution is described by the PBE: ∂ fi(t,x) ∂t + vi ⋅∇fi(t,x) = Gij f j d (11) j ∑ f j − f j 0 d f j is the deviational distribution function defined as and the G is the scattering where matrix. The original form of the PBE is known to be difficult to solve. The advection and scattering terms are in differential and integral forms and the unknown, , is a function in many dimensions including time, real space, and reciprocal space domains. The equation has been often simplified assuming steady state, a constant temperature gradient in an infinitely large sample, and very small deviation from the equilibrium distribution: ∑ vi ⋅∇T fi(t,x) (12) 0 dfi dT = Gij f j d j ) 0 dT The differential advection term in Eq. (11) was replaced with the spatially homogenous term, ( vi ⋅∇T dfi , by assuming the constant temperature gradient in an infinitely large sample. With these assumptions, the phonon distribution function is spatially homogenous except for the change due to temperature gradient. Then, the PBE could be simplified from the integro- differential equation to the homogenous integral equation which is relatively easier to solve. Recently developed ab initio framework of lattice dynamics made it possible to calculate the scattering matrix, G, from first principles [37, 38]. Also, several numerical techniques such as the full iterative method [39, 40] and the variational method [41] were developed to solve the Eq. (12). Solving Eq. (12) with ab initio phonon dispersion and scattering matrix showed an excellent predictive power for thermal conductivity of bulk samples [42]. For the hydrodynamic regime, however, the assumption of spatially homogenous distribution function is not valid. As schematically shown in Fig. 2(b), the heat flux and phonon distribution largely depends on the location in real space, and the advection term, ,in Eq. (11) cannot be homogenous. Also, the second sound is the temporal and spatial fluctuation of temperature field which requires the description under unsteady condition. Therefore, we would need to solve the PBE as an original form containing both differential and integral terms. We briefly review the past approaches several decades ago to solve the PBE with several assumptions, and also introduce recent approaches with minimal assumptions from first principles. One of the most challenging parts of solving the PBE is how to handle the integral scattering term. In the PBE, all phonon states are coupled to each other through the integral scattering term. Callaway suggested a simple form of scattering model from the fact that N- scattering and U-scattering tend to relax a phonon system to displaced and stationary Bose-Einstein distributions, respectively [43]. Although the Callaway's scattering model was from intuition without rigorous theoretical considerations, it was later shown that the model can be formally derived by ignoring the off diagonal terms of the N- and U-scattering matrices [44]. vi ⋅∇fi Early theoretical studies of phonon hydrodynamics derived macroscopic transport equations like the Navier-Stokes equation of fluid flow [4, 5, 7, 8, 32]. The work by Sussmann and Thellung [4] solved the PBE to the first order assuming no U-scattering and constructed momentum and energy balance equations. Some of Krumhansl group's work extended the transport equations to the case where U-scattering exists [5, 6]. The notable work by Guyer and Krumhansl [7, 8] solved the PBE in the eigenstate space of scattering operator which led to the concept of relaxon that will be discussed later. The derivation of these early studies were carefully examined and compared later by Hardy [32]. Although the details of derivation in the early studies are slightly different, they share the same basic idea. The idea is similar to how the Navier-Stokes equation is derived from the Boltzmann transport equation with the BGK scattering model which is analogous to the N-scattering term of the Callaway's scattering model. We briefly discuss the Sussmann and Thellung's derivation here. The momentum and energy balance equations can be simply derived from the PBE by taking momentum ( !q ) and energy ( !ω where ) as a moment of the PBE: ∂E ∂t + ∇αQα = 0 ∂Pα ∂t + ∇βpαβ = 0 E = Qα = Pα = pαβ = 1 NV 1 NV 1 NV 1 NV !ωi fi ∑ i !ωivα,i fi ∑ i !qα,i fi ∑ i !qα,ivβ,i fi ∑ i (13) (14) (15) (16) (17) (18) Qα α -direction. The are the energy density and heat flux along the The and are E -direction momentum density and the momentum flux along the -direction. Note that the α right-hand side of Eq. (13) and Eq. (14) are zero because total momentum and energy are conserved upon N-scattering. If U-scattering is considered, the momentum destroying term by U- scattering would appear in the momentum balance equation. In order to complete those momentum and energy balance equations, the phonon distribution function is required. The phonon distribution can be found by solving the PBE with the N-scattering term of Callaway's scattering model: and β pαβ Pα ∂ f ∂t + v ⋅∇f = − f − f disp τN (19)  and !f ≈ !f disp ∇f ≈ ∇f disp Eq. (19) can be further simplified if we assume . This assumption is analogous to the Chapman-Enskog expansion to the first order and is valid when N-scattering is strong [45]. To be more specific, N-scattering is considered strong when the relaxation time and mean free path of N-scattering are much smaller than the characteristic time and size of system (e.g., time period of temperature fluctuation for second sound and the sample size for steady-state heat flow). With such assumptions, it is straightforward to solve Eq. (19). Based on the phonon distribution function from Eq. (19) being plugged into Eq. (13) and Eq. (14), the following macroscopic governing equations can be derived: 1 3 ∇ ⋅u = 0 vg 2∇2 (20) !′T − ′T + 1 3 !uα + vg 2∇α ′T − vg 2τN ⎛ ⎝⎜ 2 5 ∇α∇ ⋅u + 1 5 ∇2uα ⎞ ⎠⎟ = 0 (21) f T0 ′T where is closer to where ( T − T0 is an equilibrium temperature. is the dimensionless deviational temperature defined as vg is the group velocity. The ) T0 Although early theoretical studies [4, 5, 7, 8, 32] are slightly different in the details of derivation, they are based on the same assumptions: i) N-scattering being much stronger than U- scattering such that , and ii) N-scattering being strong enough that !f ≈ !f disp . Because of these assumptions, the hydrodynamic equations derived in the early studies have several limitations. The macroscopic hydrodynamic equations may not accurately describe the following cases: i) the characteristic size of system being comparable to the mean free path of N-scattering, namely phonon transport in somewhere between ballistic and hydrodynamic limits, and ii) N-scattering being not much stronger than U-scattering, namely phonon transport in somewhere between diffusive and hydrodynamic limits. In addition, the validity of Callaway's scattering model is questionable for quantitative purposes [46, 47]. ∇f ≈ ∇f disp than and f disp f 0 As the full scattering matrix can now be calculated from first principles and the hydrodynamic phonon transport gained the renewed attention, there are two recently developed methods to solve the PBE with the full scattering matrix in both real and reciprocal spaces without the assumption of strong N-scattering. Both approaches provide a solution of the PBE without any significant assumptions and thus can be useful to study a complex transport phenomena where features of all three regimes exist to some extent [48]. The first approach is based on the eigenstates of the scattering matrix. The scattering matrix to Eq. (11) such where X i = !ωi kBT can be symmetrized by multiplying a factor, that the scattering matrix has an orthogonal set of eigenstates [44]: ⎞ ∑ ⎠⎟ vi ⋅∇fi = Gij * f j 2sinh 1 2 2sinh( X i / 2) X i ⎛ ⎝⎜ j d* (22) where d* f j is ⎛ ⎝⎜ 2sinh 1 2 X j d ⎞ ⎠⎟ f j and the scattering matrix, G*, is Gij * = ⎛ ⎜ ⎜ ⎜ ⎝ 2sinh 1 2 2sinh 1 2 X i X j ⎞ ⎟ ⎟ ⎟ ⎠ Gij (23) The orthogonal eigenstates of G* are later called relaxons [49]. The solution of the PBE, , then can be expressed as a linear combination of relaxons and the equation for the coefficient (population) of each relaxon state can be derived from the PBE [49]. An advantage of the relaxon framework is that relaxon has now a well-defined relaxation length and thus the thermal transport can be described with a simple kinetic description of relaxon particles. Phonons, if experience strong N-scattering, do not have well-defined relaxation length due to the complex interplay between N- and U-scattering processes and also its collective nature of motions. d*f The second approach employs the Monte Carlo (MC) method to solve the PBE with the full scattering matrix [50, 51]. The MC method was previously developed to solve the PBE with the single mode relaxation time approximation (SMRT) for studying quasi-ballistic phonon transport [52-54]. The MC method with the SMRT stochastically determines the occurrence of scattering based on the probability of scattering. With the full scattering matrix, the MC method stochastically determines whether a certain scattering process occurs or not and the final state of phonon particles if the scattering is determined to occur. The energy-based PBE is chosen over the regular PBE due to its advantage of strict energy conservation: ) ( Bij ωj f j vi ⋅∇ ωi f )i = (24) ∑ ( d j is the scattering matrix of the energy-based PBE, defined as Bij where exchange upon scattering is described as ( d t + Δt ( Zij Δt where the energy propagator matrix Z can be found as ( Z Δt ωi fi ) = ∑ j ) = eBΔt (26) If the off-diagonal terms of matrix B is ignored and only diagonal terms are considered, Eq. (25) is recovered to the exponential decay of energy which is equivalent to the SMRT: ωi fi ( d t + Δt Bii (27) Note from Eq. (24). The scattering in Eq. (25) describes the transfer of energy from phonon state j to i. In MC simulation, the destination state i can be stochastically determined and its detailed MC algorithm can be found in literatures [50, 51]. is the same as −1 −τi ) = exp(BiiΔt)ωi fi d t( ) ( ωi ωj )Gij . The energy )ωj f j d t( ) (25) IV. Steady-State Phonon Hydrodynamics The N-scattering itself does not directly cause thermal resistance because of its momentum conserving nature. However, the N-scattering can affect thermal resistance when combined with momentum-destroying scattering (R-scattering or diffuse boundary scattering) or thermal reservoirs that emit phonons of which distribution deviates from the displaced Bose-Einstein distribution. These situations are common in practical systems. We discuss the role of N-scattering for thermal resistance in three different cases: i) an infinitely large sample, ii) a sample with an infinite length but a finite width where diffuse boundary scattering destroys the phonon momentum along the flow direction, and iii) a sample with an infinite width but a finite length contacting hot and cold reservoirs that emit phonons with the stationary Bose-Einstein distribution. IV.1. Infinitely Large Sample It is well known that the thermal conductivity is infinitely large when the N-scattering is the only scattering mechanism and the sample is infinitely large. Assuming that a local temperature gradient is applied and phonon flow is initiated, the phonons subsequently establish the displaced Bose-Einstein distribution through many N-scattering events. Then, the N-scattering does not further alter the displaced Bose-Einstein distribution and the phonons can continue to flow even without any temperature gradient, resulting in the infinite thermal conductivity. This leads to the simple statement that N-scattering itself does not cause thermal resistance. This simple statement, however, is true only when the distribution function is homogenous in space as in the infinitely large sample. If there is a significant spatial variation of the distribution function, the N-scattering can cause thermal resistance. This will be discussed in IV.3. Even when the distribution function is homogenous in space, N-scattering contributes to thermal resistance if U-scattering also exists. In general, phonon states with a small wavevector have very weak U-scattering. However, the small wavevector phonons can be scattered into larger wavevector states through N-scattering and then can be seen by U-scattering. A recent study on the thermal transport in SWCNTs [55] quantitatively shows the effect of N-scattering on thermal conductivity. The thermal conductivity of (10,10) SWCNT is 10,000 W/m-K when only U- scattering is considered, but it is significantly reduced to 2,000 W/m-K when both N- and U- scattering processes are included. IV.2. Sample with an Infinite Length and a Finite Width We consider a sample with an infinite length and a finite width to discuss the thermal resistance when N-scattering is combined with diffuse boundary scattering. As shown in Fig. 6, we consider a constant temperature gradient along the length direction, which drives the phonon flow. Fig. 6. Schematic of phonon flow in an infinitely long sample with a finite width The major mechanisms of thermal resistance in diffusive and ballistic regimes are U- scattering and diffuse boundary scattering, respectively. In the hydrodynamic regime, we have a different mechanism for thermal resistance: viscous damping effect which is a result of combined N- and diffuse boundary scattering. The drift velocity near boundaries is smaller than that in the middle of a sample due to diffuse boundary scattering. Thus, the drift velocity exhibits a gradient along the transverse direction (y-direction in Fig. 6). Due to the drift velocity gradient, phonon momentum is transferred from the middle of the sample to the boundaries through N-scattering processes and then finally is destroyed by the diffuse boundary scattering. The viscous damping term can be seen in the second-order derivative term in Eq. (21). discussed in section III, an expression for the phonon hydrodynamic viscosity ( [51]: Based on the momentum balance equation from the PBE with its first-order solution ) can be derived µph 0 i∑ qx 2vy,i 2 fi i∑ qxvx,i fi 0 ( ( fi 0 +1 fi 0 +1 )τN,i )ωi µph = (28) A notable difference between ballistic and hydrodynamic regimes is that the momentum transfer to the boundary in hydrodynamic regime is impeded by N-scattering. As N-scattering rate is increased, the rate of momentum transfer to the boundary, which determines the extent of viscous damping, is decreased. This can be seen in the phonon hydrodynamic viscosity as a function of temperature in Fig. 7. As temperature increases, the N-scattering rate is increased, resulting in the lower hydrodynamic phonon viscosity. The extent of viscous damping also depends on the width of sample as indicated in the second-order derivative term in Eq. (21). The rate of momentum transfer rate in hydrodynamic regime is proportional to 1/W2 where W is the width of a sample, while the rate in ballistic regime is proportional to 1/W. ] s [ y t i s o c s v n o n o h P i 10-9 10-10 10-11 50 100 200 Temperature [K] 300 Fig. 7. Temperature dependence of phonon hydrodynamic viscosity of suspended graphene calculated with phonon dispersion and scattering rates from first-principles calculation [51]. The viscous damping effect of hydrodynamic regime causes peculiar dependences of thermal conductivity on temperature and sample width, which are distinguished from the ballistic and diffusive cases. In the ballistic regime, the thermal conductivity is linearly proportional to the sample width. The thermal conductivity of diffusive regime is constant regardless of a sample width. However, the thermal conductivity of hydrodynamic regime superlinearly increases with the sample width due to the viscous damping that decreases as W2. In addition, the thermal conductivity of hydrodynamic regime increases with temperature much faster than that of ballistic regime as the viscous damping is weakened as temperature increases. The peculiar dependence of thermal conductivity on temperature was observed in solid He at low temperature, verifying the existence of phonon Poiseuille flow [11]. Recently these dependences have been predicted at much higher temperature in graphene [3, 51, 56] and graphite [20], and experimentally observed in SrTiO3 [57]. The peculiar dependences of thermal conductivity on temperature and sample width can be observed only when the actual transport phenomena are close to those in the ideal hydrodynamic regime without U-scattering. The thermal transport in graphitic materials at intermediate temperature above 100 K can exhibit all three different mechanisms of thermal resistance: U- scattering, direct diffuse boundary scattering, combined diffuse boundary and N-scattering. The significance of each mechanism can be evaluated using the momentum balance. The temperature gradient in Fig. 6 drives phonon flow and generates excess phonon momentum ( ). This momentum is balanced by momentum destructions by three different mechanisms: diffuse boundary scattering without internal phonon scattering (i.e, ballistic effect, ), diffuse boundary scattering combined with N-scattering (i.e., viscous damping or hydrodynamic effect, ), and ). The momentum balance direct momentum destruction by U-scattering (i.e., diffusive effect, can be expressed as Φ∇T ΦD ΦB ΦH Φ∇T = ΦB + ΦH + ΦD (29) T 2.03 Fig. 8(a) shows that thermal conductivity of an infinitely long graphene has a temperature dependence of when temperature is below 90 K, much larger than that of the ballistic case T 1.68 . This temperature range agrees with the momentum balance analysis of the same sample, shown in Fig. 8(b). It is clear that below 90 K, is the major mechanism of the momentum destruction, indicating that viscous damping is significant at this condition [48]. ΦH (a) ] - K m W / [ y t i v i t c u d n o c l a m r e h T 105 104 ~T2.03 W=10µm 103 50 1.0 0.8 0.6 0.4 0.2 ΦB ΦH ΦD W=10µm (b) n o i t c u r t s e d m u t n e m o m d e z i l a m r o N 100 200 Temperature [K] infinitely long graphene sample with the width of 10 µm from the MC solution of the PBE with Fig. 8 Temperature dependence of (a) thermal conductivity, and (b) the momentum balance in an Temperature [K] 300 100 150 200 250 400 0.0 50 ab initio full three-phonon scattering matrix. IV.3. Sample with an Infinite Width and a Finite Length Contacting Hot and Cold Reservoirs When an infinitely wide sample contacts hot and cold reservoirs as in Fig. 9, the phonons emitted from the reservoirs do not follow the displaced Bose-Einstein distribution. They follow a Bose-Einstein distribution distorted by a spectral transmission function at the interface between the sample and the reservoir. The N-scattering processes change this non-displaced Bose-Einstein distribution (i.e., non-collective) to the displaced Bose-Einstein distribution (i.e., collective). As entropy is always generated when the distribution function is changed by scattering processes as shown in Eq. (6), N-scattering causes thermal resistance near the interface between the sample and the reservoir where the emitted phonon flow becomes collective. The region where the thermal resistance occurs is within the order of the mean free path of N-scattering from the boundary. Fig. 10 shows the formation of collective phonon flow at the cost of temperature drop near the boundaries, resulting in the thermal resistance by N-scattering. Assuming N-scattering is the only scattering mechanism, the N-scattering far from the boundaries does not cause any temperature drop as the distribution function is already the displaced Bose-Einstein distribution. The thermal resistance due to the transition between non-collective and collective phonon flows depends on materials. Fig. 11 compares three-dimensional Debye phonon dispersion and graphite in terms of the reduction of phonon heat flux by N-scattering from the purely ballistic case. For the 3D Debye case, the reduction of heat flux is relatively small; the heat flux reduction by N-scattering is only around 5% for all three temperatures, 100, 200, and 300 K. However, for graphite, the reduction of heat flux is substantial; the heat flux is reduced by 20, 30, and 40% at 100, 200, and 300 K, respectively. Other graphitic materials such as SWCNTs and graphene show similar reduction of heat flux. Hot reservoir T+ D 0T Sample phonon flow Cold reservoir T- D 0T x = 0 ( f T 0 0 T + D ) xf + = = 0 x L= ( f T 0 0 - = = x L f T - D ) Fig. 9 Schematic of sample geometry contacting hot and cold reservoirs (cid:62) (cid:44) (cid:60) (cid:1) (cid:70) (cid:83) (cid:86) (cid:85) (cid:66) (cid:83) (cid:70) (cid:81) (cid:78) (cid:70) (cid:85) (cid:1) (cid:77) (cid:66) (cid:79) (cid:80) (cid:74) (cid:85) (cid:66) (cid:87) (cid:70) (cid:69) (cid:74) 5×10-4 0 -5×10-4 0 8×10-4 6×10-4 4×10-4 2×10-4 0 20 5 10 15 Position (µm) d r i f t v e o c i t y , l u x [ m / s ] Fig. 10 The profile of deviational temperature, defined as the difference between local temperature and global equilibrium temperature, and the drift velocity. The sample is (20,20) SWCNT contacting hot and cold reservoirs that have the deviational temperature of 0.001 and -0.001 K, respectively. The profile is calculated by Monte Carlo method of the PBE assuming the Callaway's scattering model. The rate of N-scattering is assumed 1010 s-1 and U-scattering is ignored [58]. (a) 1.0 (b) 1.0 B ' ' q / H ' ' q 0.8 0.6 Debye 0.4 0.1 100K 200K 300K 10 1 Kn-1 B ' ' q / H ' ' q 0.8 0.6 0.4 0.1 graphite 1 Kn-1 10 ) and without any internal scattering ( ′′qB Fig. 11 The ratio between heat flux with N-scattering ( ) as a function of inverse Knudsen number in (a) three-dimensional Debye model and (b) graphite. The heat flux is calculated with the Monte Carlo solution of the PBE with the Callaway's scattering model. The rate of N-scattering is assumed 1010 s-1 and U-scattering is ignored [58]. Hq¢¢ The large thermal resistance by N-scattering for graphitic materials can be explained with their non-linear phonon dispersion with many phonon branches. The rate of entropy generation due to scattering, Eq. (6) can be written as follows assuming the Callaway's scattering model and the stationary Bose-Einstein distribution for phonons emitted from the reservoirs. ) ( )ωi qx,i vx,i * − ′ux (30) 0 +1 ∑ "2 ( fi fi 0 2 !Sscatt = ΔT T ⎛ ⎝⎜ ⎞ ⎠⎟ τNkBT 2NV i vx,i is * ωi qx,i where from the momentum conservation: and ′ux is the drift velocity per temperature difference and can be found  ′ux = i ∑ qx,i ωi fi ( ∑ qx,i 2 fi 0 i 0 ( fi ) fi 0 +1 ) 0 +1 (31) The Eq. (30) shows that the temperature difference of two reservoirs drives the phonon flow with the displacement of ′ux is a constant for all phonon modes (e.g., one-dimensional which may vary depending on phonon modes while the drift velocity is the same for all phonon modes. If vx,i * * vx,i Debye phonon dispersion), * vx,i is the same as ′ux by Eq. (31) and the entropy generation would be zero. However, if * vx,i significantly varies with phonon states, the entropy generation is expected to be large. For the three-dimensional Debye model, the * vx,i wavevector and thus causes small thermal resistance as can be seen in Fig. 11(a). The ratio in this case does not change with temperature as the variance of varies with the direction of phonon ′′qB is associated with the direction Hq¢¢ / * vx,i is significant of phonon wavevector only. For graphtic materials, however, the variance of compared to the Debye model as a result of non-linear dispersion with many branches, resulting in the large thermal resistance that depends on temperature in Fig. 11(b). This indicates the resistance due to the transition between non-collective and collective phonon flows is determined by the shape of phonon dispersion. vx,i * V. Unsteady Phonon Hydrodynamics (Second Sound) The fundamental difference between N- and U-scattering in terms of momentum conservation leads to the different response upon temporal perturbation to a phonon system. One simple form of the perturbation is a heat pulse being applied to one end of a sample as shown in Fig. 12. The heat pulse causes the increased local phonon density and the response of phonon system is largely different depending on the transport regime. For the diffusive regime, the energy balance equation with the Fourier's law indicates that the peak position of heat pulse cannot move forward and remains at its original location. Then, the thermal energy of the heat pulse diffuse into the sample and finally the sample reaches an equilibrium with a slightly elevated temperature for the entire region. For the ballistic regime, the heat pulse can propagate through the sample as there is no phonon scattering. However, the shape of heat pulse can spread out in space unless all phonon modes have the same group velocity along the heat pulse propagation direction. This is expected to be particularly significant in graphitic materials where flexural phonon modes with a quadratic dispersion are important for thermal energy transport. For the hydrodynamic regime, the heat pulse leads to the local fluctuation of temperature field which can propagate as a wave through the sample. An analogous phenomenon in fluid system is the propagation of pressure pulse in space, which is acoustic sound. From the similarity of the two phenomena, the temperature pulse propagation in the form of wave in hydrodynamic regime is called second sound. Fig. 12. Propagation of a heat pulse in diffusive and hydrodynamic regimes [3] The second sound was first studied with superfluid He II in which phonon is an elemental excitation of the system. The speed of second sound in liquid He was predicted by Landau using the two fluid theory of rotons and phonons [59] and later confirmed by an experiment [60]. The predicted temperature wave was named as second sound by Landau in order to distinguish it from the first sound which is the ordinary acoustic sound (i.e., pressure wave propagation). Later it was shown that the same speed of second sound can be directly derived by using phonon gas model without roton [61, 62], which motivated the studies of second sound in crystalline solids. The second sound in solids can be observed with two different methods: heat-pulse experiment [12-15, 63] and light scattering method [10, 64-69]. In the heat pulse experiment, a heat pulse was applied to one end of a few mm long sample and the temperature to the opposite end was recorded as a function of time. At sufficiently low temperature such that internal phonon- phonon scattering is negligibly weak, two peaks of temperature pulse were observed, each of which represents the ballistic transport of transverse and longitudinal phonons. No significant dispersion of the temperature peak was observed because three dimensional bulk materials where long wavelength phonons have a linear dispersion relation were used. With slightly increased temperature (around 15 K for NaF [14]), another peak in addition to those two peaks were observed. The delay time of the new peak agree well with the predicted speed of second sound and the third peak was considered second sound. As temperature is further increased, the third peak disappears, indicating that U-scattering becomes significant. The light scattering method measures the inelastic light scattering by a local change of dielectric constants due to second sound wave. A challenge lies in very weak coupling between light and thermal fluctuation at low temperatures. To solve this problem, a relatively strong thermal fluctuation field was induced by an optical grating method and the second sound in NaF was successfully measured [67]. The measured speed of second sound agrees well with that from the previous heat pulse experiments. Later, the light scattering measurements were carried out without inducing thermal fluctuation field for SrTiO3. The SrTiO3 has soft transverse optical phonons with small wavevector that are strongly anharmonic and thus cause strong N-scattering [68, 69]. The measured spectrum at around 30 K exhibits a doublet with a frequency shift (~ 20 GHz) that is comparable to the expected frequency of second sound in this temperature range. As the conditions for the clear observation of second sound is narrow in the variable space, the second sound measurements critically require a priori knowledge on the wavelength and frequency of second sound as well as the speed of second sound. The wavelength and frequency of second sound are determined by the mean free path and scattering rate of N- and U-scattering processes. If the pulse duration is much longer than the rate of U-scattering, the pulse can be destroyed by the U-scattering and thus cannot propagate as a second sound. If the pulse duration is much shorter than the rate of N-scattering, phonons will travel with their own group velocity and do not have a chance to establish the collective motion because of the lack of N-scattering. In this case, the pulse also cannot maintain its original shape and the thermal energy smears out. vII = vg 3 The speed, frequency, and wavelength of second sound were theoretically studied by calculating the dispersion relation of second sound. The speed of second sound was derived for the simplest case where Debye phonon dispersion is assumed and there is no U-scattering, giving the well-known relation for the speed of second sound, is the speed of second sound [61, 62]. The speed of second sound was also derived for more realistic phonon dispersion consisting of one longitudinal and two degenerate transverse acoustic branches, all having the Debye-type dispersion [4]. Later theoretical studies considered the possible mechanisms for attenuation of second sound and predicted the possible second sound frequency ranges [5-7]. In the literature, two different types of second sound called drifting and driftless second sounds were discussed [22, 70]. The driftless second sound differs from the second sound we discuss and does not require strong N-scattering; it occurs when all eigenstates of scattering operator have the similar relaxation time such that collective-looking thermal transport can occur. To our best knowledge, there was no experimental observation of the driftless second sound. , where vII The dispersion relation of second sound can be derived from the momentum and energy balance equations in Eq. (13) and Eq. (14). If U-scattering is considered, the momentum destruction by U-scattering needs to be added to the right hand side of Eq. (14). An example dispersion relation of second sound in (20,20) SWCNT is shown in Fig. 13(a). The real and imaginary frequencies represent the propagation and attenuation of a pulse, respectively. The imaginary frequency in the limit of small wavevector (i.e., long wavelength) is mostly determined by the rate of U-scattering. As the wavevector is increased (i.e., wavelength becomes shorter), the viscous damping effect by N-scattering becomes strong, causing the significant attenuation of second sound. Fig. 13(b) shows the required length of a sample. For the second sound to propagate, the sample length should be larger than the wavelength of second sound but smaller than its Im Ω( ) relaxation length defined as are the speed of second sound and the imaginary part of second sound frequency, respectively. vIIIm Ω( where ) vII and (a) ) z H ( y c n e u q e r f d n u o s d n o c e S 1010 109 108 Real Imaginary T=100 K (20,20) SWCNT 105 106 Second sound wavevector (m-1) (b) ) m ( h t g n e L 10-4 10-5 10-6 10-7 Relaxation Length Wavelength T=100 K (20,20) SWCNT 105 106 Second sound wavevector (m-1) Fig. 13. The propagation and attenuation of second sound in (20,20) SWCNT. (a) the dispersion relation of second sound showing propagation (real) and attenuation (imaginary) of second sound. (b) the comparison between relaxation length and wavelength of second sound. [18] VI. Summary and Future Perspectives In this chapter, we briefly reviewed the past and recent studies on hydrodynamic phonon transport. We first discussed the displaced Bose-Einstein distribution representing the collective motion of phonon particles as an equilibrium state under N-scattering. Then, we introduced several approaches to solve the Peierls-Boltzmann transport equation for the case where N-scattering is significant. Based on the solution of the Peierls-Boltzmann transport equation, we then showed how N-scattering affects thermal phonon transport in both steady state and transient cases. For the steady state cases, we discuss three-scenarios; when N-scattering is combined with i) U-scattering, ii) diffuse boundary scattering, and iii) thermal reservoirs that emit phonons following non- displaced Bose-Einstein distribution functions. In all cases, N-scattering affects thermal transport indirectly. For the first case where N-scattering is combined with U-scattering, it transfers energy from small wavevector states where U-scattering is relatively weak to large wavevector states where U-scattering is strong, thereby contributing to thermal resistance. For the second case, the N-scattering impedes the momentum transfer to the boundaries which acts as a momentum sink by diffuse boundary scattering. We discussed that stronger N-scattering leads to less viscous damping effect and larger thermal conductivity. For the last case, the N-scattering itself causes thermal resistance when the distribution function is not homogenous in space due to thermal reservoirs emitting phonons with non-displaced distribution. The thermal resistance occurs while those non-collective phonon flows become collective through N-scattering processes. The thermal resistance by the transition between collective and non-collective phonon flows depends on the shape of phonon dispersion; while the thermal resistance due to this effect is small for Debye phonon dispersion, it can be significant in graphitic materials because of their highly non-linear phonon dispersion with many branches. The second sound was discussed as a representative phenomenon of phonon hydrodynamics in transient case. The N-scattering causes the damping of second sound even without U-scattering. If the fluctuation of temperature field is fast in time and space domains, for example the frequency and wavelength of second sound are shorter than the rate and mean free path of N-scattering respectively, the fluctuation can be largely damped. Thus, the N-scattering imposes the limit of frequency and wavelength of second sound for its propagation. Although the recently developed ab initio framework for phonon transport has been proved for its high accuracy and predictive power [42], the significant hydrodynamic phonon transport in graphitic materials need to be experimentally confirmed. The significant contributions from the flexural phonon modes to thermal transport were experimentally shown [71], but its strong N- scattering due to extremely large anharmonicity for small wavevector states has not been verified. The explicit observation of hydrodynamic phonon transport has several challenges. First, the measurements need to be done in much smaller length and time scale compared to the previous studies performed several decades ago. The characteristic length and time scale of hydrodynamic phonon transport scales with the mean free paths and rate of internal phonon scattering. As those previous studies measured the hydrodynamic phonon transport at extremely low temperature below 15 K, the internal phonon scattering was weak; therefore the Poiseuille flow was measured with several mm size sample [11] and the second sound propagation was measured with the time scale of µs [14]. However, as the hydrodynamic phonon transport in graphitic materials is expected to occur at much higher temperature, the internal phonon scattering is accordingly strong. Thus, the experiments need to be done with sub-mm size samples and ns temporal resolution. Recent advancements on the microscale platform for the measurement of thermal conductivity [72, 73] as well as the ultrafast spectroscopy technique [74-76] are perhaps well suited for the measurement of hydrodynamic phonon transport in graphitic materials. Second, we would need a large sample with minimal defects. The ab initio simulation shows that the sample size should be at least 10 µm for measuring phonon Poiseuille flow and second sound at 100 K [51], but typical graphitic material samples with this sample size contain many defects. Interestingly, the observation of second sound was reported very recently using a highly oriented pyrolytic graphite (HOPG) sample [77]. This study used the transient grating method to generate the standing wave of second sound and could measure the fluctuation of temperature which is associated with the second sound. As the second sound is in standing-wave form in this study, it does not need to propagate throughout the entire sample and could be measured with less significant damping. The observation of phonon Poiseuille flow is expected to be more challenging compared to the second sound case. The theoretical prediction of phonon Poiseuille flow assumed infinitely long samples for the condition of fully developed phonon flow [51]. If a sample has a finite length, there would be so called entrance effect which is due to the transition from spatially uniform phonon flow to parabolic phonon flow near the entrance. This would require a sample with length being much larger than width. In the previous study, the phonon Poiseuille flow was predicted with the width of 10 µm, thus length should be much longer than this value. The recent prediction of significant hydrodynamic phonon transport indicates that the hydrodynamic regime is practically important for high thermal conductivity materials where N- scattering is often strong and cannot be ignored. Although the clear observation of hydrodynamic phonon transport is expected at sub-room temperatures, the hydrodynamic phonon transport is still important for understanding the thermal transport. As shown in Fig. 3, the mean free path of N- scattering and U-scattering have a large gap in the length ranges from sub-µm to µm for 300 K. If the sample size lies in this gap which is common in the practical applications of high thermal conductivity materials for thermal management, the diffusive-ballistic phonon transport may not correctly describe the thermal transport phenomena; it ignores the thermal resistance due to the momentum transfer and formation of collective phonon flow by N-scattering. Therefore, the hydrodynamic regime needs to be considered another limit of thermal transport in addition to ballistic and diffusive limits which were extensively studied in the past [78, 79]. The detailed mechanisms of how N-scattering contributes to thermal resistance when combined with other scattering processes have not been rigorously discussed in the past. We think this is partially because of the lack of available numerical tools; it has been very challenging to solve the Peierls- Boltzmann transport equation in both real and reciprocal spaces with minimal assumptions. With the recently developed ab initio frameworks for solving the Peierls-Boltzmann transport equation in both real and reciprocal spaces [49, 51], it is now possible to quantitatively study the influence of N-scattering on the overall thermal transport process when it is combined with other scattering processes. This would complete the understanding of phonon transport in high thermal conductivity materials and lead to the better design of thermal devices using those high thermal conductivity materials. Acknowledgement We acknowledge support from National Science Foundation (Award No. 1705756 and 1709307). References [1] W. Nernst, "Die theoretischen grundlagen des neuen wärmesatzes", Knapp, Halle (1917). [2] R. Peierls, "Zur kinetischen Theorie der Wärmeleitung in Kristallen", Annalen der Physik 395 (1929). [3] S. Lee, D. Broido, K. Esfarjani, and G. Chen, "Hydrodynamic phonon transport in suspended graphene", Nature Communications 6 (2015). [4] J. A. Sussmann, and A. Thellung, "Thermal conductivity of perfect dielectric crystals in the absence of Umklapp processes", Proc. Phys. Soc. 81 (1963). [5] E. W. Prohofsky, and J. A. Krumhansl, "Second-Sound Propagation in Dielectric Solids", Physical Review 133 (1964). [6] R. A. Guyer, and J. A. Krumhansl, "Dispersion Relation for Second Sound in Solids", Physical Review 133 (1964). [7] R. A. Guyer, and J. A. Krumhansl, "Thermal conductivity, second Sound, and phonon hydrodynamic phenomena in nonmetallic crystals", Physical Review 148 (1966). [8] R. A. Guyer, and J. A. Krumhansl, "Solution of the linearized phonon Boltzmann equation", Physical Review 148 (1966). [9] R. Gurzhi, "Thermal conductivity of dielectrics and ferrodielectrics at low temperatures", Journal of Experimental and Theoretical Physics 46 (1964). [10] A. Griffin, "On the detection of second sound in crystals by light scattering", Physics Letters 17 (1965). [11] L. Mezhov-Deglin, "Measurement of the thermal conductivity of crystalline He4", J. Exp. Theor. Phys. 49 (1965). [12] C. C. Ackerman, and W. C. Overton, "Second sound in solid helium-3", Phys. Rev. Lett. 22 (1969). [13] T. F. McNelly, S. J. Rogers, D. J. Channin, R. J. Rollefson et al., "Heat pulses in NaF: onset of second sound", Physical Review Letters 24 (1970). [14] H. E. Jackson, C. T. Walker, and T. F. McNelly, "Second sound in NaF", Physical Review Letters 25 (1970). [15] V. Narayanamurti, and R. C. Dynes, "Observation of second sound in bismuth", Phys. Rev. Lett. 28 (1972). [16] N. Ashcroft, and N. Mermin, Solid State Physics (Brooks and Cole, 1976). [17] H. E. Jackson, and C. T. Walker, "Thermal Conductivity, Second Sound, and Phonon- Phonon Interactions in NaF", Physical Review B 3 (1971). [18] S. Lee, and L. Lindsay, "Hydrodynamic phonon drift and second sound in a (20,20) single- wall carbon nanotube", Physical Review B 95 (2017). [19] A. Cepellotti, G. Fugallo, L. Paulatto, M. Lazzeri, F. Mauri, and N. Marzari, "Phonon hydrodynamics in two-dimensional materials", Nature Communications 6 (2015). [20] Z. Ding, J. Zhou, B. Song, V. Chiloyan, M. Li, T.-H. Liu, and G. Chen, "Phonon Hydrodynamic Heat Conduction and Knudsen Minimum in Graphite", Nano Letters 18 (2018). [21] P. K. Schelling, and P. Keblinski, "Thermal expansion of carbon structures", Physical Review B 68 (2003). [22] H. Beck, P. F. Meier, and A. Thellung, "Phonon hydrodynamics in solids", physica status solidi (a) 24 (1974). [23] V. L. v. Gurevich, Transport in phonon systems (Elsevier Science Publishers, Amsterdam, 1986), Modern Problems in Condensed Matter Sciences. [24] D. D. Joseph, and L. Preziosi, "Heat waves", Reviews of Modern Physics 61 (1989). [25] F. X. Alvarez, D. Jou, and A. Sellitto, "Phonon hydrodynamics and phonon-boundary scattering in nanosystems", JOURNAL OF APPLIED PHYSICS 105 (2009). [26] A. Ziabari, P. Torres, B. Vermeersch, Y. Xuan et al., "Full-field thermal imaging of quasiballistic crosstalk reduction in nanoscale devices", Nature Communications 9 (2018). [27] P. Torres, A. Ziabari, A. Torelló, J. Bafaluy et al., "Emergence of hydrodynamic heat transport in semiconductors at the nanoscale", Physical Review Materials 2 (2018). [28] Y. Guo, and M. Wang, "Phonon hydrodynamics for nanoscale heat transport at ordinary temperatures", Physical Review B 97 (2018). [29] Y. Dong, B.-Y. Cao, and Z.-Y. Guo, "Generalized heat conduction laws based on thermomass theory and phonon hydrodynamics", Journal of Applied Physics 110 (2011). [30] V. A. Cimmelli, A. Sellitto, and D. Jou, "Nonlinear evolution and stability of the heat flow in nanosystems: Beyond linear phonon hydrodynamics", Physical Review B 82 (2010). [31] A. Sellitto, F. X. Alvarez, and D. Jou, "Temperature dependence of boundary conditions in phonon hydrodynamics of smooth and rough nanowires", Journal of Applied Physics 107 (2010). [32] R. J. Hardy, and D. L. Albers, "Hydrodynamic approximation to the phonon Boltzmann equation", PHYSICAL REVIEW B 10 (1974). [33] Y. Guo, and M. Wang, "Phonon hydrodynamics and its applications in nanoscale heat transport", Physics Reports 595 (2015). [34] J. M. Ziman, Electrons and Phonons: the Theory of Transport Phenomena in Solids (Oxford University Press, UK, 1960). [35] Z. Ding, J. Zhou, B. Song, M. Li, T.-H. Liu, and G. Chen, "Umklapp scattering is not necessarily resistive", Physical Review B 98 (2018). [36] A. A. Maznev, and O. B. Wright, "Demystifying umklapp vs. normal scattering in lattice thermal conductivity", (2013). [37] D. A. Broido, M. Malorny, G. Birner, N. Mingo, and D. A. Stewart, "Intrinsic lattice thermal conductivity of semiconductors from first principles", Applied Physics Letters 91 (2007). [38] K. Esfarjani, G. Chen, and H. T. Stokes, "Heat transport in silicon from first-principles calculations", Physical Review B 84 (2011). [39] M. Omini, and A. Sparavigna, "Beyond the isotropic-model approximation in the theory of thermal conductivity", Physical Review B 53 (1996). [40] M. Omini, and A. Sparavigna, "Heat transport in dielectric solids with diamond structure", Nuovo Cimento D Serie 19 (1997). [41] G. Fugallo, M. Lazzeri, L. Paulatto, and F. Mauri, "Ab initio variational approach for evaluating lattice thermal conductivity", Phys. Rev. B 88 (2013). [42] L. Lindsay, C. Hua, X. L. Ruan, and S. Lee, "Survey of ab initio phonon thermal transport", Materials Today Physics 7 (2018). [43] J. Callaway, "Model for lattice thermal conductivity at low temperatures", Physical Review 113 (1959). [44] J. A. Krumhansl, "Thermal conductivity of insulating crystals in the presence of normal processes", Proc. Phys. Soc. 85 (1965). [45] W. G. Vincenti, and C. H. Kruger, Introduction to physical gas dynamics (Wiley, New York, 1965). [46] P. B. Allen, "Improved Callaway model for lattice thermal conductivity", Physical Review B 88 (2013). [47] J. Ma, W. Li, and X. Luo, "Examining the Callaway model for lattice thermal conductivity", Physical Review B 90 (2014). [48] X. Li, and S. Lee, "Crossover of ballistic, hydrodynamic, and diffusive phonon transport in suspended graphene", Physical Review B 99 (2019). [49] A. Cepellotti, and N. Marzari, "Thermal Transport in Crystals as a Kinetic Theory of Relaxons", Physical Review X 6 (2016). [50] C. D. Landon, and N. G. Hadjiconstantinou, "Deviational simulation of phonon transport in graphene ribbons with ab initio scattering", Journal of Applied Physics 116 (2014). [51] X. Li, and S. Lee, "Role of hydrodynamic viscosity on phonon transport in suspended graphene", Physical Review B 97 (2018). [52] A. Majumdar, "Microscale Heat Conduction in Dielectric Thin Films", Journal of Heat Transfer 115 (1993). [53] J.-P. M. Péraud, and N. G. Hadjiconstantinou, "Efficient simulation of multidimensional phonon transport using energy-based variance-reduced Monte Carlo formulations", Physical Review B 84 (2011). [54] Q. Hao, G. Chen, and M.-S. Jeng, "Frequency-dependent Monte Carlo simulations of phonon transport in two-dimensional porous silicon with aligned pores", JOURNAL OF APPLIED PHYSICS 106 (2009). [55] L. Lindsay, D. Broido, and N. Mingo, "Lattice thermal conductivity of single-walled carbon nanotubes: Beyond the relaxation time approximation and phonon-phonon scattering selection rules", Physical Review B 80 (2009). [56] Y. Guo, and M. Wang, "Heat transport in two-dimensional materials by directly solving the phonon Boltzmann equation under Callaway's dual relaxation model", Physical Review B 96 (2017). [57] V. Martelli, J. L. Jiménez, M. Continentino, E. Baggio-Saitovitch, and K. Behnia, "Thermal Transport and Phonon Hydrodynamics in Strontium Titanate", Physical Review Letters 120 (2018). [58] S. Lee, X. Li, and R. Guo, "Thermal Resistance by Transition Between Collective and Non- Collective Phonon Flows in Graphitic Materials", Nanoscale and Microscale Thermophysical Engineering (2019). [59] L. Landau, "Theory of the Superfluidity of Helium II", Physical Review 60 (1941). [60] V. Peshkov, "Determination of the velocity of propagation of the second sound in helium II", J. Phys. USSR 10 (1946). [61] J. Ward, and J. Wilks, "The velocity of second sound in liquid helium near the absolute zero", Philos. Mag. 42 (1951). [62] J. C. Ward, and J. Wilks, "III. Second sound and the thermo-mechanical effect at very low temperatures", Philos. Mag. 43 (1952). [63] S. J. Rogers, "Transport of Heat and Approach to Second Sound in Some Isotopically Pure Alkali-Halide Crystals", PHYSICAL REVIEW B 3 (1971). [64] R. A. Guyer, "Light scattering detection of thermal waves", Physics Letters 19 (1965). [65] A. Griffin, "Brillouin Light Scattering from Crystals in the Hydrodynamic Region", Reviews of Modern Physics 40 (1968). [66] R. K. Wehner, and R. Klein, "Scattering of light by entropy fluctuations in dielectric crystals", Physica 62 (1972). [67] D. W. Pohl, and V. Irniger, "Observation of Second Sound in NaF by Means of Light Scattering", Physical Review Letters 36 (1976). [68] B. Hehlen, A.-L. Pérou, E. Courtens, and R. Vacher, "Observation of a Doublet in the Quasielastic Central Peak of Quantum-Paraelectric SrTiO3", Physical Review Letters 75 (1995). [69] A. Koreeda, R. Takano, and S. Saikan, "Second Sound in SrTiO3", PHYSICAL REVIEW LETTERS 99 (2007). [70] R. J. Hardy, "Phonon Boltzmann Equation and Second Sound in Solids", PHYSICAL REVIEW B 2 (1970). [71] J. H. Seol, I. Jo, A. L. Moore, L. Lindsay et al., "Two-dimensional phonon transport in supported graphene", Science 328 (2010). [72] L. Shi, D. Li, C. Yu, W. Jang et al., "Measuring Thermal and Thermoelectric Properties of One-Dimensional Nanostructures Using a Microfabricated Device", Journal of Heat Transfer 125 (2003). [73] E. Ou, X. Li, S. Lee, K. Watanabe, and T. Taniguchi, "Four-Probe Measurement of Thermal Transport in Suspended Few-layer Graphene with Polymer Residue", Journal of heat Transfer (in press). [74] D. G. Cahill, "Analysis of heat flow in layered structures for time-domain thermoreflectance", REVIEW OF SCIENTIFIC INSTRUMENTS 75 (2004). [75] J. A. Johnson, A. A. Maznev, J. Cuffe, J. K. Eliason et al., "Direct Measurement of Room- Temperature Nondiffusive Thermal Transport Over Micron Distances in a Silicon Membrane", Physical Review Letters 110 (2013). [76] J. R. Salcedo, A. E. Siegman, D. D. Dlott, and M. D. Fayer, "Dynamics of Energy Transport in Molecular Crystals: The Picosecond Transient-Grating Method", Physical Review Letters 41 (1978). [77] S. Huberman, R. A. Duncan, K. Chen, B. Song et al., "Observation of second sound in graphite at temperatures above 100 K", arXiv preprint arXiv:1901.09160 (2019). [78] D. G. Cahill, P. V. Braun, G. Chen, D. R. Clarke et al., "Nanoscale thermal transport. II. 2003 -- 2012", Appl. Phys. Rev. 1 (2014). [79] D. G. Cahill, W. K. Ford, K. E. Goodson, G. D. Mahan et al., "Nanoscale thermal transport", J. Appl. Phys. 93 (2003). 

1510.02759

2

1510

Topological semi-metals with line nodes and drumhead surface states

In an ordinary three-dimensional metal the Fermi surface forms a two-dimensional closed sheet separating the filled from the empty states. Topological semimetals, on the other hand, can exhibit protected one-dimensional Fermi lines or zero-dimensional Fermi points, which arise due to an intricate interplay between symmetry and topology of the electronic wavefunctions. Here, we study how reflection symmetry, time-reversal symmetry, SU(2) spin-rotation symmetry, and inversion symmetry lead to the topological protection of line nodes in three-dimensional semi-metals. We obtain the crystalline invariants that guarantee the stability of the line nodes in the bulk and show that a quantized Berry phase leads to the appearance of protected surfaces states with a nearly flat dispersion. By deriving a relation between the crystalline invariants and the Berry phase, we establish a direct connection between the stability of the line nodes and the topological surface states. As a representative example of a topological semimetal with line nodes, we consider Ca$_3$P$_2$ and discuss the topological properties of its Fermi line in terms of a low-energy effective theory and a tight-binding model, derived from ab initio DFT calculations. Due to the bulk-boundary correspondence, Ca$_3$P$_2$ displays nearly dispersionless surface states, which take the shape of a drumhead. These surface states could potentially give rise to novel topological response phenomena and provide an avenue for exotic correlation physics at the surface.

cond-mat.mes-hall

cond-mat

Topological semi-metals with line nodes and drumhead surface states Y.-H. Chan,1 Ching-Kai Chiu,2, 3, 4 M. Y. Chou,1, 5 and Andreas P. Schnyder6, ∗ 1Institute of Atomic and Molecular Sciences, Academia Sinica, Taipei 10617, Taiwan 2Department of Physics and Astronomy, University of British Columbia, Vancouver, BC, Canada V6T 1Z1 3Quantum Matter Institute, University of British Columbia, Vancouver BC, Canada V6T 1Z4 4Condensed Matter Theory Center, Department of Physics, University of Maryland, College Park, MD 20742, USA 5School of Physics, Georgia Institute of Technology, Atlanta, GA 30332, USA 6Max-Planck-Institut fur Festkorperforschung, Heisenbergstrasse 1, D-70569 Stuttgart, Germany (ΩDated: October 29, 2015) In an ordinary three-dimensional metal the Fermi surface forms a two-dimensional closed sheet separating the filled from the empty states. Topological semimetals, on the other hand, can exhibit protected one-dimensional Fermi lines or zero-dimensional Fermi points, which arise due to an intri- cate interplay between symmetry and topology of the electronic wavefunctions. Here, we study how reflection symmetry, time-reversal symmetry, SU(2) spin-rotation symmetry, and inversion symme- try lead to the topological protection of line nodes in three-dimensional semi-metals. We obtain the crystalline invariants that guarantee the stability of the line nodes in the bulk and show that a quan- tized Berry phase leads to the appearance of protected surfaces states with a nearly flat dispersion. By deriving a relation between the crystalline invariants and the Berry phase, we establish a direct connection between the stability of the line nodes and the topological surface states. As a repre- sentative example of a topological semimetal with line nodes, we consider Ca3P2 and discuss the topological properties of its Fermi line in terms of a low-energy effective theory and a tight-binding model, derived from ab initio DFT calculations. Due to the bulk-boundary correspondence, Ca3P2 displays nearly dispersionless surface states, which take the shape of a drumhead. These surface states could potentially give rise to novel topological response phenomena and provide an avenue for exotic correlation physics at the surface. I. INTRODUCTION The study of band structure topology of insulating and semi-metallic materials has become an increasingly im- portant topic in modern condensed matter physics [1– 5]. The discovery of spin-orbit induced topological insulators has revealed that a non-trivial momentum- space topology of the electronic bands can give rise to new states of matter with exotic surface states [6–11] and highly unusual magneto-transport properties [12– 14]. Recently, due to the experimental detection of arc surface states in Weyl semi-metals [15], considerable at- tention has focused on the investigation of topological semi-metals [16–31]. While in ordinary three-dimensional metals filled and empty states are separated by two- dimensional Fermi sheets, topological semi-metals can exhibit zero-dimensional Ferm points or one-dimensional Fermi lines. Classic examples of topological semi-metals are the Weyl and Dirac semi-metals which exhibit two-fold and four-fold degenerate Fermi points, respectively. Weyl points can occur in the absence of any symmetry be- sides translation, whereas Dirac points are topologically stable only in the presence of time-reversal symmetry together with a crystal lattice symmetry, such as rota- tion or reflection. For example in the Dirac materials Cd3As2 [32–37] and Na3Bi [38–42], the gapless property ∗ a.schnyder@fkf.mpg.de of the Dirac points is protected by a C4 and C3 crystal rotation symmetry, respectively. Correspondingly, the stability of Weyl points is guaranteed by a Chern num- ber, while Dirac points are protected by a crystalline in- variant, e.g., a mirror number [3]. Due to their topologi- cal characteristics these point-node semi-metals display a number of exotic transport phenomena, such as negative magneto-resistance and chiral magnetic effect [24, 43–46]. Probably even more interesting than semi-metals with point nodes are topological materials with line nodes, since they support weakly dispersing surface states that could provide an interesting platform for exotic corre- lation physics [47–49]. Moreover, these semi-metals are expected to exhibit long-range Coulomb interaction [50] and graphene-like Landau levels [51]. In nodal line semi- metals the valence and conduction bands cross along one- dimensional lines in momentum space forming a ring- shaped Fermi line. From the general classification of gap- less topological materials [3] it follows, that line nodes in semi-metals are stable against gap opening only in the presence of a lattice symmetry, such as, e.g., reflec- tion [18–20]. That is, the two bands that cross at (or near) the Fermi level of a nodal line semi-metal have opposite crystal symmetry eigenvalues, which prevents hybridization. For example, in non-centrosymmetric PbTaSe2 [52, 53] and TlTaSe2 [54] the reflection about the Ta atomic planes protects the topological nodal lines. Similarly, the band crossings in Cu3PdN [55], ZrSiS [56], and Ca3P2 [57] are protected by point group symmetries. Since the latter three systems are symmetric under both inversion and time reversal, their nodal rings are four-fold 5 1 0 2 t c O 7 2 ] l l a h - s e m . t a m - d n o c [ 2 v 9 5 7 2 0 . 0 1 5 1 : v i X r a degenerate, i.e., of “Dirac type”. In contrast, PbTaSe2 and TlTaSe2 lack inversion symmetry and hence exhibit “Weyl rings”, which are only two-fold degenerate. In this paper, by considering Ca3P2 as a representa- tive example of a topological semi-metal, we discuss the stability of topological Fermi lines in terms of crystalline topological invariants that take on nonzero quantized val- ues. These topological numbers measure the global phase structure of the electronic wavefuncitons in the presence of symmetry constraints. We derive and compute the Z- and Z2-type crystalline invariants for both a tight- binding model (Sec. II) and a low-energy effective de- scription of Ca3P2 (Sec. III). It follows from our anal- ysis that the four-fold degenerate Dirac ring of Ca3P2 [Fig. 1(d)] is protected against gap opening by reflection symmetry and SU(2) spin-rotation symmetry. The Dirac ring can be split into two two-fold degenerate Weyl rings by spin-rotation symmetry breaking perturbations, see Figs. 5 and 6. We find that the stability of both the Dirac ring and the Weyl ring are guaranteed by a Z-type mirror invariant (Sec. II B). The Fermi ring of Ca3P2 can also be stabilized by time-reversal symmetry combined with inversion, instead of reflection, in which case the protection is due to a Z2-type topological number. Unlike in crystalline topological insulators [58–62], the crystalline invariants for nodal line semi-metals are not directly linked with the appearance of surface states. Nevertheless, as we show in Sec. II C and Fig. 3, there appear topological ingap states at the surface of Ca3P2, which arise from a quantized Berry phase, rather then the crystalline invariant. Since the Berry phase is equal to π for any closed path that interlinks with the Fermi line, surface states with a nearly flat dispersion occur within two-dimensional regions of the surface Brillouin zone. These surface states take the form of a drumhead that is bounded by the projected Fermi lines (Fig. 3). We derive in Sec. II D an important relation between the Z-type mirror invariant and the Berry phase, which es- tablishes a direct connection between the appearance of the nearly flat surface states and the topological stability of the bulk Fermi line. It follows from this relation that drumhead boundary states are a generic feature of topo- logical nodal line semi-metals, occurring in both Weyl and Dirac ring systems (Figs. 3, 5, and 6). In the presence of disorder or interactions the sur- face states of nodal line semi-metals can scatter and interact with quasiparticles in the bulk, since there is no full gap in the system. Hence, impurity scatter- ing or electron-electron correlations might potentially de- stroy the boundary modes. For nearly flat surface states the effects of interactions are particularly strong, since their large density of states enhances correlation effects. Hence, even relatively weak interactions may lead to ex- otic symmetry broken states at the surface, such as sur- face magnetism or surface superconductivity (Sec. IV). Regarding the effects of disorder, we find that bulk im- purities do not destroy the surface states as long as: (i) the disorder strength is considerably smaller than the en- 2 ergy gap separating valence from conduction bands and (ii) the disorder respects reflection symmetry on average. The remainder of this paper is organized as follows. In Sec. II we discuss the topological features of nodal line semimetals in terms of a tight-binding model. We start in Sec. II A by deriving a twelve band tight-binding Hamilltonian for Ca3P2 using maximally localized Wan- nier functions. This is followed by a discussion of the topological stability of the Dirac ring in Sec. II B. We show in Sec. II C that a non-zero quantized Berry phase leads to the appearance of nearly flat surface states. The relation between the Berry phase and the crystalline topological invariant is derived in Sec. II D. Sec. II E is de- voted to the study of time-reversal and inversion breaking perturbations, which split the Dirac ring into two Weyl rings. To show that the topological features discussed in Sec. II are generic to any nodal line semi-metal, we discuss in Sec. III an effective continuum model that de- scribes the low-energy physics near a general topological Fermi line. We evaluate the crystalline invariant for this continuum model in Sec. III A. In Sec. III B we study how time-reversal and inversion breaking terms split the Fermi line. Finally, in Sec. IV we conclude the paper and give an outlook on future research. Sec. IV also contains a brief discussion of the effects of disorder on the topo- logical surface states. Some technical details have been relegated to four appendices. II. TIGHT-BINDING CALCULATIONS In this section, we examine the band structure topol- ogy of Ca3P2 in terms of a tight-binding model with twelve bands. Although the analysis below is performed specifically for Ca3P2, the principles discussed in this sec- tion are valid more generally and can be applied to any material with the same symmetries as Ca3P2. A. Tight-binding model for Ca3P2 Recently, a new polymorph of Ca3P2 has been syn- thesized which crystallizes in a hexagonal lattice struc- ture with space group P 63/mcm [57]. Figures 1(a) and 1(b) display the crystal structure of this polymorph of Ca3P2, which contains two layers with three Ca and three P atoms separated by four interstitial Ca atoms. X-ray diffraction measurements show that the Ca site is only partially occupied, yielding a Ca2+–P3− charge-balanced compound. To determine the electronic band structure we perform first principles calculations with the WIEN2k code [63] using as an input the experimental crystal structure of Ref. [57]. For the exchange-correlation functional we choose the generalized-gradient approximation of Perdew-Burke-Ernzerhof type [64]. The full Brillouin zone is sampled by 21 × 21 × 22 k-points and the plane- wave cut-off is set to RKmax = 7. We treat the par- 3 model. Hence, the tight-binding Hamiltonian is defined in terms of a twelve-component Bloch spinor R(cid:105) , eik·(R+sα) φα ψα k(cid:105) = (cid:88) (2.1) 1√ N R where α is the orbital index, R denotes the lattice vec- tors, and sα represents the position vectors of the six Ca (α = 1, . . . , 6) and the six P sites (α = 7, . . . 12), as specified in Figs. 1(a) and 1(b). For completeness, the numerical values of the position vectors sα are given in Table I of Appendix A. At this stage of the discussion, we ignore the spin degree of freedom of the Bloch spinor, since spin-orbit coupling is negligibly small for the light elements Ca and P. Using the spinor (2.1), we construct the matrix elements of the Bloch Hamiltonian as eik·(R+sα−sβ )tαβ H αβ(k) = (cid:104)ψα (cid:88) kHψβ R , (2.2) k(cid:105) = R where tαβ R is the hopping amplitude from orbital α in the unit cell at the origin to orbital β in the unit cell at posi- tion R. To simplify the form of the matrix elements (2.2) and have a single-valued Hamiltonian, we absorb a mo- mentum dependent phase factor in the definition of the basis orbitals, i.e., we let ψα k(cid:105). We observe that Hamiltonian (2.2) has a nested block structure (cid:18)HCaCa HCaP k(cid:105) → eik·sαψα (cid:19) (cid:18)hll (cid:19) ij hlu ij hul ij huu ij ij H(k) = ij , hul , Hij = HPCa HPP , (2.3) ij with fixed i, j ∈ {Ca, P} and where the sub-blocks hmn fixed m, n ∈ {l, u} are 3 × 3 matrices. The outer blocks Hij represent hopping processes among and between the Ca and P orbitals, whereas the inner blocks (huu ij , hll ij) and (hlu ij ) describe intralyer and interlayer hoppings, respectively. The detailed form of the matrix elements hmn is specified in Appendix A 1, where we also de- scribe how the hopping parameter values are determined from a maximally localized Wannier function (MLWF) method [66, 67]. In Fig. 1(d) we plot the energy isosurface of Hamil- tonian (2.2) at E = EF ± 20 meV, which shows that the tight-binding model correctly captures the fourfold degenerate Dirac ring of Ca3P2. Comparing the first- principles band structure of Fig. 1(c) with the tight- binding bands displayed in Fig. 2, we find that the tight- binding model closely reproduces the bands with dom- inant Ca-dz2 and P-px orbital character. In particular, the linear dispersion close to the Dirac ring agrees well with the first-principles results. 1. Symmetries As we will see in the following sections, time-reversal, inversion, reflection, and SU(2) spin-rotation symmetry play a crucial role for the protection of the Dirac ring. Let us therefore discuss how these symmetries act on the tight-binding Hamiltonian. FIG. 1. Crystal structure and electronic bands of Ca3P2. (a) Crystal structure of Ca3P2, which contains two planes with three Ca atoms (blue) and three P atoms (red) that are separated by interstitial Ca atoms (black). The gray dashed lines indicate the unit cell. (b) Top and side view of the crys- tal structure. The P-px and Ca-dz2 orbitals included in the tight-binding model are shown schematically. (c) Calculated electronic band structure of Ca3P2. The weights of the P-px and Ca-dz2 orbitals that are located within the layers are indi- cated by the width of the corresponding band. The weight of the Ca-dz2 orbital is multiplied by two to make it more visible on the scale of the plot. (d) Fermi ring of Ca3P2 as obtained from the tight-binding model, Eq. (2.2). The bulk and sur- face Brillouin zones are outlined by the green and black lines, respectively. tial occupancy of the Ca atoms within the virtual crys- tal approximation [65]. Figure 1(c) shows the calculated band structure of Ca3P2 within an energy range of ±3 eV around the Fermi energy EF. To obtain the orbital char- acter of the bands we introduce a local coordinate system for each Ca and P site, whose definition is illustrated in Fig. 1(b). In each coordinate frame the x axis is oriented along the c direction, whereas the z axis lies with the ab plane, pointing towards the lower left edge of the unit cell [Fig. 1(b)]. With these definitions, we find that the bands close to the Fermi energy mainly originate from the Ca-dz2 and P-px orbitals that are located within the layers [Fig. 1(c)]. The other orbitals of the in-plane atoms (Ca-dxy, Ca-dxz, Ca-dyz, Ca-dx2−y2, P-py, and P-pz), as well as all the orbitals of the Ca interstitials, contribute insignificantly to the low-energy bands and can be ne- glected for the construction of the tight-binding model. Guided by these observations, we use the six Ca-dz2 and the six P-px orbitals that are located within the two layers as a basis set for the low-energy-tight binding dz2pxCa1Ca4Ca3Ca6Ca2Ca5mirror planeP4P6P1P2Ca1Ca2Ca6Ca4P6Ca6Ca4P4P1P6P2P4P3P5(a)(upper plane)mirror plane(lower plane)(b)dz2pxAkxkykzMKK'ΓMKK'Γ(d)(c)LHFirst of all, since we did not include the spin degree of freedom in Eq. (2.2), the tight-binding model is fully SU(2) spin-rotation invariant. That is, our model is di- agonal in spin space with Hamiltonian (2.2) represent- ing the diagonal element. As a consequence, the time- reversal operator is simply given by the identity matrix times the complex conjugation operator K, i.e., T = 1K, which acts on the Hamiltonian as −1H(−k)T = H(k). T (2.4) Hence, Hamiltonian (2.2) belongs to symmetry class AI, since T 2 = +1. According to the classification of Ref. 3 Fermi rings in this symmetry class are unstable in the ab- sence of lattice symmetries. However, as we will discuss below, reflection symmetry or a combination of inversion with time-reversal symmetry can produce a topological protection of the Dirac ring. The two layers of the crystal structure of Ca3P2, in- dicated in green and brown in Fig. 1(a), are reflection planes. For brevity, we only discuss the lower reflec- tion plane [colored in green in Fig. 1(a)], but the fol- lowing analysis also holds, mutatis mutandis, for the up- per plane. The invariance of the tight-binding Hamilto- nian (2.2) under reflection about the lower plane implies −1(kz)H(kx, ky,−kz)R(kz) = H(kx, ky, kz), R (2.5a) with the kz-dependent reflection operator R(kz) = τz ⊗ ei kz = τz ⊗ (cid:18)1 (cid:19) 2 (ρz−ρ0)c ⊗ 13×3 ⊗ 13×3, 0 0 e+ikzc (2.5b) where c is the length of the lattice vector along the (001) direction. Here, the two sets of Pauli matrices τα and ρα describe the orbital (Ca-dz2, P-px) and the layer (l, u) degrees of freedom, respectively. The form of the reflec- tion operator R(kz) follows from the observations that (i) the P-px orbitals are odd under reflection, while the Ca-dz2 orbitals are even; and (ii) the mirror symmetry maps the orbitals in the upper layer to the next unit cell, which gives rise to the phase factor e+ikzc. Finally, we find that the tight-binding model is also inversion sym- metric. That is, Hamiltonian (2.2) satisfies −1H(−k)I = H(k), (2.6) with the spatial inversion operator I = τ0 ⊗ ρx ⊗ 13×3. I B. Topological protection of the Fermi ring Let us now discuss how reflection symmetry (2.5) leads to the topological protection of the Dirac ring. First, we observe that for k within the reflection plane kz = 0, π the mirror operator R(kz) commutes with Hamilto- nian (2.2), i.e., [R(kz), H(kx, ky, kz)] = 0 for kz = 0, π. Therefore, it is possible to block-diagonalize H(k) within 4 FIG. 2. Band structure of the tight-binding model. Pan- els (a) and (b) show the energy bands of Hamiltonian (2.2) along high-symmetry lines within the mirror planes kz = 0 and kz = π/c, respectively [cf. Fig. 1(d)] . The reflection eigenvalues of the bands are indicated by color, with blue and red corresponding to R = +1 and R = −1, respectively. the mirror planes with respect to R. In this block- diagonal basis each eigenstate of H(k) has either mir- ror eigenvalue R = +1 or R = −1. As we can see from Fig. 2(a), the two bands that cross at the Dirac point have opposite mirror eigenvalues, which prevent hy- bridization between them. In other words, any term that couples the two bands breaks reflection symmetry. The stability of the band crossing is guaranteed by a mirror invariant of type M Z [18]. This mirror index is given by the difference of occupied states with eigenvalue R = +1 on either side of the Dirac ring, i.e., N 0 MZ = n+,0 occ (k(cid:107) > k0) − n+,0 occ (k(cid:107) < k0), (2.7) where k(cid:107) = (kx, ky) is the in-plane momentum and n+,0 occ (k(cid:107)) = k(cid:107) < k0 (inside the ring) k(cid:107) > k0 (outside the ring) (2.8) (cid:26) 1, 0, denotes the number of occupied states at (k(cid:107), 0) in the mirror eigenspace R = +1. In passing, we note that Hamiltonian (2.2) is a mem- ber of symmetry class AI with R+ in the terminology of Ref. 18, since T 2 = +1 and R commutes with T . However, nodal lines with codimension p = 2 in class AI with R+ are unstable, since for this class there does not exist any zero-dimensional invariant defined at time- reversal invariant momenta within the mirror plane. Nev- ertheless, the Dirac band crossing is protected, since the Hamiltonian can also be viewed as a member of class A with R. The mirror invariant for the latter class [i.e., Eq. (2.8)], which is defined for any in-plane momentum k(cid:107), can be non-zero even in the presence of time-reversal symmetry. Besides reflection symmetry, the product of inversion and time-reversal symmetry IT also protects the Dirac line. This will be discussed at the end of Sec. II C and in Sec. III B 1 in terms of a low-energy con- tinuum model. −2024Energy (eV)(a) Γk0MKk0ΓR=+1R=−1kz=0ALHA−2024Energy (eV)(b) R=+1R=−1kz=π/cC. Surface states and Berry phase In this section, we present the surface spectrum of Ca3P2 as obtained from the tight-binding model (2.2) and show that, due to a non-zero Berry phase, there ap- pear nearly flat ingap states at the surface. Figure 3(a) displays the surface band structure for the (001) surface in a three-dimensional slab geometry with 60 unit cells. The surface momentum is varied along a high-symmetry path, which is drawn in red in the surface Brillouin zone of Fig. 1(d). Using an iterative Green’s function method [68] we compute the momentum resolved surface density of states for a semi-infinite (001) slab, which is shown in Fig. 3(b). As indicated by the green area in Fig. 3(d) and by the green and yellow lines in Figs. 3(a) and 3(b), respectively, the surface state is nearly disper- sionless, taking the shape of a drumhead that is bounded by the projected Dirac ring. We note that nearly or com- pletely flat surface states have recently also been studied in photonic crystals [69], in noncentrosymmetric super- conductors [70–73], in bernal graphite [74], and in topo- logical crystalline insulator heterostructures [48]. In contrast to crystalline topological insulators the sur- face states of the semimetal (2.2) are not directly related to the mirror invariant (2.7), but are connected to a non- zero Berry phase. To make this connection explicit, we decompose the (001) slab considered in Fig. 3 into a fam- ily of one-dimensional systems parametrized by the in- plane momentum k(cid:107) = (kx, ky). For fixed k(cid:107), the Berry phase is defined as P(k(cid:107)) = −i (cid:104)uj(k)∂kzuj(k)(cid:105)dkz, (2.9) (cid:90) π (cid:88) −π Ej <EF where the sum is over filled Bloch eigenstates uj(k)(cid:105) of Hamiltonian (2.2). As was shown by King-Smith and Vanderbilt [75], the Berry phase P(k(cid:107)) is related to the charge qend at the end of the one-dimensional system with fixed in-plane momentum k(cid:107), i.e., P(k(cid:107)) mod e. e 2π qend = (2.10) Hence, when P(k(cid:107)) (cid:54)= 0 an ingap state appears at k(cid:107) in the surface Brillouin zone. For the tight-binding Hamil- tonian (2.2) we find that there are two different sym- metries which each quantize the Berry phase (2.9) to 0 or π, namely, the reflection symmetry (2.5) and the product of time-reversal and inversion symmetry IT , see Appendix B. In Fig. 3(c) we numerically compute P(k(cid:107)) using the tight-binding wave functions of Hamil- tonian (2.2). We obtain that the Berry phase equals π for k(cid:107) inside the projected Dirac ring, while it is zero for k(cid:107) outside the ring. This indicates that surface states occur within the projected Dirac ring, which is in agree- ment with the surface spectrum of Figs. 3(a) and 3(b). The Berry phase is defined modulo 2π, since large gauge transformations of the wave functions change it by 2π. As a result, P protects only single, but not multiple, sur- face states at a given k(cid:107). 5 FIG. 3. Drumhead surface states and Berry phase. (a) Surface band structure of Ca3P2 as obtained from the tight-binding model (2.2) for the (001) surface in slab geom- etry with 60 unit cells. The surface state is highlighted in green. (b) Momentum-resolved surface density of states of Hamiltonian (2.2) for the (001) surface. White and dark red correspond to high and low density, respectively. (c) Vari- ation of the Berry phase (2.9) of Hamiltonian (2.2) along high-symmetry lines of the (001) surface Brillouin zone [see Fig. 1(d)]. (d) Surface spectrum of the low-energy effective model (3.1) for the (001) face as a function of surface mo- menta kx and ky. The bulk states at kz = 0 with reflection eigenvalues R = +1 and R = −1 are colored in blue and red, respectively. The drumhead surface state is indicated by the green area. Remarkably due to the IT symmetry, the Berry phase P along any closed loop in the three-dimensional Brillouin zone is quantized (see Appendix B). This allows us to interprete the Berry phase as a topological invari- ant which guarantees the stability of the Dirac line in the presence of the IT symmetry. That is, for a loop inter- linking with the Dirac ring, we find that P = ±π which shows that the Dirac band crossing is protected by the product of inversion with time-reversal symmetry. The Berry phase represents a Z2-type invariant, since it is de- fined only up to multiples of 2π. In contrast, the mirror number (2.7) is a Z-type invariant, which can take on any integer number. Therefore, only the mirror invari- ant NMZ can give rise to the stability of multiple Dirac lines at the some location in the Brillouin zone. D. Relation between Berry phase and mirror invariant The analysis of the previous section suggests that the topological stability of the Dirac ring is closely related to the appearance of surface states. In order to put this connection on a firmer footing, we present here a relation (a)(b)(c)(d)kx(1/a)ky(1/a)E(eV)between the mirror invariant and the Berry phase P(k(cid:107)). Namely, we find that 6 (−1)n+,0 occ (k(cid:107))+n+,π occ (k(cid:107))ei∂R = eiP(k(cid:107)) (2.11a) (cid:90) π (cid:88) Ej <EF 0 for all in-plane momenta k(cid:107) = (kx, ky), where ∂R = i (cid:104)uj(k)R † (kz) [∂kz R(kz)]uj(k)(cid:105)dkz (2.11b) denotes the change in phase of the reflection operator R(kz) along the reflection direction kz. The invariants n+,0 occ (k(cid:107)) and n+,π occ (k(cid:107)) correspond to the number of oc- cupied states at (k(cid:107), 0) and (k(cid:107), π), respectively, with mirror eigenvalue R = +1. Formula (2.11), whose proof is derived in Appendix B, is one of the main results of this paper. For concreteness we have assumed in (2.11) that reflection symmetry R(kz) maps z to −z. But re- lationn (2.11) is valid more generally, i.e., for any re- flection symmetric semimetal, in particular also for line- node materials with strong spin-orbit coupling, such as PbTaSe2 [52, 53]. We observe that in general the reflection operator only depends on the momentum along the reflection direc- tion [i.e., on kz in the case of Eq. (2.5)], but is inde- pendent of the in-plane momenta k(cid:107). Hence, we infer from Eq. (2.11) that when the mirror invariant n+,0 occ (k(cid:107)) [or n+,π occ (k(cid:107))] changes by one as the in-plane momentum k(cid:107) is moved across the topological Dirac line, the Berry phase increases by π, since ∂R does not depend on k(cid:107). As a consequence, a drumhead surface state appears either inside or outside the projected Dirac ring. This proofs the direct connection between the stability of the Dirac ring and the existence of drumhead surface states. For the tight-binding model of Ca3P2, Eq. (2.2), we find that the phase change ∂R of the reflection operator (2.5) eval- uates to 3π independent of k(cid:107). Figure 2(b) shows that the number of occupied states with momentum (k(cid:107), π) and mirror eigenvalue R = +1 is n+,π occ (k(cid:107)) = 3 for all k(cid:107). Using relation (2.11) together with Eq. (2.8), it follows that the Berry phase P equals π inside and 0 outside the Dirac ring, which agrees with the explicit calculation of P, see Fig. 3(c). In closing this section, we note that for certain highly symmetric lattice models [60, 76] the reflection operator R is completely momentum independent, in which case formula (2.11) simplifies to occ (k(cid:107))(cid:3) π = P(k(cid:107)) (mod 2π), (2.12) (cid:2)n+,0 occ (k(cid:107)) + n+,π for all k(cid:107) [77]. Hence, in this case the Berry phase, and therefore the location of the surface states, is fully deter- mined by the mirror invariant (2.8). This is useful, since the mirror number (2.8) is easier to compute than the Berry phase, for which one needs to determine the mo- mentum dependence of the tight-binding wave functions. Arc surface state and spin Chern number. FIG. 4. (a) ky dependence of the spin Chern number (2.14) of Hamil- tonian (2.2) in the presence of the mirror and time-reversal symmetry breaking perturbation (2.13). (b) Surface and bulk spectra of the low-energy model (3.1) perturbed by the mass term (3.2) with d = 0.9 eVA and θ0 = −π/4, which breaks reflection and time-reversal symmetry. The bulk states and the arc state at the (001) surface are indicated in gray and green, respectively. E. Symmetry-breaking perturbations We have seen that the stability of the Dirac ring of Ca3P2 is protected by SU(2) spin-rotation symmetry, re- flection symmetry, and the product of inversion and time- reversal symmetry IT . In this section, we study how the breaking of these symmetries modifies the bulk and sur- face spectrum of Ca3P2. 1. Reflection and time-reversal symmetry breaking First, we consider a reflection and time-reversal break- ing perturbation with the following nonzero matrix ele- ments (cid:104)ψ1 kHψ9 k(cid:105) = +0.2 sin(k · r0) and (cid:104)ψ4 kHψ12 k (cid:105) = −0.2 sin(k · r0), (2.13a) (2.13b) where r0 = (0.5, 0.5, 0) is a vector within the reflection plane along the diagonal direction. This term is odd in momentum k and couples the dz2 orbitals at the Ca1 and Ca4 sites with the px orbitals at the P3 and P6 sites [cf. Figs. 1(a) and 1(b)]. It follows from Eqs. (2.5) and (2.6) that perturbation (2.13) breaks reflection and time-reversal symmetry, but respects inversion symme- try. Therefore, Eq. (2.13) gaps out the Dirac ring except for two points along the diagonal direction (1, −1, 0), where it vanishes [see Fig. 4(b)]. These two gap closing points are Dirac nodes (or Weyl nodes, if one disregards the spin degree of freedom), whose stability is guaranteed by the spin Chern number [78] Cs(ky) = 1 2πi ∂kxA(j) z − ∂kz A(j) x dkxdkz,(2.14) (cid:90) (cid:88) (cid:104) Ej <EF T 2 (cid:105) −0.5−0.3−0.10.10.30.500.51kyChern number(a)(2π/a)(a)E(eV)ky(1/a)kx(1/a)(b)µ = (cid:104)uj∂kµuj(cid:105) is the Berry connection. We where A(j) find that Cs(ky) evaluates to +1 for kxkz planes inbe- tween the two Dirac points, while it is zero otherwise [Fig. 4(a)]. By the bulk-boundary correspondence, the nonzero spin Chern number (2.14) implies the appear- ance of an arc state in the surface Brillouin zone connect- ing the projections of the two Dirac nodes [green area in Fig. 4(b)]. As perturbation (2.13) is turned to zero, the arc state transforms into the drumhead surface state of Fig. 3. 2. Spin-rotation symmetry breaking Second, we study the effects of SU(2) spin-rotation symmetry breaking induced, for example, by spin-orbit coupling. For Ca3P2 the spin-orbit interactions are negli- gible due to the small atomic number of Ca and P. How- ever, there are a number of topological semimetals with heavy elements, such as PbTaSe2 and TlTaSe2, for which spin-orbit coupling is strong. Spin-orbit interactions can modify the energy spectrum of nodal line semimetals in two different ways: either they open up a full gap in the spectrum, or they split the Dirac ring into two Weyl rings. Here, we study the latter possibility. In order to do so, we need to explicitly include the spin degree of freedom in Hamiltonian (2.3), i.e., we consider H(k) = H(k) ⊗ σ0 + Hsb(k), (2.15) where σ0 operates in spin space and Hsb represents a spin-rotation symmetry breaking term, which we specify below. Time-reversal symmetry acts on H according to Eq. (2.4), but with the modified time-reversal operator T = T ⊗ iσy. Similarly, the reflection operator and the spatial inversion operator are changed to R = R⊗σz and I = I ⊗ σ0, respectively. To split the four-fold degener- ate Dirac ring of Eq. (2.15) into two two-fold degenerate Weyl rings, it is necessary to also break time-reversal or inversion symmetry, besides spin-rotation symmetry. a. Time-reversal breaking perturbation The stag- gered Zeeman field Hsb(k) = hz τz ⊗ ρ0 ⊗ 13×3 ⊗ σz (2.16) breaks both time-reversal and spin-rotation symmetry, but satisfies inversion and reflection symmetry. It de- scribes an external staggered magnetic field with oppo- site signs on the Ca and P sites. According to the ter- minology of Ref. [18], Hamitlonain (2.15) perturbed by Eq. (2.16) is a member of class A with R, which exhibits an integer number of equivalence classes distinguished by a mirror invariant. In Figs. 5(a) and 5(c) we present the bulk energy bands of Hamiltonian (2.15) with an applied staggered Zeeman field of strength hz = 0.1 eV. The bulk spectrum displays two Weyl rings, whose stability is guar- anteed by the mirror number (2.7). Figures 5(b) and 5(d) show the surface energy spectrum at the (001) face. We find that there are two drumhead surface states which 7 Bulk bands and drumhead surface states of a FIG. 5. spinful time-reversal breaking line-node semimetal. Panels (a) and (b) show the bulk bands and the surface density of states of Hamiltonian (2.15) in the presence of the staggered Zeeman term (2.16) with hz = 0.1 eV. The momentum in panel (a) is varied within the mirror plane kz = 0 along high- symmetry lines of the Brillouin zone. (c) Energy isosurfaces of Hamiltonian (2.15) with hz = 0.1 eV at EF ± 5 meV and kz = 0. (d) Surface and bulk spectra of the low-energy ef- fective model (3.3) perturbed by the time-reversal breaking term (3.4) with ν(cid:107)hz eff = 0.07 eV. The drumhead states at the (001) surface are colored in green. The reflection eigen- values of the bulk bands at kz = 0 in panels (a), (c), and (d) are indicated by color, with blue and red corresponding to R = +1 and R = −1, respectively. are bounded by the projections of the two Weyl rings. In accordance with the discussion of Secs. II C and II D [cf. Eq. (2.11)] the single surface state that appears be- tween the projections of the outer and inner Weyl rings is protected by the Berry phase (2.9), which takes on the nonzero quantized value P = ±π. The two surface states that exist inside the projection of the inner Weyl ring, on the other hand, are topologically unstable. b. kσ Hψ6 Inversion breaking perturbation To break inver- sion and spin-rotation symmetry we consider a perturba- tion with the following nonzero matrix elements (cid:104)ψ1 kσ(cid:105) = +0.6i sgn(σ)eik·(s6−s1)(cid:2)1 + eik·ez(cid:3) kσ(cid:105) = −0.3i sgn(σ)eik·(s12−s7+R110)(cid:2)1 + eik·ez(cid:3) , and (cid:104)ψ7 kσ Hψ12 (2.17b) where ψα kσ(cid:105) denotes the Bloch spinor with orbital index α and spin index σ = ±. The vectors sα are the posi- tion vectors of the atoms in the unit cell and are given in Table I of Appendix A. Perturbation (2.17) couples the orbitals at the Ca1 and P1 sites with the orbitals at the Ca6 and P6 sites, respectively. Using Eqs. (2.4), (2.5), (2.17a) −4−2024−4−2024kx (1/a)ky (1/a)(c) R=+1R=−1−2024Energy (eV)ΓMKΓ(a) R=+1R=−1(a)(b)(c)(d)ky(1/a)E(eV)kx(1/a)8 form of this low-energy description is universal, since it is entirely dictated by symmetry. We start by discussing Dirac rings, which arise in semimetals with conserved SU(2) spin-rotation symmetry. Spin rotation breaking semimetals with Weyl nodal lines will be discussed in Sec. III B 2. Consider the following low-energy Hamiltonian with spin-rotation symmetry Heff (k) = ν(cid:107)(k2(cid:107) − k2 0)τz + νzkzτy + f (k)τ0, (3.1) x + k2 y = k2 which describes a Dirac ring within the kz = 0 plane, located at k2(cid:107) := k2 0. In Eq. (3.1) we suppress the spin degree of freedom, since any spin-dependent terms are forbidden by symmetry. The Pauli matrices τi operate in orbital space and the function f (k) is re- stricted by symmetry to be even in k. We assume that f (k) = ν0(k2(cid:107) − k2 0) + V0, neglecting any terms of higher order in k. To make a connection with the previous sec- tion, we fit the parameters ν0, ν(cid:107), νz, k0, and V0 to the low-energy band structure of the DFT calculations of Sec. II A [see Fig. 1(c)]. We find that the momentum parameter k0 equals k0 = 0.206 A−1, the chemical po- tential is V0 = 0.095 eV, and the velocities are given by ν0 = −0.993 eVA2, ν(cid:107) = 4.34 eVA2, and νz = 2.50 eVA. Employing Eqs. (2.4), (2.5), and (2.6), one can show that the low-energy Hamiltonian Heff satisfies time-reversal, reflection, and inversion symmetry, with the symmetry operators Teff = τ0K, Reff = τz, and Ieff = τz, respec- tively. Before we discuss in the next section the topolog- ical stability of the Dirac line (3.1), let us remark that Heff(k) can be converted in a straightforward manner to a lattice model, see Appendix C. In Figs. 3(d), 4(b), 5(d), and 6(d) we use the lattice version of Eq. (3.1) to plot the surface states. Observe that there are some minor differences in the shape of the surface states between the thigh-binding model (2.2) and the effective theory (3.1) [compare Fig. 3(b) with Fig. 3(d)]. We attribute this dif- ference to the omission of longer range hopping terms in Eq. (3.1). A. Topological protection of the Fermi ring As mentioned in Sec. II B, Dirac nodal lines are pro- tected by either reflection symmetry R or the product of inversion with time-reversal symmetry IT . Let us now discuss this in terms of the low-energy theory (3.1). a. Z classification due to reflection symmetry Con- sidering only reflection symmetry and disregarding the spin degree of freedom, Hamiltonian (3.1) belongs to class A with R. Since the codimension of the Dirac ring is p = 2, it is classified by an MZ invariant (see Table II of Ref. [18]), i.e., by the mirror number (2.7), which mea- sures the difference of occupied states with mirror eigen- value Reff = +1 on either side of the Dirac ring. The two bands that cross at the nodal line have opposite reflec- tion eigenvalues, which prohibits hybridization between them. Indeed, we find that the hybridization term τx FIG. 6. Bulk bands and drumhead surface states of a spinful inversion breaking line-node semimetal. Panels (a) and (b) display the bulk bands and the surface density of states of tight-binding model (2.15) in the presence of the inversion breaking term (2.17). The momentum in panel (a) is varied within the mirror plane kz = 0 along high-symmetry lines. (c) Energy isosurfaces of Hamiltonian (2.15) perturbed by Eq. (2.17) at EF ±5 meV and kz = 0. (d) Surface and bulk spectra of the low-energy effective model (3.3) perturbed by the inversion breaking term (3.6) with δ = 0.025 eVA. The drumhead states at the (001) surface are indicated in green. The mirror eigenvalues of the bulk bands at kz = 0 in panels (a), (c), and (d) are represented by color, with blue and red corresponding to R = +1 and R = −1, respectively. and (2.6) one can check that the term (2.17) satisfies re- flection and time-reversal symmetry, but breaks inversion symmetry. Since T 2 = −1 and { T , R} = 0, Hamilto- nian (2.15) perturbed by Eq. (2.17) is a member of class AII with R− of Ref. [18], for which a mirror invariant can be defined. The bulk bands at kz = 0 of Hamil- tonian (2.15) in the presence of the inversion-breaking term (2.17) are presented in Figs. 6(a) and 6(b). We observe that the Dirac ring is split into two Weyl rings, 3,−1, 0) axis. As in the pre- which intersect on the ( vious cases, the Weyl nodal lines are protected by the nonzero mirror number (2.7). Figures 6(b) and 6(d) show the surface spectrum at the (001) surface, which exhibits two drumhead surface states. As before, we find that only the single surface state which occurs between the projections of the inner and outer rings is protected by the Berry phase (2.9). √ III. LOW-ENERGY CONTINUUM THEORY OF NODAL LINE SEMIMETALS In this section we present a low-energy effective the- ory for a general topological nodal line semimetal with time-reversal, reflection, and inversion symmetry. The −4−2024−4−2024(c)kx (1/a)ky (1/a) R=+1R=−1−2024Energy (eV)ΓMKΓ(a) R=+1R=−1(a)(b)(c)(d)ky(1/a)E(eV)kx(1/a)eff. breaks reflection symmetry Reff. We note that the mirror invariant (2.7) is of Z type and can therefore protect mul- tiple Dirac crossings in the Brillouin zone. To verify this for the low-energy model (3.1), we enlarge the matrix di- mension of Hamiltonian Heff by considering Heff ⊗ 1n×n, which respects reflection symmetry with the enlarged re- eff = Reff⊗ 1n×n. Hybridization terms flection operator R(cid:48) for the enlarged Hamiltonian are of the form τx⊗ A, with A an arbitrary n × n Hermitian matrix. However, since eff)−1(τx ⊗ A)R(cid:48) eff = −τx ⊗ A, all of these terms break (R(cid:48) reflection symmetry R(cid:48) b. Z2 classification due to IT symmetry Besides re- flection, also the product of inversion with time-reversal symmetry IeffTeff prohibits hybridization between the two bands, since the hybridization term τx is not invari- ant under IeffTeff = τzK. But in the presence of IeffTeff, Dirac nodal lines are classified as Z2 instead of MZ. To see this, consider two copies of Hamiltonian Heff, i.e., Heff ⊗ µ0, which satisfies IT symmetry with the doubled operator IeffTeff ⊗ µ0. Here, µα denotes an additional set of Pauli matrices. The Dirac rings of this doubled Hamil- tonian are topologically unstable, since the symmetry- preserving term τx ⊗ µy gaps out the nodal lines. As discussed at the end of Sec. II C, the product of inver- sion with time-reversal symmetry IT quantizes the Berry phase P to 0 or π [19, 79]. Hence, P can be interpreted as a Z2 topological invariant that guarantees the stabil- ity of the nodal ring. In contrast to the mirror invariant, the integration path that enters in the definition of this Z2 number [cf. Eq. (2.9)], is not confined to the mirror plane. For any integration path that interlinks with the nodal line, P = ±π signals the stability of the Dirac ring. In closing we note that, while the low-energy the- ory (3.1) accurately captures the topological stability of the nodal ring of a given semi-metal, it does not neces- sarily correctly reproduce the location of the drumhead surface state. That is, in order to determine whether the drumhead surface state is located inside or outside the projected Dirac ring, it is necessary to compute the Berry phase of all the occupied states. This information is not contained in the low-energy model (3.1), cf. Ap- pendix C. B. Symmetry-breaking perturbations In analogy to the discussion of Sec. II E, we now study how different symmetry breaking perturbations trans- form the Dirac ring (3.1) into Dirac points or Weyl rings. 1. Reflection and time-reversal symmetry breaking The Dirac line node of Heff can be deformed into two Dirac points by the perturbation d sin(θ(cid:107) − θ0)k(cid:107)τx, (3.2) 9 (cid:113) k2 x + k2 which breaks reflection and time-reversal symmetry, but respects inversion symmetry. Here, θ(cid:107) = tan−1(ky/kx) and k(cid:107) = y denote polar angle and abso- lute value of the in-plane momentum k(cid:107), respectively. The term (3.2) gaps the Dirac ring except at k = ±k0(cos θ0, sin θ0, 0). These two gap closing points are Dirac nodes with opposite chiralities, which are pro- tected by the spin Chern number (2.14). Due to the bulk-boundary correspondence an arc state appears at the surface, connecting the projected Dirac points in the surface Brillouin zone. This is illustrated in Fig. 4(b), where we set θ0 = −π/4 and d = 0.9 eVA, which mim- ics the effects of perturbation (2.13) for the tight-binding Hamiltonian (2.2). From the arc surface state of the above Dirac semimetal one can infer the existence of the drumhead surface state of Heff, since the two transform into each other by letting d tend to zero in Eq. (3.2). Moreover, the one-dimensional set of Dirac nodes, induced by Eq.(3.2) and parametrize by θ0, can be interpreted as the Dirac ring of Heff. That is, as we let θ0 in Eq. (3.2) run from 0 to π a nodal ring is created. For each fixed θ0 there is an arc surface state connecting the two points k(cid:107) = ±k0(cos θ0, sin θ0) in the surface Brillouin zone. Hence, a drumhead surface state is generated when θ0 is varied from 0 to π. From this argument one infers that drumhead states also appear at surfaces for which the Berry phase (2.9) is not quantized (cf. Sec. II C), since the appearance of arc states does not depend on any crys- tal symmetries. 2. Spin-rotation symmetry breaking In the absence of SU(2) spin-rotation symmetry, the Dirac ring of Heff is topologically unstable. To discuss this, we consider as in Sec. II E 2 a spinful version of Hamiltonian (3.1) Heff(k) = Heff(k) ⊗ σ0 + H sb eff(k), (3.3) where the Pauli matrices σα describe the spin degree of freedom and H sb eff denotes a spin-rotation symmetry Heff is invariant under the same sym- breaking term. metries as the spinful tight-binding Hamiltonian (2.15). That is, it satisfies time-reversal, reflection, and inversion symmetry with the operators T = τ0⊗iσyK, R = τz⊗σz, and I = τz ⊗ σ0, respectively. We find that, the Dirac nodal lines of Heff can be gapped out by the spin-rotation symmetry breaking mass terms τx⊗σx and τx⊗σy, which preserve reflection symmetry R as well as I T symmetry. These perturbations turn Hamiltonian (3.3) into a trivial insulator. However, there exist also other spin-rotation symmetry breaking terms that deform the Dirac ring into two Weyl rings. These perturbation terms break either time-reversal symmetry or inversion symmetry. a. Time-reversal breaking perturbation First, we add a spin-rotation and time-reversal breaking term to the Hamiltonian Heff, which takes the form of a stag- gered Zeeman field H sb effτz ⊗ σz. eff(k) = ν(cid:107)hz (3.4) (cid:112) This perturbation respects reflection and inversion sym- It splits the Dirac ring into two Weyl rings metry. that are located within the mirror plane kz = 0 at 0 ± hz k2 k(cid:107) = eff. The stability of these Weyl nodal lines is guaranteed by the mirror invariant (2.7), which evalu- ates to  1, k(cid:107) < (cid:112) (cid:112) (cid:112) 0 − hz k2 0 − hz k2 eff < k(cid:107) < k2 eff < k(cid:107) 0 + hz 0, 1, eff (cid:112) n+,0 occ (k(cid:107)) = k2 0 + hz eff .(3.5) In Fig. 5(d) we plot the surface spectrum of Heff in the presence of the staggered Zeeman term with ν(cid:107)hz eff = 0.07 eV. There appear two drumhead surface states which are bounded by the two projected Weyl rings. b. Inversion breaking perturbation Alternatively, the Dirac ring can be split into Wely rings by an inversion breaking perturbation. To show this, we consider H sb eff(k) = δ(kx + 3ky)τz ⊗ σz, (3.6) √ which respects reflection and time-reversal symmetry. In the presence of this term Hamiltonian (3.3) exhibits two Weyl rings within the mirror plane kz = 0 with in- plane momenta given by the equation (kx± δ/2)2 + (ky ± √ 3δ/2)2 = k2 0 + δ2. These two Weyl rings intersect on 3,−1, 0) axis, where the gap term (3.6) vanishes the ( [cf. Fig. 6(c)]. We find again that these Weyl rings are protected by the mirror number (2.7), with √  1, (kx ± δ (kx ∓ δ 0, otherwise n+,0 occ (k(cid:107)) = 2 )2 + (ky ± √ 2 )2 + (ky ∓ √ 3δ 3δ 2 )2 > k2 2 )2 < k2 0 + δ2 & 0 + δ2 .(3.7) Fig. 6(d) shows the surface spectrum of Heff perturbed by Eq. (3.6). As for the tight-binding model with the inversion-breaking term (2.17), there appear two drum- head surface states. We note that PbTaSe2 [52, 53] and TlTaSe2 [54] are examples of inversion breaking semi- metals with Weyl nodal lines. The low-energy physics of these materials can be described by the effective the- ory (3.3) perturbed by a term of the form (3.6). IV. SUMMARY AND DISCUSSION In this paper we have studied the topological stabil- ity of Dirac and Weyl line nodes of three-dimensional semimetals in the presence of reflection symmetry, time- reversal symmetry, inversion symmetry, and SU(2) spin- rotation symmetry. We have shown that when spin- rotation symmetry is preserved, the Dirac line is pro- tected by either reflection symmetry or the product of inversion with time-reversal symmetry IT . In the for- mer case, the nodal lines are classified by an M Z invari- ant [18], which takes the form of a mirror number, see 10 Eq. (2.7). In the latter case the stability of the Dirac line is guaranteed by a quantized nonzero Berry phase, which leads to a Z2 classification, see Eq. (2.9). As a representa- tive example of a line node semimetal, we have considered Ca3P2 [56], which exhibits a topologically stable Dirac ring at the Fermi energy. By means of a tight-binding model derived from ab initio DFT calculations, we have computed the mirror number and the quantized Berry phase for this material (Fig. 3) and shown that the Dirac band crossing is protected by reflection or IT symmetry. The band topology of this Dirac line semimetal was also discussed in terms of a low-energy effective theory, see Eq. (3.1). Even though the mirror invariant (2.7) does not di- rectly give rise to topological surface states, Dirac line semimetals generically exhibit drumhead surface states which are due to a quantized Berry phase. By deriving a relation between the mirror number (2.7) and the Berry phase (2.9), we have established a direct connection be- tween the existence of drumhead surface states and the topological stability of Dirac nodal lines in the bulk, see Eq. (2.11). Using the ab initio derived tight-binding model, we have computed the surface spectrum of Ca3P2, showing that its drumhead surface state is weakly dis- persing with an effective mass m∗ (cid:39) 4me [Fig. (3)(b) and (3)(d)]. In Ca3P2 spin-rotation symmetry is conserved to a very good approximation, since spin orbit coupling for the light elements Ca and P is very small. However, there are nodal line semimetals with heavy atoms, such as PbTaSe2 and TlTaSe2, in which spin-rotation symme- try is broken, due to the non-negligible spin-orbit inter- actions. In these systems the four-fold degenerate Dirac rings are unstable. Two-fold degenerate Weyl rings, on the other hand, can be protected against gap opening by reflection symmetry, provided either time-reversal or re- flection symmetry is broken. We have shown that the sta- bility of these Weyl rings is guaranteed by the mirror in- variant (2.7). Similar to the Dirac nodal line semimetals, Weyl ring semimetals support drumhead surface states (Figs. 5 and 6). The region in the surface Brillouin zone where these drumhead states appear are bounded by the projected Weyl rings. Determining the stability of the drumhead surface states against disorder and interactions needs a care- ful analysis of different types of scattering and interac- tion processes, involving both states near the bulk line nodes and surface states. The drumhead surface state of Ca3P2 has a relatively weak dispersion (Fig. 3), which gives rise to a large density of states thereby enhanc- ing interaction effects. Therefore, even small interactions may lead to unusual symmetry-broken states at the sur- face, such as surface superconductivity [47, 48] or surface magnetism [49]. Disorder scattering, on the other hand, breaks the crystalline symmetries that protect the sur- face states. Moreover, it mixes bulk and surface states, since there is no full gap in the bulk energy spectrum. For the case of crystalline topological insulators it has been shown that the surface states are robust against disor- der when the disorder respects the crystal symmetries on average [80]. In appendix D, we study this question in terms of a one-dimensional reflection symmetric toy model with a quantized Berry phase. In order to infer how impurity scattering affects the topological proper- ties, we determine the charge that is accumulated at the two ends of this one-dimensional system [79]. We find that even in the presence of disorder that respects re- flection symmetry on average, the end charges remain to a good approximation quantized to ±e/2. Due to Eq. (2.10), which relates the end charges to the Berry phase, this indicates that the bulk topological properties remain unaffected by this type of disorder. This finding suggest that the drumhead surface states of nodal line semimetals are not destroyed by impurities, as long as the disorder respects reflection symmetry on average and its strength is smaller than the energy gap between the conduction and valence bands. In conclusion, Dirac ring and Weyl ring semimetals are a new type of topological material which is charac- terized by a non-zero mirror invariant and a quantized Berry phase. The nontrivial band topology of these sys- tems manifests itself at the surface in terms of protected drumhead surface states. There are many interesting av- enues for future research on line node semimetals. For example, the drumhead states may give rise to unusual correlation physics at the surface. Another promising di- rection for future work is the study of novel topological response phenomena in these systems. Note added. — Upon completion of this work, we be- came aware of a study by Yamakage et al. [81], which dis- cusses the topology of line node semi-metals in terms of a k-independent reflection operator, using a k-dependent gauge transformation. The authors thank M. Franz, G. Bian, T. Heikkila, T. Hyart, L. Schoop, and D.-H. Xu for useful discussions. The support of the Max-Planck-UBC Centre for Quan- tum Materials is gratefully acknowledged. APS was sup- ported in part by the National Science Foundation under Grant No. NSF PHY11-25915. CCK was also supported by Microsoft and LPS-MPO-CMTC during the last stage of this work. YHC is supported by a Thematic Project at Academia Sinica. MYC acknowledges the support by the U.S. Department of Energy, Office of Science, Office of Basic Energy Sciences, under Award No. DE-FG02- 97ER45632. Appendix A: Details of the tight-binding model In this Appendix we give a detailed description of the tight-binding Hamiltonian of Sec.II. ACKNOWLEDGMENTS HCa = , (A1) 11 TABLE I. Position vectors sα of each orbital. All vectors are given in the crystal coordinate system, which is indicated by the red/green arrows in Fig. 7. The lattice vectors are a = [7.150, -4.218, 0.000], b = [0.000, 8.256, 0.000], and c = [0.000, 0.000, 6.836] in the unit of A. α 1 2 3 4 5 6 7 8 9 10 11 12 Orbital sα Ca1 Ca2 Ca3 Ca4 Ca5 Ca6 P1 P2 P3 P4 P5 P6 ( 0.2029 , 0.0 , 0.25 ) ( -0.2029 , -0.2029 , 0.25 ) ( 0.0 , 0.2029 , 0.25 ) ( -0.2029 , 0.0 , -0.25 ) ( 0.2029 , 0.2029 , -0.25 ) ( 0.0 , -0.2029 , -0.25 ) ( 0.6215 , 0.0 , 0.25 ) ( -0.6215 , -0.6215 , 0.25 ) ( 0.0 , 0.6215 , 0.25 ) ( -0.6215 , 0.0 , -0.25 ) ( 0.6215 , 0.6215 , -0.25 ) ( 0.0 , -0.6215 , -0.25 ) 1. Matrix elements The matrix elements given below closely follows Eq. 2.2. The position vectors sα of each orbital are listed in Table I. We illustrate each hopping terms in Fig. 7. a. Ca-Ca matrix elements In the HCa block, we can further divide orbitals in each atomic species into those belong to the lower layer and the upper layer, Ca, H uu where sub-blocks H ll The Hamiltonian matrix H ll Ca are 3× 3 matrices. Ca have 3 inde- pendent intra-layer hopping terms, the nearest-neighbor, second nearest-neighbor, and third nearest neighbor hop- pings, td2, td4, and td5 as shown in Fig. 7. Ca and H uu (cid:19) (cid:18) H ll Ca H lu Ca† H uu Ca H lu Ca Ca, and H lu hc,ll 11 hc,ll 21 hc,ll hc,ll 31 hc,ll hc,ll 12 hc,ll 22 hc,ll 32 hc,ll 23 13 33  , H ll Ca = where 12 = eik·s1,2(td2 + td4c4 hc,ll 13 = eik·s1,3(td2 + td4c4 hc,ll 23 = eik·s2,3(td2 + td4c4 hc,ll 22 = hc,ll 12 + td5c5 13 + td5c5 23 + td5c5 12) 13) 23) and hc,ll 11 = hc,ll 33 = µd. We define phase factors (A2) (A3) (A4) (A5) where 12 = eik·s7,8(tp5 + tp1a1 hp,ll 12) 13 = eik·s7,9(tp5 + tp1a1 hp,ll 13) 23 = eik·s8,9(tp5 + tp1a1 hp,ll 23) 12 (A15) (A16) (A17) αβ are phase factors from 22 = hp,ll 11 = hp,ll 33 = µp. ai and hp,ll hopping tpi with matrix index α and β, 12 = eik·R100 + eik·R110 a1 13 = eik·R100 + eik·R0−10 a1 23 = eik·R−1−10 + eik·R0−10. a1 H uu P can be defined similarly. H lu P contains 3 independent inter-plane hopping inte- grals tp2, tp3, and tp4. hp,lu 11 hp,lu 21 hp,lu 31  , hp,lu 12 hp,lu 22 hp,lu 32 hp,lu 13 hp,lu 23 hp,lu 33 H lu P = 11 = tp2eik·s7,10 (eik·R100 + eik·R101) hp,lu (A22) 22 = tp2eik·s8,11 (eik·R−1−10 + eik·R−1−11) (A23) hp,lu 33 = tp2eik·s9,12 (eik·R010 + eik·R011), hp,lu (A24) 12 = eik·s7,11 (tp3c0 + tp4a4 hp,lu 12) 13 = eik·s7,12 (tp3c0 + tp4a4 hp,lu 13) 21 = eik·s8,10 (tp3c0 + tp4a4 hp,lu 21) 23 = eik·s8,12 (tp3c0 + tp4a4 hp,lu 23) 31 = eik·s9,10 (tp3c0 + tp4a4 hp,lu 31) 32 = eik·s9,11 (tp3c0 + tp4a4 hp,lu 32). The corresponding phase factors are, c0 = 1 + eik·R001 12 = eik·R0−10 + eik·R0−11 a4 13 = eik·R110 + eik·R111 a4 21 = eik·R0−10 + eik·R0−11 a4 23 = eik·R−100 + eik·R−101 a4 31 = eik·R110 + eik·R111 a4 32 = eik·R−100 + eik·R−101. a4 c. Ca-P matrix elements (A13) (A18) (A19) (A20) (A21) (A25) (A26) (A27) (A28) (A29) (A30) (A31) (A32) (A33) (A34) (A35) (A36) (A37) FIG. 7. (a) Definitions of hopping integrals between two Ca orbitals. (b) Definitions of hopping integrals between two P orbitals and one Ca and one P orbital. Orbitals in the first Brillouin zone are labeled. Dark blue and red color represent orbitals in the lower plane while light blue and pink orbitals lies in the upper plane. ci αβ for hopping integral tdi with matrix indices α and β 12 = eik·R100 + eik·R110 c4 13 = eik·R100 + eik·R0−10 c4 23 =eik·R−1−10 + eik·R0−10 c4 12 = eik·R010 + eik·R0−10 c5 13 = eik·R110 + eik·R−1−10 c5 23 = eik·R−100 + eik·R100 c5 (A6) (A7) (A8) (A9) (A10) (A11) where and where Rijk is the lattice vector connecting the unit cell in the (i, j, k) direction and sl,m = sm − sl. H uu Ca is defined similarly. H lu Ca contains 2 independent inter-plane hopping inte- grals td1 and td3. H lu Ca = c0 td3eik·s1,4 td1eik·s1,5 td1eik·s1,6 td1eik·s2,4 td3eik·s2,5 td1eik·s2,6 td1eik·s3,4 td1eik·s3,5 td3eik·s3,6  ,(A12) where c0 is (1 + eik·R001). b. P-P matrix elements We apply similar division of layer indices for HP ma- trix. HP = (cid:18) H ll (cid:19) . P H lu P † H uu P H lu P P and H uu The Hamiltonian matrix H ll P have 2 indepen- dent hopping integrals tp1 and tp5 coupling orbitals in the same layer. hp,ll 13 11 hp,ll hp,ll 21 hp,ll hp,ll 31 hp,ll 12 hp,ll 22 hp,ll 32 hp,ll 23 33  , H ll P = Finally, the inter-orbital hopping matrix V describes the hybridization between Ca and P orbitals. We again divide V into four 3× 3 matrices according to their layer indices, (A14) V = (A38) (cid:18) V ll V lu (cid:19) V ul V uu Ca1Ca4Ca3Ca6Ca2Ca5td1td2td3td4td5P1P4P3P6P2P5tp4tp2tp3tp1tp5tdp1tdp3tdp4tdp2(a)(b). The V ll and V uu blocks only have diagonal elements, which can be written down with the hopping integrals tdp4, eik·s1,7 eik·s4,10 0 0 0 0  (A39)  , (A40) 0 eik·s2,8 0 0 0 eik·s3,9 0 eik·s5,11 0 0 0 eik·s6,12 V ll = b1 tdp4 and V uu = −b1 tdp4 where the phase factor b1 = (eik·R001 − eik·R00−1). We note that the minus sign in V uu is due to the opposite orientation of px orbitals in the different layer. Also due to the opposite inversion symmetry eigenvalue of the px and the dz2 orbital, hopping integrals vanish if both of them lie in the same plane. Hence, only hopping integrals from different unit cell contributes to diagonal elements. Inter-layer coupling tdp1, tdp2, and tdp3 contributes to V lu and V ul matrices, V lu 11 V lu V lu 21 V lu V lu 31 V lu 12 V lu 13 22 V lu 23 32 V lu 33  , (A41) V lu = where 11 = −tdp1eik·s1,10(eik·R101 − eik·R100) V lu (A42) 22 = −tdp1eik·s2,11(eik·R−1−11 − eik·R−1−10 )(A43) V lu 33 = −tdp1eik·s3,12(eik·R011 − eik·R010). V lu (A44) and off-diagonal elements 12 = eik·s1,11(tdp2b2 + tdp3b3 V lu 12) 13 = eik·s1,12(tdp2b2 + tdp3b3 V lu 13) 21 = eik·s2,10(tdp2b2 + tdp3b3 V lu 21) 23 = eik·s2,12(tdp2b2 + tdp3b3 V lu 23) 31 = eik·s3,10(tdp2b2 + tdp3b3 V lu 31) 32 = eik·s3,11(tdp2b2 + tdp3b3 V lu 32). The phase factors are, b2 = −(eik·R001 − eik·R000) 12 = −(eik·R0−11 − eik·R0−10 ) b3 13 = −(eik·R111 − eik·R110) b3 21 = −(eik·R0−11 − eik·R0−10 ) b3 23 = −(eik·R−101 − eik·R−100) b3 31 = −(eik·R111 − eik·R110) b3 32 = −(eik·R−101 − eik·R−100) b3 (A45) (A46) (A47) (A48) (A49) (A50) (A51) (A52) (A53) (A54) (A55) (A56) (A57) αβ belongs to hopping tdpi between index α and , where bi β. Similarly, we have V ul = V ul 11 V ul V ul 21 V ul V ul 31 V ul 12 V ul 13 22 V ul 23 32 V ul 33  , (cid:1) (cid:88) (cid:104)uk,j∂kuk,j(cid:105)dk Ej <EF −π (cid:104)uk,j∂kuk,j(cid:105)dk + (cid:90) 0 (cid:88) (cid:88) Ej <EF Ej <EF 0 P = −i(cid:0)(cid:90) π (cid:90) π (cid:90) π (cid:88) = −i +i 0 0 = 13 where 11 = −tdp1eik·s4,7 (eik·R−10−1 − eik·R−100) (A59) V ul 22 = −tdp1eik·s5,8 (eik·R11−1 − eik·R1110) V ul (A60) 33 = −tdp1eik·s6,9 (eik·R0−1−1 − eik·R0−10 ). (A61) V ul and off-diagonal elements ∗ 12 = eik·s4,8(tdp2b2 + tdp3b3 V ul 12) ∗ 13 = eik·s4,9(tdp2b2 + tdp3b3 V ul 13) ∗ 21 = eik·s5,7(tdp2b2 + tdp3b3 V ul 21) 23 = eik·s5,9(tdp2b2 + tdp3b3 ∗ V ul 23) ∗ 31 = eik·s6,7(tdp2b2 + tdp3b3 V ul 31) ∗ 32 = eik·s6,8(tdp2b2 + tdp3b3 V ul 32) where ∗ denotes the complex conjugate. , (A62) (A63) (A64) (A65) (A66) (A67) 2. Tight-binding parameters We list parameters of the tight-binding model in the unit of eV below. The hopping integrals between two Ca orbitals are td1 = −0.2031, td2 = −0.6388, td3 = −0.0786, td4 = −0.216, and td5 = 0.0516. Those be- tween two P orbitals are tp1 = −0.041, tp2 = −0.4077, tp3 = −0.0479, tp4 = −0.1067, and tp5 = 0.0548. Fi- nally, the hopping amplitudes between Ca and P orbitals are tdp1 = 0.1415, tdp2 = 0.0379, tdp3 = 0.0443 and tdp4 = 0.0376. The chemical potentials are µd = 2.6808 and µp = −1.2186 for Ca and P respectively. Appendix B: topological number and Berry phase To show that the Berry phase in the kz direction is quantized and is related to n+ occ in Eq. (2.11), we recall some basic facts of inversion symmetry. We assume no degeneracies so the inversion symmetry acts the wave- functions uk,j(cid:105) in the unique expression (k ≡ kz) u−k,j(cid:105) = e −iαj k Rkuk,j(cid:105) (B1) The reflection operator obeys R−kRk = ±1 for spinless/spin-1/2 systems respectively. For spin-1/2, we † redefine Rk → −iRk so that R−kRk = 1. Also, R kRk = 1. Let us rewrite the Berry phase (cid:104)uk,jR −iαj k Rkuk,j(cid:105)dk † keiαj (cid:90) π k ∂ke (cid:88) π − αj (αj 0) + i (cid:104)uk,jR k∂kRkuk,j(cid:105)dk † (B2) (A58) Ej <EF 0 Ej <EF The reflection symmetry operator has a generic block- diagonalized from [3] Rk = Ui1j1ein1k ⊕ Ui2j2 ein2k ⊕ . . . ⊕ UiN jN einN k,(B3) where Uiljl constant a ≡ 1. is a unitary matrix and we use the lattice R Hence, ∂R is just mπ, where m is an integer † k∂kRk = in1δi1j1 ⊕ in2δi2j2 ⊕ . . . ⊕ inN δiN jN .(B4) (cid:90) π (cid:88) k∂kRkuk,j(cid:105)dk = −(cid:88) (cid:104)uk,jR nlmlπ, (B5) † i 0 Ej <EF l=1 where ml is the number of the occupied states in Uiljl block. Consider left hand side of Eq. (2.11) (cid:0)(cid:104)uπ,jRπuπ,j(cid:105) − (cid:104)u0,jR0u0,j(cid:105)(cid:1) occ − n+,0 n+,π occ = = 1 2 1 2 (cid:88) (cid:88) Ej <0 Ej <0 (eiαj π − eiαj 0) (B6) † k0 = Rk0, where k0 = −k0, such as 0, π so eiαj k0 = Since R ±1 and then (cid:88) i(cid:80) Ej <EF 14 (cid:105)dk (cid:104)uk,j∂kuk,j(cid:105)dk phase is quantized P = − i = − i (cid:73) (cid:88) (cid:73) (cid:88) (cid:88) Ej <EF Ej <EF (cid:104)u ∗ k,lj U k∂keiβj kUu ∗ k,lj † e −iβj (cid:73) (cid:88) = + − βj−) − i (βj Ej <EF Ej <EF (cid:104)u ∗ k,lj ∂ku ∗ k,lj (cid:105)dk, (B9) where βj∓ represent the phases at the beginning and end of the integration path respectively. The first summation is 2nπ. Since the jth and the ljth states share the same energy and each state in the second summation should be orthogonal, we safely change the index lj to j in the summation. We use the identity k,j(cid:105) = (cid:104)∂kuk,juk,j(cid:105) = −(cid:104)uk,j∂kuk,j(cid:105)(B10) k,j∂ku (cid:104)u ∗ ∗ The Berry phase is quantized (cid:88) P = + − βj−) = nπ (βj (B11) occ − n+,0 n+,π occ ≡ 1 π π − αj (αj 0) (mod 2) Ej <EF (B7) occ = e occ −n+,0 Thus, (−1)n+,π . By using Eq. (B2), (B7), we obtain the relation in Eq. (2.11) be- tween the topological invariants and the Berry phase P is either 0 or π (mod 2π) since 2nπ phase can be cancelled by a large U (1) gauge transformation. Ej <EF (αj π−αj 0) ***** Similarly, IT symmetry, the composite symmetry of time-reversal and inversion, also quantizes the Berry phase when dk is integrated along any closed loop. Since time-reversal and inversion operators both flip k, the composite symmetry operators keep the same k. The integration path can be arbitrarily chose to preserve IT symmetry. Unlike the Berry phase under reflection sym- metry, the integration path should be strictly in the kz direction to preserve reflection symmetry. IT symmetry operator is the combination of a unitary matrix and complex conjugation TI = U K; the unitary matrix U might be k-dependent. To simplify the prob- lem, we assume U is k-independent, which is the case of Ca3P2 tight-binding model. The relation of wavefunc- tions under IT symmetry is given by k Uuk,lj(cid:105) (B8) We note that uk,j(cid:105) and uk,lj(cid:105) in the same energy level might be orthogonal or identical. Let us show the Berry uk,j(cid:105) = eiβj Appendix C: Toy model of topological nodal lines The tight-binding model of Ca3P2 provides the way to investigate topological nodal lines in a realistic model. However, to capture the physical features of the nodal lines only the low energy theory is needed. We extend the low energy theory to a simple lattice model in order to provide an economic way to investigate topological nodal lines. Although the space group of Ca3P2 is P 63/mcm, we consider square lattice and extend and transfer the low energy equation (3.1) with spins to the lattice form H lattice spinful(k) = +(cid:0) ν(cid:48) ν(cid:48) (cid:107) a2 g(k(cid:107))τzσ0 + a2 g(k(cid:107)) + V0 0 νz c sin ckzτyσ0 (cid:1)τ0σ0 + Hcos kz (C1) where g(k(cid:107)) = 1 + cos ak0 − cos akx − cos aky, the lattice constants a = 8.26 Aand c = 6.84 A, ν(cid:48) , and ν(cid:48) 0 = 2ν0ak0 sin ak0 Hcoskz = (1 − cos ckz)(cid:0)Zτ τzσ0 + Z0τ0σ0 . Furthermore, we define 2ν(cid:107)ak0 sin ak0 (C2) (cid:107) = (cid:1) in the simplest form so that the Berry phase inside the nodal ring is nonzero when the spin degree of freedom is neglected. By fitting the energy spectrum from the DFT calculation as kz = 0, π, we have Zτ = 0.287 eV and Z0 = −0.156 eV. 15 tion along the 1d BZ, is quantized. The toy model in momentum space can be simply written as H(p) = (µ + cos p)σx + sin pσy + δ cos p1, (D1) which preserves inversion symmetry by satisfying Eq. (2.6) with inversion symmetry operator I = σx. Bro- ken chiral symmetry caused by δ cos p1 destroys the def- inition of winding number so the Berry phase is the only valid topological invariant. Furthermore, by Eq. (2.4) time-reversal symmetry is preserved with time-reversal operator T = K. IT symmetry also guarantees the quan- tized Berry phase. By choosing µ = 0.5 and δ = 0.1, the Berry phase P = π leads to the presence of charge ±e/2 at each end, which is one of the topological fea- tures of this inversion symmetric insulator. The sign of the charge depends on the occupation of the end mode. Hence, we can numerically compute the charge on one of the ends. If the charge is no longer ±e/2 under disorder, the topology is destroyed by disorders. We add inversion symmetry breaking disorder rjc † jσzcj to the Hamiltonian in real space (cid:88) (cid:2) µ H = 2 j † † jσxcj + c j+1 c σx + δ1 + iσy 2 cj + h.c.],(D2) where rj is a random number from −∆ + m to ∆ + m. When m = 0, the average < rj >= 0 indicates the aver- age disorder preserves inversion symmetry. As shown in Eq. (8) (a) when m = 0, the charge on one end is ±e/2 on average. When inversion symmetry is broken on average, the charge is no longer quantized and then the topologi- cal phase is destroyed. Fig. 8 (b) the standard deviation of the disorder is proportional to the deviation of the end disorder. Thus, the quantized end charges survive when disorder on average is zero and the fluctuation is small enough. FIG. 8. The numerical result for the 1d inversion-symmetric topological insulator with 200 sites. We consider the half- filling scenario and compute the absolute value of charge ac- cumulated on the first 10 sites under gaussian disorder as fixed disorder average m and disorder random deviation ∆ 3000 times. (a) As ∆ = 0.02, the end charge is not quantized when the average disorder is not zero. In the special condition that the average disorder vanishes so inversion symmetry on average is preserved, the end charge on average is ±e/2. (b) The standard deviation of the disorder grows as the deviation of the end disorder grows. Appendix D: Quantized end charge under disorders To understand the robustness of the topology under disorder we consider the toy model of a 1d inversion- symmetric topological insulator. We note that in a 1d system inversion symmetry is equivalent to reflection symmetry; reflection symmetric nodal lines with fixed kx, ky is equivalent to the 1d inversion symmetric topo- logical insulator; the Berry phase, which is the integra- [1] M. Z. Hasan and C. L. Kane, Rev. Mod. Phys. 82, 3045 [9] Y. Ando, Journal of the Physical Society of Japan 82, (2010). 102001 (2013). [2] X.-L. Qi and S.-C. Zhang, Rev. Mod. Phys. 83, 1057 [10] M. Z. Hasan and J. E. Moore, Annu. Rev. Condens. Mat- (2011). ter Phys. 2, 55 (2011). [3] C.-K. Chiu, J. C. Y. Teo, A. P. Schnyder, and S. Ryu, ArXiv e-prints (2015), arXiv:1505.03535 [cond-mat.mes- hall]. [4] A. P. Schnyder, S. Ryu, A. Furusaki, and A. W. W. Ludwig, Phys. Rev. B 78, 195125 (2008). [11] M. Z. Hasan, S.-Y. Xu, and M. Neupane, Topologi- cal Insulators, Topological Dirac semimetals, Topological Crystalline Insulators, and Topological Kondo Insulators (John Wiley and Sons, 2015). [12] X.-L. Qi, T. L. Hughes, and S.-C. Zhang, Phys. Rev. B [5] S. Ryu, A. P. Schnyder, A. Furusaki, and A. W. W. 78, 195424 (2008). Ludwig, New J. Phys. 12, 065010 (2010). [13] X.-L. Qi, T. L. Hughes, and S.-C. Zhang, Nat. Phys. 4, [6] D. Hsieh, D. Qian, L. Wray, Y. Xia, Y. S. Hor, R. J. Cava, 273 (2008), 0710.0730. and M. Z. Hasan, Nature (London) 452, 970 (2008). [7] D. Hsieh, Y. Xia, D. Qian, L. Wray, J. H. Dil, F. Meier, J. Osterwalder, L. Patthey, J. G. Checkelsky, N. P. Ong, A. V. Fedorov, H. Lin, A. Bansil, D. Grauer, Y. S. Hor, R. J. Cava, and M. Z. Hasan, Nature 460, 1101 (2009). [8] Y. Xia, D. Qian, D. Hsieh, L. Wray, A. Pal, H. Lin, A. Bansil, D. Grauer, Y. S. Hor, R. J. Cava, and M. Z. Hasan, Nat. Phys. 5, 398 (2009). [14] I. Garate and M. Franz, Phys. Rev. B 81, 172408 (2010). [15] S.-Y. Xu, C. Liu, S. K. Kushwaha, R. Sankar, J. W. Krizan, I. Belopolski, M. Neupane, G. Bian, N. Ali- doust, T.-R. Chang, H.-T. Jeng, C.-Y. Huang, W.-F. Tsai, H. Lin, P. P. Shibayev, F.-C. Chou, R. J. Cava, and M. Z. Hasan, Science 347, 294 (2015). [16] S. Matsuura, P.-Y. Chang, A. P. Schnyder, and S. Ryu, New J. Phys. 15, 065001 (2013). 00.0250.05000.0050.0100.0150.0200.0250.0300.0350.040disorder deviation ∆charge deviation on the end00.050.100.380.400.420.440.460.480.500.52disorder average mcharge deviation on the end(a)(b)16 [17] Y. X. Zhao and Z. D. Wang, Phys. Rev. Lett. 110, 240404 (2013). [18] C.-K. Chiu and A. P. Schnyder, Phys. Rev. B 90, 205136 (2014). [19] C. Fang, Y. Chen, H.-Y. Kee, and L. Fu, Phys. Rev. B 92, 081201 (2015). [20] A. A. Burkov, M. D. Hook, and L. Balents, Phys. Rev. B 84, 235126 (2011). [21] Y. X. Zhao and Z. D. Wang, Phys. Rev. B 89, 075111 (2014). [22] O. Vafek and A. Vishwanath, Annual Review of Con- densed Matter Physics 5, 83 (2014). [23] W. Witczak-Krempa, G. Chen, Y. B. Kim, and L. Ba- lents, Annual Review of Condensed Matter Physics 5, 57 (2014). [40] S.-Y. Xu, C. Liu, S. K. Kushwaha, T.-R. Chang, J. W. Krizan, R. Sankar, C. M. Polley, J. Adell, T. Balasub- ramanian, K. Miyamoto, N. Alidoust, G. Bian, M. Neu- pane, I. Belopolski, H.-T. Jeng, C.-Y. Huang, W.-F. Tsai, H. Lin, F. C. Chou, T. Okuda, A. Bansil, R. J. Cava, and M. Z. Hasan, ArXiv e-prints (2013), arXiv:1312.7624 [cond-mat.mes-hall]. [41] S.-Y. Xu, C. Liu, S. K. Kushwaha, R. Sankar, J. W. Krizan, I. Belopolski, M. Neupan, e, G. Bian, N. Ali- doust, T.-R. Chang, H.-T. Jeng, C.-Y. Huang, W.-F. Tsai, H. Lin, P. P. Shibayev, F. Chou, R. J. Cava, and M. Z. Hasan, Science (2014), 10.1126/science.1256742. [42] C.-K. Chiu and A. P. Schnyder, ArXiv e-prints (2015), arXiv:1501.06820 [cond-mat.mes-hall]. [43] A. A. Zyuzin and A. A. Burkov, Phys. Rev. B 86, 115133 [24] P. Hosur and X. Qi, Comptes Rendus Physique 14, 857 (2012). (2013). [44] C.-X. Liu, P. Ye, and X.-L. Qi, Phys. Rev. B 87, 235306 [25] S.-Y. Xu, I. Belopolski, N. Alidoust, M. Neupane, G. Bian, C. Zhang, R. Sankar, G. Chang, Z. Yuan, C.-C. Lee, S.-M. Huang, H. Zheng, J. Ma, D. S. Sanchez, B. Wang, A. Bansil, F. Chou, P. P. Shibayev, H. Lin, S. Jia, and M. Z. Hasan, Science 349, 613 (2015), http://www.sciencemag.org/content/349/6248/613.full.pdf. [26] L. Lu, Z. Wang, D. Ye, L. Ran, L. Fu, J. D. Joannopou- and M. Soljaci´c, Science 349, 622 (2015), los, http://www.sciencemag.org/content/349/6248/622.full.pdf. [27] B. Q. Lv, H. M. Weng, B. B. Fu, X. P. Wang, H. Miao, J. Ma, P. Richard, X. C. Huang, L. X. Zhao, G. F. Chen, Z. Fang, X. Dai, T. Qian, and H. Ding, Phys. Rev. X 5, 031013 (2015). (2013). [45] S. A. Parameswaran, T. Grover, D. A. Abanin, D. A. Pesin, and A. Vishwanath, Physical Review X 4, 031035 (2014). [46] M. M. Vazifeh and M. Franz, Phys. Rev. Lett. 111, 027201 (2013). [47] N. B. Kopnin, T. T. Heikkila, and G. E. Volovik, Phys. Rev. B 83, 220503 (2011). [48] E. Tang and L. Fu, Nat Phys 10, 964 (2014). [49] G. Z. Magda, X. Jin, I. Hagymasi, P. Vancso, Z. Osvath, P. Nemes-Incze, C. Hwang, L. P. Biro, and L. Tapaszto, Nature 514, 608 (2014). [50] Y. Huh, E.-G. Moon, and Y. B. Kim, ArXiv e-prints [28] Z. Gao, M. Hua, H. Zhang, and X. Zhang, ArXiv e-prints (2015), arXiv:1506.05105. (2015), arXiv:1507.07504 [cond-mat.mes-hall]. [51] J.-W. Rhim and Y. B. Kim, Phys. Rev. B 92, 045126 [29] Y. Kim, B. J. Wieder, C. L. Kane, and A. M. Rappe, (2015). Phys. Rev. Lett. 115, 036806 (2015). [52] M. N. Ali, Q. D. Gibson, T. Klimczuk, and R. J. Cava, [30] H. Weng, Y. Liang, Q. Xu, R. Yu, Z. Fang, X. Dai, and Phys. Rev. B 89, 020505 (2014). Y. Kawazoe, Phys. Rev. B 92, 045108 (2015). [31] Y. Chen, Y. Xie, S. A. Yang, H. Pan, F. Zhang, (2015), and S. Zhang, ArXiv e-prints M. L. Cohen, arXiv:1505.02284 [cond-mat.mtrl-sci]. [32] S. Jeon, B. B. Zhou, A. Gyenis, B. E. Feldman, I. Kimchi, A. C. Potter, Q. D. Gibson, R. J. Cava, A. Vishwanath, and A. Yazdani, Nat. Mater. 13, 851 (2014). [33] Z. Wang, H. Weng, Q. Wu, X. Dai, and Z. Fang, Phys. Rev. B 88, 125427 (2013). [34] M. Neupane, S.-Y. Xu, R. Sankar, N. Alidoust, G. Bian, C. Liu, I. Belopolski, T.-R. Chang, H.-T. Jeng, H. Lin, A. Bansil, F. Chou, and M. Z. Hasan, Nat. Commun. 5, 3786 (2014). [35] S. Borisenko, Q. Gibson, D. Evtushinsky, V. Zabolot- nyy, B. Buchner, and R. J. Cava, Phys. Rev. Lett. 113, 027603 (2014). [36] Z. K. Liu, J. Jiang, B. Zhou, Z. J. Wang, Y. Zhang, H. M. Weng, D. Prabhakaran, S.-K. Mo, H. Peng, P. Dudin, T. Kim, M. Hoesch, Z. Fang, X. Dai, Z. X. Shen, D. L. Feng, Z. Hussain, and Y. L. Chen, Nat. Mater. 13, 677 (2014). [37] T. Liang, Q. Gibson, M. N. Ali, M. Liu, R. J. Cava, and N. P. Ong, Nat Mater 14, 280 (2015). [38] Z. K. Liu, B. Zhou, Y. Zhang, Z. J. Wang, H. M. Weng, D. Prabhakaran, S.-K. Mo, Z. X. Shen, Z. Fang, X. Dai, Z. Hussain, and Y. L. Chen, Science 343, 864 (2014). [39] Z. Wang, Y. Sun, X.-Q. Chen, C. Franchini, G. Xu, H. Weng, X. Dai, and Z. Fang, Phys. Rev. B 85, 195320 (2012). [53] G. Bian, T.-R. Chang, R. Sankar, S.-Y. Xu, H. Zheng, T. Neupert, C.-K. Chiu, S.-M. Huang, G. Chang, I. Be- lopolski, D. S. Sanchez, M. Neupane, N. Alidoust, C. Liu, B. Wang, C.-C. Lee, H.-T. Jeng, A. Bansil, F. Chou, H. Lin, and M. Zahid Hasan, ArXiv e-prints (2015), arXiv:1505.03069 [cond-mat.mes-hall]. [54] G. Bian, T.-R. Chang, H. Zheng, S. Velury, S.-Y. Xu, T. Neupert, C.-K. Chiu, D. S. Sanchez, I. Belopolski, N. Alidoust, P.-J. Chen, G. Chang, A. Bansil, H.-T. Jeng, H. Lin, and M. Zahid Hasan, ArXiv e-prints (2015), arXiv:1508.07521 [cond-mat.mes-hall]. [55] R. Yu, H. Weng, Z. Fang, X. Dai, and X. Hu, Phys. Rev. Lett. 115, 036807 (2015). [56] L. M. Schoop, M. N. Ali, C. Strasser, V. Duppel, S. S. P. Parkin, B. V. Lotsch, and C. R. Ast, ArXiv e-prints (2015), arXiv:1509.00861 [cond-mat.mtrl-sci]. [57] L. S. Xie, L. M. Schoop, E. M. Seibel, Q. D. Gib- (2015), son, W. Xie, and R. J. Cava, ArXiv e-prints arXiv:1504.01731 [cond-mat.mtrl-sci]. [58] J. C. Y. Teo, L. Fu, and C. L. Kane, Phys. Rev. B 78, 045426 (2008). [59] L. Fu, Phys. Rev. Lett. 106, 106802 (2011). [60] C.-K. Chiu, H. Yao, and S. Ryu, Phys. Rev. B 88, 075142 (2013). [61] T. Morimoto and A. Furusaki, Phys. Rev. B 88, 125129 (2013). [62] K. Shiozaki and M. Sato, Phys. Rev. B 90, 165114 (2014). [63] P. Blaha, K. Schwarz, G. K. Madsen, D. Kasnicka, and J. Luitz, Wien2k, An Augmented Plane Wave + Lo- 17 cal Orbitals Program for Calculating Crystal Properties, edited by K. Schwarz (Techn. Universitat Wien, Austria, 2001). [73] A. P. Schnyder and P. M. R. Brydon, Journal of Physics: Condensed Matter 27, 243201 (2015). [74] P. Esquinazi, T. Heikkila, Y. Lysogorskiy, D. Tayurskii, [64] J. P. Perdew, K. Burke, and M. Ernzerhof, Phys. Rev. and G. Volovik, JETP Letters 100, 336 (2014). Lett. 77, 3865 (1996). [75] D. Vanderbilt and R. D. King-Smith, Phys. Rev. B 48, [65] L. Nordheim, Annalen der Physik 401, 607 (1931). [66] J. Kunes, R. Arita, P. Wissgott, A. Toschi, H. Ikeda, and K. Held, Computer Physics Communications 181, 1888 (2010). [67] N. Marzari, A. A. Mostofi, J. R. Yates, I. Souza, and D. Vanderbilt, Rev. Mod. Phys. 84, 1419 (2012). [68] M. P. L. Sancho, J. M. L. Sancho, J. M. L. Sancho, and J. Rubio, Journal of Physics F: Metal Physics 15, 851 (1985). 4442 (1993). [76] T. L. Hughes, E. Prodan, and B. A. Bernevig, Phys. Rev. B 83, 245132 (2011). [77] This relation also holds if there is spin degeneracy, in which case ∂R is zero since the two spin components give opposite contribution. [78] T. Fukui, Y. Hatsugai, and H. Suzuki, Journal of the Physical Society of Japan 74, 1674 (2005). [79] S. T. Ramamurthy and T. L. Hughes, ArXiv e-prints [69] L. Lu, L. Fu, J. D. Joannopoulos, and M. Soljacic, Nat (2015), arXiv:1508.01205. Photon 7, 294 (2013). [80] M. Diez, D. I. Pikulin, I. C. Fulga, and J. Tworzyd(cid:32)lo, [70] A. P. Schnyder and S. Ryu, Phys. Rev. B 84, 060504 New Journal of Physics 17, 043014 (2015). (2011). [71] Y. Tanaka, M. Sato, and N. Nagaosa, Journal of the Physical Society of Japan 81, 011013 (2012). [72] A. P. Schnyder, C. Timm, and P. M. R. Brydon, Phys. Rev. Lett. 111, 077001 (2013). [81] A. Yamakage, Y. Yamakawa, Y. Tanaka, and (2015), arXiv:1510.00202 Y. Okamoto, ArXiv e-prints [cond-mat.mes-hall]. 

1511.01742

1

1511

From Coupled Rashba Electron and Hole Gas Layers to 3D Topological Insulators

We introduce a system of stacked two-dimensional electron and hole gas layers with Rashba spin orbit interaction and show that the tunnel coupling between the layers induces a strong three- dimensional (3D) topological insulator phase. At each of the two-dimensional bulk boundaries we find the spectrum consisting of a single anistropic Dirac cone, which we show by analytical and numerical calculations. Our setup has a unit-cell consisting of four tunnel coupled Rashba layers and presents a synthetic strong 3D topological insulator and is distinguished by its rather high experimental feasibility.

cond-mat.mes-hall

cond-mat

From Coupled Rashba Electron and Hole Gas Layers to 3D Topological Insulators Luka Trifunovic,1, 2 Daniel Loss,1 and Jelena Klinovaja1 1Department of Physics, University of Basel, Klingelbergstrasse 82, CH-4056 Basel, Switzerland 2Dahlem Center for Complex Quantum Systems and Physics Department, Freie Universitat Berlin, Arnimallee 14, 14195 Berlin, Germany (Dated: April 29, 2021) We introduce a system of stacked two-dimensional electron and hole gas layers with Rashba spin orbit interaction and show that the tunnel coupling between the layers induces a strong three- dimensional (3D) topological insulator phase. At each of the two-dimensional bulk boundaries we find the spectrum consisting of a single anistropic Dirac cone, which we show by analytical and numerical calculations. Our setup has a unit-cell consisting of four tunnel coupled Rashba layers and presents a synthetic strong 3D topological insulator and is distinguished by its rather high experimental feasibility. PACS numbers: 73.21.Ac; 73.20.-r; 03.65.Vf Introduction. Since the discovery of the quantum Hall effect there has been immense theoretical interest fo- cused on understanding topological phases of quantum matter [1, 2]. The interest was not solely concentrated on classification of these novel phases [3], which goes beyond the Landau paradigm of phase transitions, but also on potential applications of the topologically ordered phases, in particular for storing quantum information in a manner that is resilient to local imperfections [4]. Addi- tionally, the electronic surface states of a strong topolog- ical insulator (TI) [2, 5], being an example of a 3D topo- logical phase of matter, forms a two-dimensional (2D) topological metal, which is 'half' of an ordinary metal [2]. Such 2D topological metals are notable for the fact that their electrons cannot be localized even in presence of strong disorder, as long as the bulk energy gap of the parent strong 3D TI is intact [6]. There are strong indications that certain materials, such as semiconducting alloys, behave as strong 3D TIs [2]. Despite great success in this field, both theo- retically and experimentally, there are still certain issues that need to be resolved, in particular that strong TIs suf- fer from bulk conduction due to chemical imperfections. Thus, there is a strong need for synthetic materials where one has enough control over the system parameters in or- der to achieve a topological phase with a sufficiently large bulk gap which excludes bulk conduction. One of the very successful approaches for theoretically constructing 2D topological phases of matter is using anisotropic hopping or a coupled wire construction [7 -- 21]. Apart from being very intuitive, this approach al- lows non-perturbative treatment of the electron-electron interactions and is thus suitable for study of fractional topological phases. Recently, a strong effort was made to extend this approach to the study of 3D TIs, where topological phases related to weak TIs were obtained, as well as Weyl semimetal phases [21 -- 23]. Despite the great theoretical insight this approach gives, its main drawback in the case of 3D systems is that the resulting setups are FIG. 1. Panel a) shows the setup consisting of a stack of layers arranged in the xy-plane and tunnel coupled along the z axis. The layers colored in green (orange) denote electron (hole) 2DEGs with Rashba SOI and at chemical potential µτ . The brightness of the color encodes two possible values of the Rashba SOI ατ . Panel b) shows the dispersion of a 2DEG with Rashba SOI. rather complex and thus not easy to realize experimen- tally. In this paper we take a different approach, instead of coupled wires [7 -- 22] we introduce a construction of coupled 2D layers, see Fig. 1. Each layer is a simple 2D electron gas (2DEG) with Rashba spin-orbit interac- tion (SOI) [24]. By generalizing the coupled wires ap- proach [17] to coupled layers we arrive at a rather simple realization of a strong 3D TI. Model. We consider a system consisting of tunnel cou- pled layers of 2DEGs stacked along the z-axis, see Fig. 1. In each 2DEG we include SOI and we assume it to be of Rashba type [25]. In our model, we work with two differ- ent values of SOI that could be chosen almost arbitrary (see below) and do not require special tuning. In contrast to that, the chemical potential µτ in each layer should be individually tuned to the value determined by the corre- sponding SOI. Our setup has a unit cell consisting of four Rashba 2DEG layers. A single 2DEG layer with Rashba SOI is described by a)unitcellb)the following Hamiltonian [24] H0 = −2(∂2 x + ∂2 y)/2m0 − iα(σx∂y − σy∂x), 2 (1) where α is the strength of the Rashba SOI and m0 the electron mass in the given band. We can diagonalize the above Hamiltonian by taking the local spin quantization axis s = (− sin θ, cos θ) to be always perpendicular to the momentum k = (kx, ky) ≡ k(cos θ, sin θ), E∓(k) = 2k2/2m0 ∓ αk, (2) where the upper (lower) sign corresponds to the spin ori- entation being along (opposite to) s chosen for α > 0 and to the lower (higher) energy for a fixed k, where the corresponding spinors are given by (cid:19) (cid:18) 1±ieiθ ∓; θ(cid:105) = 1√ 2 . (3) We note here that the spin orientation is clockwise (an- ticlockwise) for +; θ(cid:105) (−; θ(cid:105)). The dispersion relation Eq. (2) is depicted in Fig. 1b, and the shape of the Fermi surfaces and the spin orientations in Fig. 2b. The setup we consider herein consists of four stacked layers composing the unit cell, which then periodically repeats in z-direction with spacing az between layers. Each of the four layers of the unit cell is labeled by two indices η = ±1 and τ = ±1. The index η = 1 (η = ¯1) corresponds to an electron (hole) dispersion relation. The index τ refers to two different values of the SOI, α1 and α¯1, where without loss of generality we assume that 0 < α1 < α¯1. The ordering of the layers inside the unit cell is shown in Fig. 2a. Two electron layers are followed by two hole layers as the SOI magnitude ατ alternates from layer to layer. (cid:82) dxdy Hn(x, y), where N is the total number (cid:80)N Hn = Hn0 + Hnt with Hn0 =(cid:80){τ,η=1,¯1} Hnητ , where of unit cells and the Hamiltonian density is given by The total Hamiltonian of the system is H = n=1 (cid:88) (cid:104) − η2 Hnητ = Ψ† nητ σ 2m0 σ,σ(cid:48) − iτ ατ (σx∂y − σy∂x) (cid:105) (∂2 x + ∂2 y) + ητ µτ σσ(cid:48)Ψnητ σ(cid:48). (4) The electron (hole) annihilation operator Ψnητ σ(r) acts on particles with spin σ at the position r = (x, y) of the (nητ )-layer. The chemical potential µτ is calculated from the crossing point at k = 0 determined by the SOI energy Eso,τ = 2k2 so,τ /2m0 with the SOI wavevector kso,τ = m0ατ /2. The dispersion relation (for fixed θ) of each layer is shown in Fig. 2a and can be easily generalized to all directions of k. In the following, we fix the chemical potentials as µ1 = µ¯1 = Eso,¯1−Eso,1. This choice ensures that the interior (exterior) Fermi surfaces have the same radius kF i = kso,¯1 − kso,1 (kF e = kso,¯1 + kso,1) across all the layers. Additionally, we need to assume that µτ (cid:29) t. FIG. 2. Panel a) shows the dispersion relation of each layer for fixed θ. The chemical potentials µτ are chosen such that inner and outer Fermi surfaces have the same radii across different layers. The arrows indicate where the tunneling be- tween the layers opens up gaps (small green circles). Note that the bottom and top layers stay gapless and have a dis- persion consisting of a single Dirac cone with spin locked to momentum due to time reversal invariance. Panel b) shows the interior and exterior Fermi surface of each layer with the cuts for ky = const. The fields for interior (exterior) left and right movers Si n[π−θe]ητ ) have in general different spin orientations. n[π−θi]ητ (Se nθeητ , Se nθiητ , Si The tunneling between the layers is assumed to be spin-independent and takes the following form, Hnt = tΨ† nητ σ(r)Ψn(cid:48)η(cid:48)τ(cid:48)σ(r) + H.c., (5) (cid:88) σ(cid:104)τ η;n(cid:48)τ(cid:48)η(cid:48)(cid:105) where the summation runs over all neighbouring layers. First, we demonstrate that the top and bottom layers host gapless modes with a helical Dirac spectrum. For the moment, we assume that the system is infinite and translationally invariant in x- and y- directions and we a)b)introduce momenta kx and ky, which are good quantum numbers. Alternatively, due to rotation invariance, one can change to polar coordinates with momenta kr and kθ. This allows us to treat the problem as effectively one- dimensional if the orbital degree of freedom is integrated out, see Fig. 2. The wavefunction can be represented close to the Fermi surface in terms of slowly varying fields Se/i nθητ , Ψnθητ σ(x, y) = αδθητ σSδ nθητ eikF δ(x cos θ+y sin θ) (6) (cid:88) δ=e,i with the angle θ ∈ [0, 2π), αeθητ σ = (cid:104)σ − τ · η, θ(cid:105) and αiθητ σ = (cid:104)ση, θ(cid:105). The kinetic term can be rewritten as βδηυF τ (Sδ nθητ )† ∂ ∂r Sδ nθητ , (7) H0 = − i(cid:88) δ=i,e nθ;τ,η=±1 where βe = 1 and βi = τ . We also take into account that the Fermi velocities υF τ = ∂Eτ /∂kµτ are differ- ent. The tunneling terms induce couplings between inte- rior/exterior Fermi surfaces of different layers, [n−1]θ¯11)†Se nθ¯1¯1)†Si nθ11)†Se [(Se nθ1¯1)†Si nθ¯11] + H.c. nθ¯1¯1 + (Se nθ11 + (Si (cid:88) nθ + (Si Ht = t nθ1¯1 (8) Here, we keep only non-oscillating terms and take into account the spin conservation during the tunneling, see Fig. 2a. Importantly, all coupling terms in Eq. (8) in- volve fields with opposite signs of Fermi velocities and each field, exept for the ones belonging to the top and bottom layers, has a partner to which it is coupled. This results in the opening of gaps at the Fermi level such that the bulk spectrum is fully gapped. However, the exterior Fermi surface field Se 1θ1¯1 of the top-most layer and the interior Fermi surface field Si N θ¯11 of the bottom- most layer do not have partners in Eq. (8) and, thus, stay gapless as all the remaining layers are insulating. As was noted above, Se N θ¯11 describe the helical Dirac cones in which spin direction is locked to the momentum direction. In our case, the spin direction stays always perpendicular to the momentum, see Fig. 2b. Such sur- face states are the hallmark of a strong 3D TI [2]. 1θ1¯1 and Si Since the rotational symmetry is broken, it is far from obvious that the surface states exist on any 2D bound- ary. To this end, we demonstrate that helical surface states also exist if a hard-wall boundary is added, say, at the plane x = 0. To this end we assume that the system is infinite in y- and z-direction. Since the sys- tem is translation invariant in y-direction (z-direction), ky (kz) is a good quantum number defined via Ψkz = ΣneinkzaΨn/ N , where a = 4az is the unit-cell size. The y-dependence of the total wavefunction is given trivially as Ψkykzητ σ(x, y) = eikyyΨkykzητ σ(x). Since both ky and kz are good quantum numbers the problem is effectively √ 3 (cid:88) one-dimensional, see Fig. 2b. To simplify the problem further, we linearize the motion in the x-direction which is achieved with the ansatz following from Eq. (6), Ψkykzητ σ(x) = αδθητ σSδ kzθητ (x)eikF δx cos θ, (9) δ=i,e θ∈{θδ,π−θδ} where Sδ kzθητ is the Fourier transform of Sδ nθητ . The above ansatz [26 -- 28] is valid for ky < kF i and t (cid:28) Eso,τ − 2(ky − τ kso,¯τ )2/2m with Ey = 2k2 y/2m0. The angles θi and θe are defined in Fig. 2b or explicitly ex- pressed by cos θδ = y/kF δ. The spin orienta- tion is determined by αδθητ σ and depends on θδ which in turn depends on ky, see Fig. 2b. F δ − k2 k2 (cid:113) After performing above linearization [26 -- 28], we arrive at the effective Hamiltonian ¯H0 = −i βδηυF τ cos θ(Sδ η,τ =±1 θ∈{θδ,π−θδ} δ=i,e (cid:88)  (cid:88) (cid:88) (cid:2)(Si (cid:88) (cid:2)(Se kzθ11)†Se θ∈{θi,π−θi} + t θ∈{θe,π−θe} ¯Ht = t kzθ1¯1)†Si kzθi11 + (Si kzθητ )†∂xSδ (cid:3) kzθi¯11 kzθ¯1¯1)†Si kzθe¯11)†Se kzθητ , (10) (11) (cid:3) + H.c. kzθ¯1¯1 + eikza(Se kzθe1¯1 It is readily noticeable from Fig. 2a, that the Hamilto- nian breaks down into 2 × 2 blocks, formed by the fields coupled by the tunneling. After inserting the ansatz kzθητ (x) ∼ eikδx, we arrive at the bulk spectrum around Sδ the interior and exterior Fermi surfaces, Eδ,± =kδ(υ1 − υ¯1) cos θδ ±(cid:113) 4t2 + k2 δ (υ1 + υ¯1)2 cos2 θδ, (12) where kδ = kx−kF δ cos θδ and δ = e, i. The bulk spectral √ υ1υ¯1/(υ1 + υ¯1). The dispersion gap is given by ∆ = 2t relation is determined by sin (2Ω) = ± 2 sin(kza/2) cos θe cos θi cos θe + cos θi , (13) and plotted in the SM. We note that E(ky, kz = 0) is independent of ky, which results in degeneracy. This de- generacy is due to fact that we only retained resonant processes in our perturbation analysis [29]. If the prob- lem is solved numerically (see below), this accidental de- generacy is lifted except at k = 0, where it is protected by time reversal symmetry. Also any perturbation in the chemical potentials lifts such a degeneracy and one is left with an single anisotropic Dirac cone. To demonstrate this explicitly, we assume a detuning δµ of chemical po- tential in the first layer. For each value of ky there is a twofold degeneracy which is lifted by such a perturbation. After performing the perturbation expansion within the (cid:19) ky kF i (cid:18) (cid:3) twofold degenerate subspace we arrive at the following dispersion relation E(ky, kz = 0) = δµ 8 1 − kF i kF e , (14) where we assumed υ1 = υ¯1 = υ, ky (cid:28) kF i, and t (cid:28) kF iυ, see the SM [30] for details. We finally address the above model numerically and study the edge states along the yz layer in the tight- binding model framework with ky and kz being good quantum numbers. The corresponding tight-binding kykzητ H0kykzητ + Hamiltonian is given by H = (cid:80) (cid:80) H0kykzητ = −(cid:88) (cid:2)η(t0 cos(kyay) + µτ /2)c Htkykz with kykz † kykzητ nσckykzητ nσ nσ † − ηt0c kykzητ (n+1)σckykzητ nσ + ¯ατ [c (cid:88) n † † kykzητ (n−1)¯1ckykzητ n1 − c kykzητ (n+1)¯1ckykzητ n1 (cid:88) † kykzητ n¯1ckykzητ n1] + H.c., † kykzη(cid:48)τ(cid:48)nσckykzητ nσ, (15) eiφτ ητ(cid:48) η(cid:48) c + 2i sin(kyay)c Htkykz = t nσ(cid:104)τ η;τ(cid:48)η(cid:48)(cid:105) where again the last sum runs over neighboring layers and ¯ατ is the spin-flip hopping amplitude, related to the phys- ical SOI parameter by ¯ατ = ατ /2ay (assuming ax = ay) and to the SOI energy by ¯Eso,τ = ¯α2 τ /t0 [31 -- 33]. Here, φ¯111¯1 = −φ1¯1¯11 = kzaz, otherwise, φτ ητ(cid:48)η(cid:48) = 0. The lat- tice constant in the i direction is ai with i = x, y, z. The † operator c kykzητ nσ is an annihilation operator acting on electron with momentum ky (kz) in the y (z) direction and with spin σ located at the point x = nax along the x direction of the ητ layer. Our numerical results confirm the strong TI phase, see Fig. 3 and Supplemental Mate- rial (SM) [30]. We again observe the single anisotropic Dirac cone, where the accidental degeneracy at kz = 0 described before is lifted by a slight detuning of the chem- ical potential or due to higher order tunneling terms not taken into account in the linearized approximation [30]. Conclusions. We introduced a coupled-layer approach to construct a strong 3D TI, where the building blocks are non-topological Rashba 2DEG layers. We showed that the bulk spectrum becomes gapped, with the gap being proportional to the tunnel coupling t between the layers -- a parameter that can be experimentally tuned. Additionally, any 2D boundary hosts gapless helical sur- face states. We calculated the dispersion relation of these surface states and found a single Dirac cone at k = 0, which together with the bulk gap constitutes a hallmark of a strong 3D TI [2]. We acknowledge support from the Swiss NF and 4 FIG. 3. a) Dispersion relation of the surface states (red) lo- calized in the yz-plane as well as bulk states (blue) obtained numerically, see Eq. (15). At small momenta, the surface states form a single anisotropic Dirac cone, but merge with the bulk states at large momenta. b) The spin orientation (red arrows) of the first layer of the unit-cell for a fixed en- ergy E/t0 = −0.03 and at the position x0 = 3ax away from the left edge. The parameter values assumed are: ¯α1 = 0.3t0, ¯α¯1 = 0.55t0, µ1 = µ¯1 = 0.2125t0, and t = 0.1t0. The spin orientation is locked to the momentum direction confirming the strong TI phase. [1] K. v. Klitzing, G. Dorda, and M. Pepper, Phys. Rev. Lett. 45, 494 (1980). [2] M. Z. Hasan and C. L. Kane, Rev. Mod. Phys. 82, 3045 (2010). [3] A. P. Schnyder, S. Ryu, A. Furusaki, and A. W. W. Ludwig, Phys. Rev. B 78, 195125 (2008). [4] A. Kitaev, Annals of Physics 303, 2 (2003). [5] B. Volkov and O. Pankratov, JETP Lett. 42, 145 (1985). [6] K. Nomura, M. Koshino, and S. Ryu, Phys. Rev. Lett. 99, 146806 (2007). [7] D. Poilblanc, G. Montambaux, M. H´eritier, and P. Led- erer, Phys. Rev. Lett. 58, 270 (1987). [8] L. P. Gor'kov and A. G. Lebed, Phys. Rev. B 51, 3285 (1995). [9] C. L. Kane, R. Mukhopadhyay, and T. C. Lubensky, Phys. Rev. Lett. 88, 036401 (2002). [10] J. Klinovaja and D. Loss, Phys. Rev. Lett. 111, 196401 (2013). [11] J. C. Y. Teo and C. L. Kane, Phys. Rev. B 89, 085101 (2014). NCCR QSIT. [12] J. Klinovaja and D. Loss, Eur. Phys. J. B 87, 171 (2014). a)b)5 [13] T. Meng, P. Stano, J. Klinovaja, and D. Loss, Eur. Phys. (1984). J. B 87, 203 (2014). [14] J. Klinovaja and Y. Tserkovnyak, Phys. Rev. B 90, 115426 (2014). [15] T. Neupert, C. Chamon, C. Mudry, and R. Thomale, Phys. Rev. B 90, 205101 (2014). [16] E. Sagi and Y. Oreg, Phys. Rev. B 90, 201102 (2014). [17] J. Klinovaja, Y. Tserkovnyak, and D. Loss, Phys. Rev. [25] Our results still hold if Rashba is replaced by Dresselhaus SOI. In case when both Rashba and Dresselhaus SOI are present, our scheme still works if one of them dominates. and [26] B. Braunecker, G. I. Japaridze, J. Klinovaja, D. Loss, Phys. Rev. B 82, 045127 (2010). [27] J. Klinovaja and D. Loss, Phys. Rev. B 86, 085408 (2012). B 91, 085426 (2015). [28] J. Klinovaja and D. Loss, Phys. Rev. Lett. 110, 126402 [18] R. A. Santos, C.-W. Huang, Y. Gefen, and D. B. Gut- (2013). man, Phys. Rev. B 91, 205141 (2015). [29] J. Klinovaja, P. Stano, and D. Loss, Phys. Rev. Lett. [19] E. Sagi, Y. Oreg, A. Stern, and B. I. Halperin, Phys. 109, 236801 (2012). Rev. B 91, 245144 (2015). [20] E. M. Stoudenmire, D. J. Clarke, R. S. K. Mong, and J. Alicea, Phys. Rev. B 91, 235112 (2015). [30] See Supplemental Material at [URL will be inserted by publisher] for the details about the analytical and numer- ical calculation. [21] S. Sahoo, Z. Zhang, and J. C. Y. Teo, arXiv:1509.07133 [31] D. Chevallier, D. Sticlet, P. Simon, and C. Bena, Phys. (2015). Rev. B 85, 235307 (2012). [22] T. Meng, arXiv:1506.01364 (2015). [23] E. Sagi and Y. Oreg, arXiv:1506.02033 (2015). [24] Y. A. Bychkov and E. I. Rashba, JETP Lett. 39, 78 [32] D. Rainis, L. Trifunovic, J. Klinovaja, and D. Loss, Phys. Rev. B 87, 024515 (2013). [33] J. Klinovaja and D. Loss, Eur. Phys. J. B 88, 62 (2015). Supplemental Material to 'From Coupled Rashba Electron and Hole Gas Layers to 3D Topological Insulators' 1Department of Physics, University of Basel, Klingelbergstrasse 82, CH-4056 Basel, Switzerland Luka Trifunovic,1,2 Jelena Klinovaja,1 and Daniel Loss1 2Dahlem Center for Complex Quantum Systems and Physics Department, Freie Universitat Berlin, Arnimallee 14, 14195 Berlin, Germany Details of the analytical calculation In order to obtain the spectrum of the surface states we fix the parameters (including the energy inside the gap) and find the eight decaying eigenstates of the Hamilto- nian. Using Eq. (9), we express them in the basis of the original fermionic fields Ψ(x), leading to eight eight- spinor solutions Φj(x) with j = 1, . . . , 8, and construct a 8 × 8 Wronskian matrix Wij(x) = [Φj(x)]i. The equa- tion det W (0) = 0 gives the spectrum of the surface states [29]. We note that for ky (cid:54)= 0, the interior and exterior branches have different velocities in x-direction. After substituting E = ∆ cos Ω with Ω ∈ [0, π] and as- suming θe, θi ∈ [0, π/2) we obtain (cid:2)4 sin2(kza/2) cos2 θe cos2 θi (16) −(cos θe + cos θi)2 sin2(2Ω)(cid:3) . det W (0) = υ2eikza/2 υ1 Thus, the dispersion from Eq. 13 in the main text, shown in Fig. 4, is obtained. chemical potential in the first layer. For each value of ky there is twofold degeneracy which is lifted by such a per- turbation. After performing the perturbation expansion in lowest order within the twofold degenerate subspace we arrive at the following dispersion relation (cid:18) kF e (cid:19) ky t , kF i kF iυ kF i , (17) (cid:115) E(ky, kz = 0) = δµf where we assumed υ1 = υ¯1 = υ and ky (cid:28) kF i. The function f (x, y) is given by (x > 1) (x2 − 1)2 f (x, y) = 8x((x2 − 1)2 + 4y(x2 + 1)) ×(cid:112)(2y(x − 1)2 − 4y3)2 + (x + 6y2)2. 1 + 4y2 (x − 1)2 (18) Since our analysis is valid for t (cid:28) kF iυ [we took only resonant terms into account in Eq. (5)] we can further simplify the above dispersion by expanding for small t/(kF iυ) and arrive at Eq. (14) of the main text. Detuning of the chemical potential In this section we show that any perturbation of the chemical potential in one of the layers lifts the degener- acy for kz = 0 which is depicted in Fig. 4. In order to demonstrate this explicitly we assume a detuning δµ of In this section we compare our numerical to analyti- cal results and additionally give more details about the numerical results. Our analytical results are valid for Numerical calculation of 2D surface states spectrum 6 FIG. 4. Dispersion relation of the surface states localized in the yz-plane for kF e/kF i = 3, obtained analytically from Eq. (13) of the main text with E/∆ = cos Ω. We plot ky up to value of 0.9kF i, the concreate range of validity of the dispersion depends on the value of t and is give below the Eq. (9) of the main text. FIG. 5. The dispersion relation of the surface states local- ized in the yz-plane obtained numerically in the tight-binding model. The parameters of the system take the following val- ues: ¯α1 = 0.3t0, ¯α¯1 = 0.55t0, µ1 = µ¯1 = 0.2125t0, and t = 0.05t0. FIG. 6. Panel a) [b)] show the kz = 0 [ky = 0] cut of the dispersion relation of the surface states localized in the yz- plane obtained numerically in the tight-binding model. The parameters of the system take the following values: ¯α1 = 0.3t0, ¯α¯1 = 0.55t0, µ1 = µ¯1 = 0.2125t0, and t = 0.1t0. t (cid:28) µτ, Eso,τ , and in this limit we obtain the degen- eracy for kz = 0, see Fig. 4. This degeneracy is lifted linearly in ky for ky (cid:28) kF i as shown in the main text. Our numerical tight-binding simulation confirms all these features, see Fig. 5. Namely, around ky = 0 the degen- eracy is linearly lifted since for the tight-binding model it is very difficult to tune the sizes of the Fermi sur- faces to be the same across the layers. Additionally, we find that there is a remaining degeneracy at ky ∼ kF i (kF i ∼ 0.2π/ay for the parameters in Fig. 5). We found that increasing the tunnel coupling between the layers, above the limit where the linearization works t ∼ Eso,1, the ky = 0 degeneracy gets completely lifted and one obtains a single Dirac cone at ky = 0, see Fig. 3 of the main text. Additionally, in Fig. 6a [Fig. 6b] we plot the cuts kz = 0 [ky = 0] of the dispersion relation which show that there is no additional structure inside of the Dirac cone. The Fig. 6a shows the behaviour of the surface states within the whole Brillouin zone from where it is seen that the dispersion relation curve of the surface state does not bend down. a)b)

1102.5258

1

1102

The catalytic potential of high-k dielectrics for graphene formation

The growth of single and multilayer graphene nano-flakes on MgO and ZrO2 at low temperatures is shown through transmission electron microscopy. The graphene nano-flakes are ubiquitously anchored at step edges on MgO (100) surfaces. Density functional theory investigations on MgO (100) indicate C2H2 decomposition and carbon adsorption at step-edges. Hence, both the experimental and theoretical data highlight the importance of step sites for graphene growth on MgO.

cond-mat.mes-hall

cond-mat

The catalytic potential of high-k dielectrics for graphene formation Andrew Scott,1, 2 Arezoo Dianat,2 Felix Borrnert,1, a) Alicja Bachmatiuk,1 Shasha Zhang (张莎莎),1 Jamie H. Warner,3 Ewa Borowiak-Pale´n,4 Martin Knupfer,1 Bernd Buchner,1 Gianaurelio Cuniberti,2,5 and Mark H. Rummeli1, 2, b) 1)Leibniz-Institut fur Festkorper- und Werkstoffforschung Dresden e. V., PF 27 01 16, 01171 Dresden, Germany 2)Technische Universitat Dresden, 01062 Dresden, Germany 3)University of Oxford, Parks Road, Oxford OX1 3PH, United Kingdom 4)Zachodniopomorski Uniwersytet Technologiczny, Pulaskiego 10, 70322 Szczecin, Poland 5)Division of IT Convergence Engineering, POSTECH, Pohang 790-784, Republic of Korea (Dated: 6 November 2018) The growth of single and multilayer graphene nano-flakes on MgO and ZrO2 at low temperatures is shown through transmission electron microscopy. The graphene nano-flakes are ubiquitously anchored at step edges on MgO (100) surfaces. Density functional theory investigations on MgO (100) indicate C2H2 decomposition and carbon adsorption at step-edges. Hence, both the experimental and theoretical data highlight the importance of step sites for graphene growth on MgO. PACS numbers: 81.05.ue, 81.16.Hc Graphene promises a great leap forward in electronic de- vice speed due to its high charge carrier mobility.1 Restricting its dimensions sufficiently (< 10 nm for room temperature ap- plications) to introduce finite size effects can open an energy gap.2,3 Confining the width and keeping the length infinitely long forms a graphene nano-ribbon (GNR) whilst restricting both the width and length forms a graphene nano-flake. Inter- esting electronic properties can also arise from bi-layer nano- ribbons/flakes and even higher order stacked flakes. Their energy gaps are not just dependent on size, but also on the edge states and electric fields.4 -- 8 Hence, the richness of elec- tronic properties afforded by graphene causes great excite- ment for its use in nano-electronic devices. However, various technical hurdles exist. For example, the implementation of graphene in nano-electronic technology is hindered by diffi- culties in obtaining defect free and clean graphene on suitable substrates. The substrates need to be insulating, ideally have a high dielectric constant, k , to prevent gate current leakage and hence allow greater miniaturization. In graphene, the carrier concentration and polarity can be modulated by an electric field. However, the fabrication of graphene devices with an atomically uniform gate dielectric, so as to provide a uniform electric field over the active graphene area without damag- ing the graphene surface remains a challenge.9 For this rea- son and the ease with which graphene can be observed, most reported graphene devices are fabricated on Si/SiO2 which forms a global back gate. Attempts to deposit dielectrics on the surface of graphene to form top gates have been made, however the deposition processes employed tend to damage the graphene, see e. g. Ref. 10. The situation for graphene based devices is further complicated due to restrictions im- posed by most of the available synthesis routes. Aside from mechanical cleaving, graphene synthesis generally involves the use of metal catalysts11 -- 13 which then necessitates post- synthesis transfer of the graphene or in the case of epitaxi- a)Electronic mail: f.boerrnert@ifw-dresden.de b)Electronic mail: m.ruemmeli@ifw-dresden.de ally grown graphene in SiC high temperatures are required.14 Other synthesis routes like chemical exfoliation or the neces- sary transfer steps from the aforementioned routes to dielec- tric substrates introduce high levels of contamination and de- fects into the graphene making such synthesis routes incom- patible with current planar fabrication technology where re- producibility is paramount. Here we report graphene forma- tion over a previously introduced class of catalytically active substrates, metal oxides.15,16 We show through transmission electron microscopy (TEM) that graphene nano-flakes can form on MgO and multilayer graphene nano-flakes on both MgO and ZrO2. As high-k dielectrics, these oxides are an at- tractive natural candidate for graphene based electronics. Fur- thermore the MgO system can catalyze graphene growth via thermal chemical vapor deposition (CVD) at the low temper- ature of 325 °C.16 Growth on ZrO2 can occur at least as low as 480 °C. All reactions were conducted in a CVD oven comprising a quartz tube and a sliding heating element. The samples were placed in an alumina boat. In the case where cyclohexane served as the feedstock, the oven loaded with MgO nano- powder was evacuated to < 1 mbar and heated to a synthe- sis temperature of 775 °C. A pressure of 100 mbar was used. The reactions were terminated after 17 s to 300 s by flushing with argon and sliding the heating element off the reaction zone. Further experiments were performed on MgO and ZrO2 using acetylene as the feedstock. After loading the samples the reactor was evacuated and then exposed to a gentle ar- gon flow (1015 mbar), and heated to temperatures ranging be- tween 325 °C and 650 °C. The argon flow was held for 10 min and then switched to a mixture of acetylene and argon. Fi- nally the oven was cooled in an argon atmosphere. In some cases we removed the oxide catalyst by ultrasonication in HCl. The TEM studies were performed on an FEI Titan3 80 -- 300 with an 80 kV acceleration voltage. The ab initio total energy calculations were carried out using density-functional theory (DFT). They were performed within the PBE generalized gra- dient approximation for the exchange-correlation functional17 and the projector augmented-wave method,18 using the Vi- 2 FIG. 2. Graphitic carbon grown over monoclinic ZrO2 from acety- lene at 480 °C and 1015 mbar. One clearly sees the anchoring of the graphitic layers onto a step of the (020) crystal face. Arrows indicate graphene sheets anchored at step sites. bon. Furthermore, as indicated by the arrow, the multi-layers are ubiquitously anchored into step edges on the (100) sur- faces. We propose that these step sites initiate the growth of graphitic layers. We argue that these step sites are not only nu- cleation sites, but also growth sites because growth appears to stop once the particle is fully encapsulated. In this regard, this work corroborates previous studies suggesting the cooperative role of oxide supports in the growth of multi walled carbon nanotubes.22 We also investigated zirconium oxide to further elucidate the atomic surface structure required for graphitic synthesis on oxides. This catalyst has recently been exten- sively explored by Steiner III et al.23 for the synthesis of car- bon nanotubes. The investigators used in situ XPS to confirm that zirconium oxide in a monoclinic baddeleyite or slightly oxygen deficient form could form sp2 carbon. In our case, the synthesis of graphitic layers is performed on nano-crystallites which following the reaction are in the baddeleyite form ev- idenced by the lattice fringes in TEM micrographs of figure 2. In both cases arrows indicate graphitic shells anchoring on step sites. In Fig. 2(b) it is possible to determine that the graphitic layers are anchored on a face with (020) as its sur- face normal. The low synthesis temperatures employed sug- gest the oxides are catalytically active for the formation of sp2 carbon. To support this concept, theoretical studies were con- ducted on MgO step sites. Based on experimental evidence, we chose a step site on the (100) surface of MgO as a model system for oxide based graphene catalysis. The exact atomic structure of graphene on MgO is undetermined. The lattice constant of MgO is 2.10 A and the spatial frequency in the re- ciprocal lattice of graphene is 2.13 A−1. The adsorption and dissociation behavior of C2H2 on (100) and step-edge of MgO surface was studied. Several adsorption sites on the (100) sur- face and step-edge of MgO are considered as initial adsorption states which are shown in Fig. 3(a). The adsorption energy is defined as the difference between the total energy of the sys- tem with the adsorbate and without the adsorbate referred to as the free energy of C2H2. The adsorption energies are shown in Fig. 3(a). In all adsorption sites, the molecule binds very weakly to the (100) surface. The distance from the surface and the center of mass of C2H2 lies between 2.5 A and 3 A. These distances are in the physisorption adsorption regime. The dissociative adsorption energy of C2H2 to C2H and H is FIG. 1. (color online) (a) -- (c) Multilayer graphene synthesized over MgO from cyclohexane at 775 °C and 100 mbar. (a) Few layer graphene shells after removal of MgO core via acid treatment. (b) Graphene lattice from single layer shell, (c) graphene layers an- chored at step-edges on the (100) surface. (d) Graphene flakes syn- thesized over MgO from acetylene at 325 °C and 1015 mbar. The attachment of graphene layers to the (100) oxide step-edges can be observed. Arrows indicate graphene sheets anchored at step sites. enna ab initio simulation package (VASP).19 The activation energies were calculated using the nudged elastic band model as developed by the Henkelman group.20 The simulation cell included a slab of four substrate layers with a lateral size of a 4 × 4 surface unit cell. This corresponds to 16 atoms on the surface layer. The step surface is prepared by removing eight atoms from the surface layer. In the case of an existing graphene on step edge surface, the lateral size of simulation cell is duplicated. In all simulations the vacuum gap between the slabs was larger than 15 A. The wave functions have been expanded into plane waves up to a kinetic energy cut-off of 400 eV. The integration in the first Brillouin zone was per- formed using Monkhorst-Pack grids21 including 25 k-points in the irreducible wedge. In all calculations, the two topmost layers of MgO have been optimized and from third layer up fixed in the bulk position until all force components were less than 0.01 eV/ A. The convergence of energy differences with respect to the used cut-off energies and k-point grids has been tested in all cases within a tolerance of 10 meV/atom. In figure 1, a variety of micrographs are presented which show we are able to grow graphene flakes on the surface of magnesium oxide and tune the growth from single to multi layers. The samples were grown from cyclohexane at 775 °C and 100 mbar. With these conditions, the reaction stops when the catalyst is fully encapsulated by graphitic layers. We found that the number of layers can increase up to nine and can be tuned via the reaction time. In figure 1 (c) we highlight the closely spaced parallel lattice fringes of the crystals are those of (100) planes on top of which there is graphitic car- 3 investigations, while still preliminary, clearly show each of these subprocesses on MgO (100) step sites (cf. Fig. 3). The experimental data show graphene nano-flakes anchored at step sites. Questions remain though, for example, what are the ef- fects of the edges of the nano-flakes, their possible interaction with the substrate and the effect of defects in the dielectric. None-the-less, the demonstrated catalytic potential of high-k dielectrics for the catalytic formation of graphene via thermal CVD at low temperatures is an advance. F.B. acknowledges the DFG (RU 1540/8-1), A.B. the A.- v.-Humboldt Stiftung and the BMBF, S.Z. the IMPRS "Dy- namical Processes in Atoms, Molecules and Solids", G.C. the South Korean Ministry of Education, Science, and Technol- ogy Program, Project WCU ITCE No. R31-2008-000-10100- 0., and M.H.R. the EU (ECEMP) and the Freistaat Sachsen. 1A. K. Geim, Science 324, 1530 (2009). 2Y.-W. Son, M. L. Cohen, and S. G. Louie, Phys. Rev. Lett. 97, 216803-1 (2006). 3P. Shemella, Y. Zhang, M. Mailman, P. M. Ajayan, and S. K. Nayak, Appl. Phys. Lett. 91, 042101-1 (2007). 4B. Sahu, H. Min, A. H. MacDonald, and S. K. Banerjee, Phys. Rev. B 78, 045404-1 (2008). 5M. P. Lima, A. Fazzio, and A. J. R. da Silva, Phys. Rev. B 79, 153401-1 (2009). 6B. Sahu, H. Min, and S. K. Banerjee, Phys. Rev. B 81, 045414-1 (2010). 7B. Sahu, H. Min, and S. K. Banerjee, Phys. Rev. B 82, 115426-1 (2010). 8F. Xia, D. B. Farmer, Y.-M. Lin, and P. Avouris, Nano Lett. 10, 715 (2010). 9B. Lee, S.-Y. Park, H.-C. Kim, K. Cho, E. M. Vogel, M. J. Kim, R. M. Wallace, and J. Kim, Appl. Phys. Lett. 92, 203102-1 (2008). 10M. C. Lemme, T. J. Echtermeyer, M. Baus, and H. Kurz, IEEE Electr. Dev. Lett. 28, 282 (2007). 11P. W. Sutter, J.-I. Flege, and E. A. Sutter, Nature Mater. 7, 406 (2008). 12A. Reina, X. Jia, J. Ho, D. Nezich, H. Son, V. Bulovic, M. S. Dresselhaus, and J. Kong, Nano Lett. 9, 30 (2009). 13X. Li, W. Cai, J. An, S. Kim, J. Nah, D. Yang, R. Piner, A. Velamakanni, I. Jung, E. Tutuc, S. K. Banerjee, L. Colombo, and R. S. Ruoff, Science 342, 1312 (2009). 14K. V. Emtsev, A. Bostwick, K. Horn, J. Jobst, G. L. Kellogg, L. Ley, J. L. McChesney, T. Ohta, S. A. Reshanov, J. Rohrl, E. Rotenberg, A. K. Schmid, D. Waldmann, H. B. Weber, and T. Seyller, Nature Mater. 8, 203 (2009). 15M. H. Rummeli, F. Schaffel, A. Bachmatiuk, D. Adebimpe, G. Trotter, F. Borrnert, A. Scott, E. Coric, M. Sparing, B. Rellinghaus, P. G. Mc- Cormick, G. Cuniberti, M. Knupfer, L. Schultz, and B. Buchner, ACS Nano 4, 1146 (2010). 16M. H. Rummeli, A. Bachmatiuk, A. Scott, F. Borrnert, J. H. Warner, V. Hoffmann, J. H. Lin, G. Cuniberti, and B. Buchner, ACS Nano 4, 4206 (2010). 17J. P. Perdew, K. Burke, and M. Ernzerhof, Phys. Rev. Lett. 77, 3865 (1996). 18P. E. Blochl, Phys. Rev. B 50, 17953 (1994). 19G. Kresse and J. Furthmuller, Phys. Rev. B 54, 11169 (1996). 20G. Henkelman, A. Arnaldsson, and H. J´onsson, Comput. Mater. Sci. 36, 354 (2006). 21H. J. Monkhorst and J. D. Pack, Phys. Rev. B 13, 5188 (1976). 22M. H. Rummeli, F. Schaffel, C. Kramberger, T. Gemming, A. Bachmatiuk, R. J. Kalenczuk, B. Rellinghaus, B. Buchner, and T. Pichler, J. Am. Chem. Soc. 129, 15772 (2007). 23S. A. Steiner III, T. F. Baumann, B. C. Bayer, R. Blume, M. A. Worsley, W. J. MoberlyChan, E. L. Shaw, R. Schlogl, A. J. Hart, S. Hofmann, and B. L. Wardle, J. Am. Chem. Soc. 131, 12144 (2009). 24S. Helveg, C. L´opez-Cartes, J. Sehested, P. L. Hansen, B. S. Clausen, J. R. Rostrup-Nielsen, F. Abild-Pedersen, and J. K. Nørskov, Nature (London) 427, 426 (2004). 25S. Hofmann, G. Cs´anyi, A. C. Ferrari, M. C. Payne, and J. Robertson, Phys. Rev. Lett. 95, 036101-1 (2005). (color online) (a) The initial adsorption sites of C2H2 on FIG. 3. (100) and step-edge of MgO surface and corresponding calculated adsorption energies for the different sites and dissociative energies of C2H2 to C2H and H. (b) Different configurations of carbon adsorp- tion on a MgO surface with an existing graphene flake. Red -- O, yellow -- Mg, blue -- C, green -- Mg in step plane, and white -- H. defined by Ediss. = EC2H+Hs − Es − EC2H2 , where EC2H+Hs is the total energy of the C2H and the H on the substrate, Es the total energy of the clean substrate, and EC2H2 the free energy of acetylene in the gas phase. The data shows that the dissociative adsorption of C2H2 on MgO (100) is an endothermic reaction (see Fig. 3). In contrast to the (100) sur- face, stronger binding is found between C2H2 and the step- edge of MgO. The dissociative adsorption energies indicate exothermic reactions on the step-edge surface. Thus, accord- ing to our calculations, the presence of the step-edge on the MgO surface is essential for the decomposition of C2H2. In addition, we have calculated the adsorption of single carbon atoms on several surface positions. For each adsorption posi- tion, the lateral position of carbon is fixed to avoid the diffu- sion of the atom to favorable adsorption sites. The most favor- able adsorption site of carbon on an MgO (100) surface is the step-edge adsorption position. In addition, the diffusion bar- rier of a carbon atom on MgO (100) has been calculated. The total energy of the initial and final configurations have been obtained using geometry optimization. The diffusion barrier is about 0.38 eV. The diffusion barrier of a hydrogen atom on the surface from on-top of one oxygen atom to the next is 0.2 eV. Thus, the hydrogen atoms simply diffuse away while the carbon remains on the surface. The importance of step sites to the growth of carbon nan- otubes is already well established for some metal catalysts. It has been argued elsewhere24,25 that the crystalline state of nickel catalyst particles evidences a surface growth mecha- nism which can divided into four distinct sub-processes -- the adsorbtion of carbon feedstock molecules on the catalyst, dis- sociation of hydrogen from the precursor, surface diffusion and addition of carbon atoms to the network. Our theoretical 

1511.06804

2

1511

First-principles study on interface magnetic structure in Nd${}_2$Fe${}_{14}$B/(Fe,Co) exchange spring magnets

The magnetic properties of Nd${}_2$Fe${}_{14}$B (NFB)/transition metal (TM = Fe, Co) multilayer systems are studied on the basis of first-principles density functional calculations. We optimize the model structure under a variety of crystallographic alignments of the NFB layer, and analyze the mechanism of interface magnetic coupling. Improvements in remanent magnetization compared to that of single NFB are observed in NFB(001)/Fe, NFB(110)/Fe, and NFB(100)/Co. On the other hand, in NFB(100)/Fe, remanence degradation due to the anti-parallel magnetization alignment between NFB and Fe layers is observed. In this system, which has the shortest optimized interlayer distance among all considered systems, an itinerant electron magnetism is required around the interface to lower the total energy, and accordingly, anti-ferromagnetic coupling is preferred. The significant difference in property between NFB(100)/Fe and NFB(100)/Co is attributed to the difference between their interface structures, optimized interlayer distances, and magnetic stiffness of TM layers.

cond-mat.mes-hall

cond-mat

First-principles study on interface magnetic structure in Nd2Fe14B/(Fe, Co) exchange spring magnets Department of Applied Physics, Tohoku University, Aoba 6-6-05, Aoba-ku, Sendai 980-8579, Japan Nobuyuki Umetsu and Akimasa Sakuma Yuta Toga ESICMM, National Institute for Materials Science, Sengen 1-2-1, Tsukuba 305-0047, Japan (Dated: August 6, 2018) The magnetic properties of Nd2Fe14B (NFB)/transition metal (TM = Fe, Co) multilayer systems are studied on the basis of first-principles density functional calculations. Assuming a collinear spin structure, we optimize the model structure under a variety of crystallographic alignments of the NFB layer, and analyze the mechanism of interface magnetic coupling. Improvements in remanent magnetization compared to that of single NFB are observed in NFB(001)/Fe, NFB(110)/Fe, and NFB(100)/Co. On the other hand, in NFB(100)/Fe, remanence degradation due to the anti-parallel magnetization alignment between NFB and Fe layers is observed. In this system, which has the shortest optimized interlayer distance among all considered systems, an itinerant electron magnetism is required around the interface to lower the total energy, and accordingly, anti-ferromagnetic cou- pling is preferred. The significant difference in property between NFB(100)/Fe and NFB(100)/Co is attributed to the difference between their interface structures, optimized interlayer distances, and magnetic stiffness of TM layers. I. INTRODUCTION Much effort has been made to improve the properties of permanent magnets, whose figure of merit is the max- imum energy product (BH)max that increases with co- ercivity and remanent magnetization. Exchange spring magnets1,2, nanocomposite materials consisting of hard and soft magnetic phases coupled by exchange interac- tion, have been promising as high-performance magnets3, and Nd -- Fe -- B-based exchange spring magnets are par- ticularly attractive from the viewpoint of low rare-earth metal content. In theoretical studies, (BH)max values of 0.6∼1.0 MJm−3 for Nd -- Fe -- B-based magnets have been predicted4 -- 7, but such high energy-product values have been difficult to achieve in real materials8 -- 16. To obtain greater values of (BH)max in real materials, the advanced design of nanostructures is required. Although the op- timal grain size and multilayer thickness of exchange spring magnets have been intensively studied17 -- 24, little is known about other crucial factors affecting the mag- netic properties of exchange spring magnets. Based on a first-principles study, Toga et al. pointed out that the magnetic properties of Nd2Fe14B (NFB)/bcc-Fe multilayer systems strongly depend on the crystallographic alignments of NFB and Fe layers25. It was shown that the NFB layer of the (001) plane is ferromagnetically coupled with Fe layers, but the (100) plane is anti-ferromagnetically coupled. Recently, these predictions were confirmed by Ogawa et al.26 by per- forming ferromagnetic resonance measurements. Their results imply that random crystallographic alignments possibly deteriorate magnetic performances because of low remanent magnetization. However, the reason for the drastically different results depending on crystallo- graphic alignments is not known at present. Moreover, it has been recognized that structure optimizations per- formed in a previous study25 are not sufficient, because those structures are only optimized with respect to the interlayer distance between the NFB and Fe layers, but not with respect to cell volume, cell shape, and ion sites. In order to identify and present reliable guidelines for fabricating high-performance magnets, the adequate op- timization of nanostructures is desirable. In the present work, we theoretically study the inter- face magnetic structure in exchange spring magnets. Our aim is to optimize the structure of the NFB/transition metal [TM = (bcc-)Fe, (hcp-)Co] multilayer systems for a variety of crystallographic alignments of NFB layers, and to analyze those interface electronic structures in order to understand crucial factors that determine the properties of high-performance magnets. In particular, we focus on NFB(100)/TM, which shows significantly different prop- erties between TM = Fe and TM = Co systems. In our calculations, we assume a collinear spin struc- ture between local moments at NFB layer and TM layer. Actually, there may be a possibility of non-collinear structure, but credible results for non-collinear structures are hard to obtain in such a case of finite sized super-cell model with periodic boundary condition. Therefore, a detailed spin structure is not predicted from our model, but the results suggest a necessity to reconsider the tacit postulation that the interface coupling in exchange spring magnets is always ferromagnetic. This paper is organized as follows. In Section II, we present our model and explain the method of structure optimization. In Section III, the results of our calculation are presented and discussed. Finally, the summary and conclusions of our work are described in Section IV. II. MODEL AND METHOD (a) NFB(001)/Fe(100) 2 (b) NFB(100)/Fe(110) (c) NFB(110)/Fe(112) We study the electronic structure of NFB/TM multi- layer systems, as shown in FIG. 1. The lattice constants of NFB are set to ah = bh = 8.8 A and ch = 12.19 A, and ion sites are determined as per a previous study27. The bcc-Fe and hcp-Co are assumed as structurally soft phases, and these plane indices facing the interface are chosen so as to match the surface size of an NFB unit cell. It is experimentally confirmed that the (001) plane and (100) plane of NFB layers match the (100) plane and (110) plane of bcc-Fe layers, respectively, but the other systems studied in this paper have not been well investi- gated experimentally. The deformed lattice constants of these soft phases are shown in TABLE I (the definitions of parameters are shown in FIG. 2). The thickness of the soft layer of our models corresponds to five atomic lay- ers, which is different from that of previously reported models25. TABLE I. Lattice constants of the soft magnetic phases of NFB/TM (TM = bcc-Fe, hcp-Co). The definitions of parameters are shown in FIG. 2. The experimental values for bcc-Fe are as = bs = cs = 2.87 A and α = β = γ = 90◦, and those of hcp-Co are as = bs = 2.51 A, cs = 4.07 A, and θ = 60◦. as[A] bs[A] cs[A] α[◦] β[◦] γ[◦] θ[◦] 2.93 2.93 2.87 90.0 90.0 90.0 2.87 2.87 2.93 90.0 90.0 90.1 2.87 2.87 2.88 90.0 90.0 90.2 - - - NFB(001)/Fe(100) NFB(100)/Fe(110) NFB(110)/Fe(112) NFB(001)/Co(0001) 2.51 2.44 4.07 NFB(100)/Co(0001) 2.51 2.51 4.07 NFB(110)/Co(0001) 2.41 2.44 4.07 - - - - - - - - - 61.0 60.2 59.5 The electronic structures are determined with first- principles density functional calculations using a plane- wave basis set. We use the Vienna Ab-initio Simulation Package (VASP)29. The ionic potentials are described by the projector-augmented-wave (PAW) method and the exchange-correlation energy of electrons is described within a generalized gradient approximation (GGA). We use the exchange-correlation functional determined by Ceperly and Alder and parameterized by Perdew and Zunger, with the interpolation formula according to Vosko et al.30. The energy cutoff is 318.6 eV in all sys- tems, and the Monkhorst-Pack k-point meshes of 1×3×3, 3 × 3 × 1, and 3 × 3 × 3 are selectively used depending on the model size. Our computations are performed until the total energy change converges to below 10−4 eV. The collinear spin structures are assumed, and magnetic sta- ble states are determined by comparing the total energy for parallel magnetization alignment (PMA) and anti- parallel magnetization alignment (APMA) between NFB and TM layers. The distance between these layers is first optimized by using the fixed lattice constants shown in TABLE I. By using force and stress, which are calculated using first principles, we perform additional optimization (d) NFB(001)/Co(0001) (e) NFB(100)/Co(0001) (f) NFB(110)/Co(0001) FIG. 1. Composite system models. Two unit cells of NFB are incorporated in the supercell of NFB(001)/Co(0001), while one unit cell is incorporated in the others. These figures are plotted using VESTA28. (a) bcc-Fe (b) hcp-Co FIG. 2. Definitions of lattice parameters (see TABLE I). with respect to the cell volume, cell shape, and ion sites. For optimization with respect to ion sites, we assume that only the ions facing the interface boundary are allowed to change positions. III. RESULTS AND DISCUSSION In FIG. 3, we show the interlayer distance dependence of the total electronic energy of NFB/TM for PMA and APMA. These are the results before optimization with respect to the cell volume, cell shape, and ion sites, while the magnetic properties (i.e., coupling constant and mag- netization) after optimization considering these factors are summarized in TABLE II, in addition to the results before optimization. The exchange coupling energy be- tween magnetization M1 and M2 is written as Eex = −J n1 · n2, (1) where J is coupling constant, and n1(2) = M1(2)/M1(2) is an unit vector. Since the difference of total energy for between APMA and PMA corresponds to the interface coupling energy, coupling constant is obtained by J = EAPMA − EPMA 2S , (2) where EAPMA and EPMA are the minimum total energy for APMA and PMA, respectively, and S is the inter- face area. For results after optimization, we divide the numerical difference by the interface area of the system with stable magnetization alignment (the results of the change rate of interface area between PMA and APMA are less than 1% in all systems). The magnetization shown in TABLE II is defined as the sum of local mag- netic moments for stable magnetization alignment per unit volume. In each system, the total electronic en- ergy after optimization is less than that before optimiza- tion by approximately 5 ∼ 10 eV. The values of mag- netization after optimization are close to those before optimization, while the values of coupling energy after magnetization are not so close to those before magne- tization (but the signs of coupling constants are con- sistent). The coupling constants of NFB(001)/Fe(100) and NFB(100)/Fe(110) before optimization are consis- tent with the previous results25. Among all our models, anti-parallel states are stable only in NFB(100)/Fe(110). The positive coupling constant of NFB(001)/Fe(100) and the negative coupling constant of NFB(100)/Fe(110) are in good agreements with the recent experimental results of Ogawa et al.26. Our results for NFB(110)/Fe(112) indicate positive coupling constant, whereas a negative coupling constant for NFB(110)/bcc-Fe has also been re- ported by Ogawa et al. However, it should be noted that the crystallographic alignment of the bcc-Fe layer has not yet been specified in that study. The comparison between the results of the TM = Fe system and those of the TM = Co system show that the optimized interlayer distance of the former models are shorter than those of the latter due to the strength difference of the nuclear interactions. The absolute value of coupling energy for TM = Fe is smaller than that for TM = Co, which may reflect the situation that the magnetic stiffness of Fe is lower than that of
a
NFB
001

Fe
100

b
NFB
001

Co
0001
3 à ae -2108.0 -2108.5 -2109.0 -2109.5 -2110.0 -2110.5 -2111.0 à ae APMA à PMA ae à ae 2.30 2.35 2.40 2.45 2.50 2.55 2.60 Interlayer Distance
Þ

d
NFB
100

Co
0001
-1227 -1228 à -1229 -1230 -1231 -1232 -1233 ae APMA à PMA ae à ae à ae 1.80 1.85 1.90 1.95 2.00 2.05 2.10 Interlayer Distance
Þ

f
NFB
110

Co
0001
-1674 -1676 à -1678 -1680 -1682 -1684 ae APMA à PMA ae à ae ae
V e
y g r e n E l a t o T
V e
y g r e n E l a t o T
V e
y g r e n E l a t o T -924.0 -924.5 à -925.0 -925.5 -926.0 -926.5 -927.0 APMA à à à ae ae PMA ae ae 2.10 2.15 2.20 2.25 2.30 2.35 2.40 Interlayer Distance
Þ

c
NFB
100

Fe
110
ae à à ae PMA ae APMA à ae à 1.70 1.75 1.80 1.85 1.90 1.95 2.00 Interlayer Distance
Þ

e
NFB
110

Fe
112
à ae APMA à ae PMA à ae à ae
V e
y g r e n E l a t o T
V e
y g r e n E l a t o T
V e
y g r e n E l a t o T -1267.0 -1267.5 -1268.0 -1268.5 -1269.0 -1231 -1232 -1233 -1234 -1235 -1236 1.90 1.95 2.00 2.05 2.10 2.15 2.20 2.10 2.15 2.20 2.25 2.30 2.35 2.40 Interlayer Distance
Þ
Interlayer Distance
Þ
FIG. 3. Total electronic energy of NFB/TM (results before optimization). The blue circles and red squares show the results for parallel magnetization alignment (PMA) and anti-parallel magnetization alignment (APMA), respectively. TABLE II. Magnetic properties before and after optimization. J and M are the results of coupling constant and magnetization after optimization with respect to the cell volume, cell shape, and ion sites, respectively. The subscripts 0 represent the results before optimization. NFB single phase NFB(001)/Fe(100) NFB(100)/Fe(110) NFB(110)/Fe(112) NFB(001)/Co(0001) NFB(100)/Co(0001) NFB(110)/Co(0001) - - J0 [J/m2] J [J/m2] M0 [T] M [T] 1.57 1.65 0.58 1.72 1.46 1.61 1.56 0.044 -0.023 0.011 0.055 0.12 0.16 0.060 -0.013 0.005 0.030 0.10 0.11 - 1.62 0.49 1.72 1.43 1.57 1.57 Co whose d-electron number is optimum for ferromag- netism. It is confirmed from TABLE II that the magne- tization of NFB(001)/Fe(100), NFB(110)/Fe(112), and NFB(100)/Co(0001) are stronger than that of the NFB single phase, and hence, these remanent magnetizations are expected to be improved in real materials. In partic- ular, the magnetizations of TM = Fe systems are higher than those of TM = Co systems because of the larger local moment of TM = Fe systems. As the degree of magnetization depends on the volume of the unit cell, a higher magnetization is exhibited for a shorter interlayer distance. (a) TM = Fe(110) (b) TM = Co(0001) 4 M2p is observed from 1.8 A to 1.85 A unlike the other local moments. We calculate M2p in more detail to be less than d = 1.90 A, and find that M2p shows varia- tional behavior with a small change in d below d = 1.83 A. It is likely that too small a value of d(= lFe 2−5) would result in overlapping atoms and the local moment being not appropriately evaluated. However, these unexpected results are not of major importance to our qualitative discussion.
a
PMA in TM=Fe
b
APMA in TM=Fe FIG. 4. Ion configurations at the NFB(100)/TM interface. The local magnetic moments of numbered ions are discussed in main text. To the unique property understand of NFB(100)/Fe(110), we analyze the local magnetic moments around the NFB(100)/TM interface. The ion positions for which the local moments are discussed here are shown in FIG. 4 (numbered sites "1," "2," and "3"). There are four Fe ions on the surface of NFB(100) in the unit cell, but we show the results of three of them because two lined up on the c-axis are at the symmetrical positions. In addition to the local moments at these sites, we also show three local moments on the surface of TM layers (numbered sites "4," "5," and "6" in FIG.4), which are the nearest neighbors to sites "1," "2," and "3," respectively. The interlayer-distance dependence of the local mag- netic moments at each site are shown in Fig. 5 [FIG. 5 (a) and (b) show the results of the TM = Fe system for PMA and APMA, respectively, and FIG. 5 (c) and (d) show the results of the TM = Co system for PMA and APMA, respectively]. We deliberately show the non-optimized results here in order to simplify the discussion. The ab- solute values of the local moment increase with increasing interlayer distance, d, owing to the electron localization, and these values approach the saturation values. Since the saturation magnetization of Co is less than that of Fe, the absolute values of M4, M5, and M6 (the magnetic moment at site "n" is represented by Mn) of the TM = Co system are smaller than those of the TM = Fe system. It is confirmed from FIG. 5 (a) that the nearest- neighbor local moments are strongly coupled to each other because the behaviors of M1p, M2p, and M3p (the subscript "p" indicates PMA) are almost similar to those of M4p, M5p, and M6p, respectively. These values reflect the strength of electron localization, which depends on inter-site distances (e.g., when d = 1.8, the inter-site dis- tance between sites "1" and "4," lFe 2−5 is 1.8 A, and lFe 3−6 is 1.91 A). The decreasing behavior of 1−4, is 2.32 A, lFe
B Μ
t n e m o M c i t e n g a M 3.0 2.5 2.0 1.5 1.0 0.5 4 p ç ae 1 p 6 p ó ò 3 p à á 0.0 1.80 3.0 2.5 2.0 1.5 1.0 0.5
B Μ
t n e m o M c i t e n g a M 2 p à 1 p ae ò á 4 p ç ó 6 p 0.0 1.80 ae ç ó ò à á ae ç ó ò à á ç ae ó ò à á ç ae ó ò 5 p á à 2 p 1.85 1.90 1.95 2.00 Interlayer Distance
Þ

c
PMA in TM=Co à ae ò 5 p á ç ó à ae ò 3 p á ç ó à ae ò á ç ó à ae ò á ç ó
B Μ
t n e m o M c i t e n g a M
B Μ
t n e m o M c i t e n g a M 1.85 1.90 1.95 2.00 Interlayer Distance
Þ
3 2 1 0 3 a ò 1 a ae à 2 a -1 -2 -3 5 a á ó 6 a ç 4 a 1.80 3 2 1 0 -1 -2 2 a à ò 3 a ae 1 a 6 a ó ç 4 a á 5 a -3 1.80 ò ae à á ó ç ae ò à á ó ç ae ò à á ó ç ae ò à á ó ç 1.85 1.90 1.95 2.00 Interlayer Distance
Þ

d
APMA in TM=Co à ò ae ó ç á à ò ae ó ç á à ae ò ó ç á à ae ò ó ç á 1.85 1.90 1.95 2.00 Interlayer Distance
Þ
FIG. 5. Magnetic moments at the numbered sites of the NFB(100)/TM interface shown in FIG. 4. The dashed lines represent the optimized interlayer distance for each system. In FIG. 5 (b), a clear increase of M1a (the subscript "a" indicates APMA) is observed from 1.8 A to 1.9 A, while M4a, the site of which is the nearest neighbor to site "1," hardly depends on d. These results indicate that M1 is easily coupled to the magnetic moments of the TM layer, rather than to those of NFB. We confirm that the local density of states (LDOS) at site "1" for APMA strongly depend on d, whereas the LDOS at site "1" for PMA hardly depend on it [see FIG. 6 (a) and (b)]. Figure 6 (b) shows that the band splitting is re- duced with decreasing d, which implies an enhancement of the electron itinerant property. A reduction in band splitting with decreasing d is also observed at site "2" for APMA, while it is not clearly observed in the other systems. We conclude that itinerant magnetism at the interface is preferred in APMA more than in PMA and that it lowers the total electron energy by overcoming the increase of exchange coupling energy. The local moments for PMA in the TM = Co sys- tem shown in FIG. 5 (c) reflects the coupling strength, which depends on inter-site distances, similar to FIG. 5 (a) (e.g., when d = 1.8, the inter-site distance between 2−5 is 2.17 A, and lCo sites "1" and "4," lCo 3−6 is 1.81 A). These local moments shown in FIG. 5 (c) are not as dependent on d as those in FIG. 5 (a) are, be- cause the magnetic stiffness of Co is greater than that 1−4, is 1.8 A, lCo of Fe and the magnetization of Co is less than that of Fe. Ma in the TM = Co system clearly increases from 1.8 A to 1.9 A, and it is confirmed from FIG. 6 (c) and (d) that the band splitting at site "1" for APMA is re- duced with decreasing d by more than that for PMA is. These results are similar to the results for the TM = Fe system, while the behavior of M2a and LDOS at site "2" are different between the TM = Co and TM = Fe sys- tems. The inter-site distance between site "2" and "5" of the TM = Co system is greater than that of the TM = Fe system (e.g., when d = 1.8, lCo 2−5 is greater than 2−5 by 0.37 A) and a reduction in band splitting than lFe with increasing d is not clearly observed at site "2" in the TM = Co system, unlike in the TM = Fe system. This band-splitting effect is not observed in other sys- tems of TM = Co, and hence, at the interface of the TM = Co system, itinerant magnetism does not grow as much as in the TM = Fe system. Moreover, a notice- able decrease in M1a, which reflects an enhancement of the itinerant property, occurs near the optimized value of d in the case of TM = Fe, whereas in the case of TM = Co, a noticeable decrease occurs at a value of d much smaller than the optimized value. Therefore, anti-parallel states are less likely to occur in real materials of TM = Co than in those of TM = Fe. The same can be said of NFB(001)/TM and NFB(110)/TM because their op- timized interlayer distances are much greater than those of NFB(100)/TM (see FIG. 3). 3−6 and lCo It should be noted that the values of M3a in both the TM = Fe and Co systems are not as dependent on d as the values of M1a are in the TM = Fe and Co sys- tems, even though lFe 3−6 are not as long as lFe 1−4 and lCo 1−4. One might think that this is attributed to the fact that M1 is more isolated than M3 owing to Nd ions [i.e., site "1" is located in the (001) plane of NFB, in which Nd ions are located, while site "3" is not located in this plane]. In practice, however, Nd ions do not affect the magnetic property of NFB(001)/Fe(110). We calcu- late the NFB(100)/Fe(110) system by replacing Nd ions, which are the nearest neighbors to sites "1" and "3", with Fe ions, and confirm that the magnetic properties and in- terface local moments of this substituted system are sim- ilar to those of the unsubstituted system. Therefore, we conclude that the interface coupling property is mainly dependent on the interface structure (i.e., inter-atomic distance, coordination number, and atomic positional re- lationship). IV. SUMMARY We studied the interface magnetic properties of Nd2Fe14B (NFB)/TM (TM = Fe, Co) multilayer ex- change spring magnets on the basis of first-principles cal- culations. Assuming a collinear spin structure, we opti- mized the structure of NFB(001)/TM, NFB(100)/TM, and NFB(110)/TM, and discussed their resultant mag- netic properties. in the remanent Improvements 5
V e
1
S O D L
V e
1
S O D L 2 1 0 -1 -2 2 1 0 -1 -2
a
PMA in TM=Fe d = 1.9 Þ d = 1.8 Þ (cid:173) ¯ -4 -2 0 2 ¶-¶F
eV

c
PMA in TM=Co d = 1.9 Þ d = 1.8 Þ (cid:173) ¯ -4 -2 0 2 ¶-¶F
eV

V e
1
S O D L
V e
1
S O D L 2 1 0 -1 -2 2 1 0 -1 -2
b
APMA in TM=Fe d = 1.9 Þ d = 1.8 Þ (cid:173) ¯ -4 -2 0 2 ¶-¶F
eV

d
APMA in TM=Co d = 1.9 Þ d = 1.8 Þ (cid:173) ¯ -4 -2 0 2 ¶-¶F
eV
FIG. 6. Local density of states (LDOS) of d-electrons at site "1" in NFB(100)/TM. The origin of the horizontal axis is the Fermi energy. magnetization were observed in NFB(001)/Fe(100), NFB(110)/Fe(112), and NFB(100)/Co(0001), and it was found, as expected, that the model with TM = Fe, rather than that with TM = Co, is advantageous to remanence. The optimized interlayer distance is short- est for NFB(100)/Fe(110), and anti-parallel magneti- zation alignment between NFB and Fe layers is pre- ferred. From the analysis of interface magnetic mo- ments and local DOS of NFB(100)/Fe(110), it was found that interface exchange splitting is reduced with de- creasing interlayer distance. Therefore, the total en- ergy of NFB(100)/Fe(110) is lowered by enhancing the interface itinerant property through anti-ferromagnetic coupling. On the other hand, in NFB(100)/Co(0001), interface magnetic moments are not so dependent on the interlayer distance, because the magnetic stiffness of Co is greater than that of Fe. From the analy- sis of local DOS, it was confirmed that the interface itinerant property of NFB(100)/Co(0001) is lower than that of NFB(100)/Fe(110) owing to the difference of in- terface structures. Moreover, the optimized interlayer distance of NFB(100)/Co(0001) is greater than that of NFB(100)/Fe(110), and it was predicted that interface anti-ferromagnetic coupling is less likely to occur in real materials of NFB(100)/Co(0001). Thus, it is concluded that the magnetic performance of exchange spring mul- tilayer magnets is strongly related to inter-site distance around the interface and magnetic stiffness of the soft magnetic phase. Proper material selection for the soft magnetic layer and more precise surface control tech- niques are indispensable for improving the properties of exchange spring magnets. 6 ACKNOWLEDGMENTS This work was supported by CREST-JST. 1 R. Coehoorn, D. B. de Mooij, and C. de Waard, J. Magn. 15 L. Jun, L. Ying, G. Jing, Z. Xiaoxi, and M. Yilong, J. Magn. Mater. 80, 101 (1989). Magn. Magn. Mater. 328, 1 (2013). 2 E. F. Kneller and R. Hawig, IEEE Trans. Magn. 27, 3588 16 W. B. Cui, Y. K. Takahashi, and K. Hono, Adv. Mater. (1991). 24, 6530 (2012); Acta Mater. 84, 405 (2015). 3 R. Skomski and J. M. D. Coey, Phys. Rev. B 48, 15812 17 T. Schrefl, J. Fidler, and H. Kronmuller, Phys. Rev. B 49, (1993). 6100 (1994). 4 T. Leineweber and H. Kronmuller, J. Magn. Magn. Mater. 18 R. F. Sabiryanov and S. S. Jaswal, Phys. Rev. B 58, 12071 176, 145 (1997). (1998). 5 N. M. Saiden, T. Schrefl, H. A. Davies, and G. Hrkac, J. 19 M. Amato, M. G. Pini, and A. Rettori, Phys. Rev. B 60, Magn. Magn. Mater. 365, 45 (2014). 3414 (1999). 6 G. P. Zhao and X. L. Wang, Phys. Rev. B 74, 012409 20 G. Asti, M. Solzi, M. Ghidini, and F. M. Neri, Phys. Rev. (2006). 7 Y. Q. Li, M. Yue, Q. Wu, T. Wang, C. X. Cheng, and H. X. Chen, J. Magn. Magn. Mater. 394, 117 (2015). 8 M. Shindo, M. Ishizone, H. Kato, T. Miyazaki, and A. Sakuma, J. Magn. Magn. Mater. 161, L1 (1996); M. Shindo, M. Ishizone, A. Sakuma, H. Kato, and T. Miyazaki, J. Appl. Phys. 81, 4444 (1997). 9 Z. C. Wang, H. A. Davies, S. Z. Zhou, M. C. Zhang, and Y. Qiao, J. Appl. Phys. 91, 7884 (2002). 10 W. Liu, Z. D. Zhang, J. P. Liu, B. Z. Cui, X. K. Sun, J. Zhou, and D. J. Sellmyer, J. Appl. Phys. 93, 8131 (2003); W. Liu, Y. C. Sui, J. Zhou, X. K. Sun, C. L. Chen, Z. D. Zhang, and D. J. Sellmyer, ibid. 97, 10K303 (2005). 11 N. Lupu, M. Grigoras, M. Lostun, and H. Chiriac, J. Appl. Phys. 105, 07A738 (2009). 12 K. P. Su, Z. W. Liu, H. Y. Yu, X. C. Zhong, W. Q. Qiu, and D. C. Zeng, J. Appl. Phys. 109, 07A710 (2011). 13 M. Urse, M. Grigoras, N. Lupu, and H. Chiriac, J. Phys.: Conf. Ser. 303, 012009 (2011). 14 D. Ogawa, K. Koike, S. Mizukami, M. Oogane, Y. Ando, T. Miyazaki, and H. Kato, J. Magn. Soc. Jpn. 36, 5 (2012). B 69, 174401 (2004). 21 E. E. Fullerton, J. S. Jiang, M. Grimsditch, C. H. Sowers, and S. D. Bader, Phys. Rev. B 58, 12193 (1998); E. E. Fullerton, J. S. Jiang, and S. D. Bader, J. Magn. Magn. Mater. 200, 392 (1999). 22 J. S. Jiang and S. D. Bader, J. Phys.: Condens. Matter 26, 064214 (2014). 23 R. Horikawa, H. Fukunaga, M. Nakano, and T. Yanai, J. Appl. Phys. 115, 17A707 (2014). 24 R. P. Pernechele, M. Solzi, and C, J. Phys. D: Appl. Phys. 47, 115002 (2014). 25 Y. Toga, H. Moriya, H. Tsuchiura, and A. Sakuma, J. Phys.: Conf. Ser. 266, 012046 (2011). 26 D. Ogawa, K. Koike, S. Mizukami, T. Miyazaki, M. Oogane, Y. Ando, and H. Kato, Appl. Phys. Lett. 107, 102406 (2015). 27 J. Herbst, Rev. Mod. Phys. 63, 819 (1991). 28 K. Momma and F. Izumi, J. Appl. Crystallogr. 41, 653 (2008). 29 G. Kresse and J. Hafner, Phys. Rev. B 47, 558 (1993); G. Kresse and J. Furthmuller, 54, 11169 (1996). 30 S. H. Vosko, L. Wilk, and M. Nusair, Can. J. Phys. 58, 1200 (1980). 

1102.1758

2

1102

Low Energy Electron Point Projection Microscopy of Suspended Graphene, the Ultimate "Microscope Slide"

Point Projection Microscopy (PPM) is used to image suspended graphene using low-energy electrons (100-200eV). Because of the low energies used, the graphene is neither damaged or contaminated by the electron beam. The transparency of graphene is measured to be 74%, equivalent to electron transmission through a sheet as thick as twice the covalent radius of sp^2-bonded carbon. Also observed is rippling in the structure of the suspended graphene, with a wavelength of approximately 26 nm. The interference of the electron beam due to the diffraction off the edge of a graphene knife edge is observed and used to calculate a virtual source size of 4.7 +/- 0.6 Angstroms for the electron emitter. It is demonstrated that graphene can be used as both anode and substrate in PPM in order to avoid distortions due to strong field gradients around nano-scale objects. Graphene can be used to image objects suspended on the sheet using PPM, and in the future, electron holography.

cond-mat.mes-hall

cond-mat

Low Energy Electron Point Projection Microscopy of Suspended Graphene, the Ultimate "Microscope Slide" J.Y. Mutus1,2, L. Livadaru1,2, J.T. Robinson3, R. Urban1,2, M.H. Salomons1,2, M. Cloutier1,2, and R. A. Wolkow1,2 1Department of Physics, University of Alberta, 11322 - 89 Avenue, Edmonton, Alberta T6G 2G7, Canada. 2National Institute for Nanotechnology, National Research Council of Canada, 11421 Saskatchewan Drive, Edmonton, Alberta T6G 2M9, 3Naval Research Laboratory, Washington, D.C. 20375 Canada October 24, 2018 Abstract Point Projection Microscopy (PPM) is used to image suspended graphene using low-energy electrons (100-200eV). Because of the low energies used, the graphene is neither damaged or contaminated by the electron beam. The transparency of graphene is measured to be 74%, equivalent to electron transmission through a sheet as thick as twice the covalent radius of sp2-bonded carbon. Also observed is rip- pling in the structure of the suspended graphene, with a wavelength of approximately 26 nm. The interference of the electron beam due to the diffraction off the edge of a graphene knife edge is observed and used to calculate a virtual source size of 4.7 ± 0.6A for the electron It is demonstrated that graphene can serve as both anode emitter. and substrate in PPM, thereby avoiding distortions due to strong field gradients around nano-scale objects. Graphene can be used to image objects suspended on the sheet using PPM, and in the future, electron holography. 1 1 Introduction Point projection microscopy, or PPM, may be the simplest implementation of electron microscopy, comprising only an electron source, a nearby sample, and an electron imaging screen some distance away (Figure 1). Electrons from the point source pass through the sample and their initial radial distri- bution naturally results in a magnified image on the detector. PPM offers many advantages over conventional electron microscopy (EM), most impor- tantly the removal of aberration-inducing lenses, extremely low acceleration voltages on the order of 100eV, and the collection of phase information to form in-line, low-energy electron point-source (LEEPS) holograms [1, 2, 3, 4]. Due to the low-energy beam, PPM can detect the smallest amount of con- tamination with higher contrast than in conventional EM techniques. Also, the low-energy electron beam neither induces damage nor deposits carbon contamination (as seen in Figure 6a) and, as a result, the same region can be imaged repeatedly for hours on end without any apparent change in opac- ity or structure, which is a significant advantage over high-energy electron microscopy known to induce structural changes to graphene [5, 6]. PPM offers many advantages for studying the structure of graphene, diagnosing the quality of samples and measuring the effective attenuation length of a single and multi-layer graphene sheets at low energies. The principal tech- nical challenges of PPM relate to the sample, which must be thinner than a few atomic layers (depending on the material) or must span a gap. Since strong electrostatic fields surround any sample in the latter case (as seen in Figure 2), electron trajectories are greatly perturbed in their vicinity and the resulting PPM images are distorted[4, 7]. 2 Why Graphene? Graphene has inspired a great deal of research recently due to its unique elec- tronic and mechanical properties. Moreover, it can also serve as a nearly ideal microscope slide for electron microscopy[8] since it is virtually trans- parent to electrons, even at low energies as demonstrated here and it is elec- trically conductive and mechanically robust. In addition, for the purpose of PPM, graphene can serve as both anode and substrate. Nanoscale objects can be deposited on the graphene, providing a nearly flat grounded plane for electrons and thus avoiding distortion due to the high field gradients that form around suspended biased nanoscale objects. These distortions complicate interpretation and holographic reconstruction of PPM images 2 [9, 7]. 3 Experimental Details Graphene was imaged in PPM using a custom-built apparatus. The micro- scope is contained in a magnetically shielded [10] ultra-high vacuum (UHV) chamber with a base pressure < 1x10−10Torr. Approaching the tip to the sample is critical in PPM and was accomplished using serial coarse and fine positioners. The coarse positioners were manufactured by Attocube Sys- tems and, like the rest of the microscope, consist entirely of non-magnetic components. These positioners move the tip to within one millimetre of the sample and have a 5 mm travel range in x, y, and z, with step sizes from a few to several hundred nanometres. Fine positioning is done using a piezo- electric tube scanner that routinely achieved sub-nanometre accuracy. The graphene sheet spans a gap in a TEM grid just below the tip. The elec- trons are transmitted through the sample towards a 2-stage chevron style micro-channel plate (MCP) and then to a phosphor screen[11]. Images are recorded with a high dynamic range, 12-bit CCD camera[12] using 10-30ms exposure times. Typically, 200-400 images are captured. The images are aligned to compensate lateral drift (few nanometers per minute) and av- eraged to improve signal-to-noise. This enhances the image contrast while averaging out the structure of the detector itself. This treatment causes the edges of some of the images to appear blurred. Several gross defects in the detector remain, such as the two large dark spots visible in Figure 3b. Fields of view ranging from tens of nanometres to several millimetres are available and useful for finding small features on relatively large sam- ples. Different regions of the sample can be examined by translating the tip relative to the sample using our combination of coarse and fine positioners. Higher magnification is achieved by moving the tip closer to the sample. Our apparatus enables us to use field ion microscopy (FIM) to employ a nitrogen-assisted etching process to shape a tip from a nearly spherical, many atom apex to a single atom[13] before imaging. With the sample grounded, a voltage in the range -80V to -1100V, depending on the tip- sample distance and overall tip shape, is applied to field-emit electrons from the tip through the sample and towards the MCP. Beam currents for imaging can range from a few pA to nA , although typically tens of pA are required to generate a sufficient signal-to-noise ratio. The manufacture and operation of the microscope will be the subject of a subsequent publication. 3 4 Sample Preparation Graphene was synthesized using low-pressure chemical vapour deposition (CVD) on copper foils[14, 15]. Briefly, 25µm thick Cu foils were heated to 1030oC under flowing hydrogen (PH2 ≈ 600 mTorr). At the growth temperature, the H2 pressure was decreased (< 50 mtorr) and methane was introduced (PCH4 ≈ 200 mtorr) for 20 minutes, after which the sample was quenched to room temperature. Subsequent to growth, graphene was transferred to metal-coated perforated silicon nitride membranes Cr(5nm)/ Au(45nm) /SiN(50nm)[16] with 2µum holes using techniques described in [14, 15]. To remove residues and contaminants the samples were then annealed in UHV at 300oC to 450oC for 45 minutes to 8 hours. The effect on the cleanli- ness of the graphene of different annealing temperatures and times is readily seen in PPM images (Figure 5). Samples with a low level of contamination - mostly agglomerated at grain boundaries - are seen only after annealing for at least 8 hours. Cleaning using UV/Ozone [17] resulted in graphene sheets that were nearly totally opaque to electrons at energies below 200 eV de- spite the graphene appearing unchanged in Raman spectroscopy, presenting a matter for further investigation. 5 Results and Discussion The PPM images of graphene presented in this work are also in-line holo- grams: part of the electron wave is scattered off the sample, this par- tially scattered wave interferes with electrons that arrive at the detector unscattered[18]. When looking at clean graphene in PPM (Figure 3) several features stand out: (I) disordered lines, (II) graphene texture, (III) transparency, and (IV) interference fringes. (I) Imaging reveals, 30 nm wide disordered lines criss-crossing the sam- ple. Scanning transmission electron micrographs (STEM) of the sample (Figure 6) indicate that these are composed of contaminants adhering to grain boundaries within the graphene films, similar to those seen in AFM and STEM[19]. Also, the lines and tears in the graphene sheet are at rel- ative angles of 60o or 120o, indicating that growth and failure within the sheets are aligned with the crystallographic directions of the graphene (e.g. Figure 3). (II) A more subtle feature is the texture of the graphene itself consistent 4 with the presence of ripples in a direction normal to the graphene sheet. With a wavelength of approximately 13 nm, this rippling is consistent with previous experimental observations of graphene[20, 21, 22] and theoretical estimates[23, 24] of the corrugation in graphene. The average radial distri- bution function (RDF) for the dark features in this area of graphene, see Fig- ure 4, also shows the presence of ripple-like structure, while for the vacuum area the RDF has no such features, as expected. Possible causes and mech- anisms of ripple formation on graphene are: (i) edge-induced stress[24, 25]; (ii) thermal fluctuations[23]; and (iii) adsorbed OH molecules or other con- taminants, altering the bond length between carbon atoms[26]. (III) Consistent with earlier reports of high electron-transparency, it is found here that very low energy electrons are transmitted through graphene with modest attenuation of about 25%. For the purposes of developing PPM and LEEPS holography, this observation demonstrates that graphene will be useful as a "microscope slide" for supporting molecular and other nano-scale samples for LEEPS holography. In addition to being largely transparent, the conductive nature of graphene remedies the field-distortion demonstrated in Figure 2 by providing a planar equipotential surface, typically at ground potential, in the vicinity of a nano-scale sample. As a result, PPM images and holograms using graphene substrates will be rendered more amenable to direct interpretation and digital reconstruction. Graphene is expected to be transparent to electrons with energies ranging from a few keV down to a few eV[27]. As electron scattering becomes stronger with decreasing energy, transparency decreases. Both elastic and inelastic scattering mechanisms are typically present in a sample and their effect is usually expressed as corresponding electron mean free paths (MFP) in the material. The overall effect of electron interaction with matter is captured as an effective attenu- ation length (EAL), denoted lEAL, which can be measured experimentally (e.g. by X-ray photoelectron spectroscopy). The values of these quantities are available in the literature and from established databases (e.g. those published by NIST[27]) for many samples and electron energies. For very thin carbon films (approaching the single-layer limit), experi- mental measurements are not abundant, but it is generally accepted that the EAL does not vary significantly from about 5A between 100 and 200 eV[28, 29]. The intensity of the electron beam after passing through a sheet of thickness h is given by, (cid:19) (cid:18) − h lEAL cos θ I(h) = Io exp , (1) where I0 is the incident beam intensity and θ is the incidence angle with 5 respect to the normal to the sheet. Assuming this continuum-limit formula may be extrapolated for a single- layer of graphene and assuming normal incidence, we calculate the trans- parency of single-layer graphene as given by T = I(h)/I0, where h is the thickness of graphene. For electrons of energy 100 eV, assuming h to be double the covalent radius of sp2-bonded carbon (1.46 A), we get a trans- parency of 75%, while assuming H to be the interlayer distance in graphite (3.35 A) yields a transparency value of 51%. In order to make an experimental estimate of the transparency of graphene we need to properly account for the profile of the field-emitted beam. As- suming the beam profile is of Gaussian form (plus a small constant), we optimized the beam parameters to maximize the uniformity of the inten- sity across the vacuum region and (by dividing our raw image by the opti- mal beam) obtained a flattened image. Using this latter image to estimate graphene transparency yields a value of 74%, very close to the theoretical estimate above using graphene thickness as double the covalent radius of sp2-bonded carbon. (IV) The ultimate resolution of this technique is limited by the coherence of the electron beam, a good measure of which is expressed by the virtual source size of our electron emitter. In our case, the resolution of the micro- scope is roughly equal to the virtual source size of the emitter[2]. The num- ber of Fresnel fringes at the graphene edges demonstrate the high coherence of the electron wave and are equivalent to those of a knife-edge interference experiment[30]. A maximum of 14 fringes have been found for the sample in Figure 3d, and the width of the fringe pattern, w, is related to the size of the virtual source of electrons. The images were taken at an energy of 124 eV (corresponding to an electron wavelength λ=1.1A) for a source-sample dis- tance of about 200 nm. The coherence angle of our beam can be estimated as the angular width of the interference pattern, γ = 2 tan−1(w/2L), where L is the source-to-detector distance. The coherence angle of our beam was estimated to be (4.3 ± 0.5) degrees. Using the van Cittert-Zernike theorem, we estimate the size of the virtual source according to [30] λ πγ . Ref f = (2) For the image in Figure 3d, this yields a value of 4.7 ± 0.6A for the virtual source size of our nanotip, in line with other experimental measurements[31, 32]. Note that this result is based on the theoretical interpretation of the knife-edge diffraction experiment performed with an opaque edge. However, our graphene edge in Figure 3d is not opaque, which could lead to errors 6 in the above estimate. A non-opaque edge should lead to a decrease in the contrast of the fringes and a washing out of the finest fringes. Therefore, the above estimate is in fact an upper limit of the source size. 6 Future Work Precise alignment and characterization of the geometry of our projection setup is required to reconstruct the holograms, this is an ongoing effort and will be present in future work. The lack of divergent electron beams has limited electron holography in the past[33], and through careful control of the shape of the apex of the tip[13] and the geometry of the PPM setup, larger coherence angles may be achieved allowing for increasingly accurate holographic reconstructions. The potential of graphene as a substrate for PPM and in-line electron holography will continue to be explored. Work is ongoing to benchmark the resolution of this technique using standards of well defined size, such as single-walled carbon nanotubes and gold nanoparticles. Holographic stud- ies of magnetic fields will also be of great interest. The fields emanating from magnetic nanoparticles and those due to various edge terminations of graphene will be of great interest. Our low energy approach offers greater sensitivity to field induced phase shifts than phase imaging techniques using high energy electrons. 7 Acknowledgements The authors would like the thank P.E Sheehan for facilitating this research and for his help with this paper. The authors are grateful for fruitful dis- cussion with a critical encouragement by B. Cho. The Microscopy group at the National Institute for Nanotechnology provided facilities for imaging the graphene sample in STEM. We would like the thank the Natural Sci- ences and Engineering Research Council (NSERC) of Canada and Alberta Innovates - Technology Futures for funding this research. References [1] HJ Kreuzer, K. Nakamura, A. Wierzbicki, H.W. Fink, and H. Schmid. Theory of the point source electron microscope. Ultramicroscopy(The Netherlands), 45(3):381 -- 403, 1992. 7 [2] Spence J.C.H. and Cowley J.M. Electron Holography at Low Energy, in Introduction to Electron Holography, Volkl, E. and Allard, L.F. and Joy, D.C., Eds. Plenum Pub Corp, 1999. [3] P. Morin, M. Pitaval, and E. Vicario. Low energy off-axis holography in electron microscopy. Physical review letters, 76(21):3979 -- 3982, 1996. [4] A. Beyer and A. Golzhauser. Low energy electron point source mi- croscopy: beyond imaging. Journal of Physics: Condensed Matter, 22:343001, 2010. [5] C.O. Girit, J.C. Meyer, R. Erni, M.D. Rossell, C. Kisielowski, L. Yang, C.H. Park, MF Crommie, M.L. Cohen, S.G. Louie, et al. Graphene at the Edge: Stability and Dynamics. Science, 323(5922):1705, 2009. [6] RF Egerton, P. Li, and M. Malac. Radiation damage in the TEM and SEM. Micron, 35(6):399 -- 409, 2004. [7] M. Prigent and P. Morin. Charge effect in point projection images of Ni nanowires and I collagen fibres. Journal of Physics D: Applied Physics, 34:1167, 2001. [8] Z. Lee, K.J. Jeon, A. Dato, R. Erni, T.J. Richardson, M. Frenklach, and V. Radmilovic. Direct Imaging of Soft- Hard Interfaces Enabled by Graphene. Nano letters, 9(9):3365 -- 3369, 2009. [9] M. Prigent and P. Morin. Charge effect in point projection images of carbon fibres. Journal of microscopy, 199(3):197 -- 207, 2000. [10] L. Livadaru, J. Mutus, and R.A. Wolkow. In-line holographic electron microscopy in the presence of external magnetic fields. Ultramicroscopy, 108(5):472 -- 480, 2008. [11] Hamamatsu Photonics. F2223-21P. [12] PCO AG. Pixelfly. [13] M. Rezeq, J. Pitters, and R. Wolkow. Tungsten nanotip fabrication by spatially controlled field-assisted reaction with nitrogen. The Journal of chemical physics, 124:204716, 2006. [14] Xuesong Li, Weiwei Cai, Jinho An, Seyoung Kim, Junghyo Nah, Dongx- ing Yang, Richard Piner, Aruna Velamakanni, Inhwa Jung, Emanuel Tutuc, Sanjay K. Banerjee, Luigi Colombo, and Rodney S. Ruoff. 8 Large-Area Synthesis of High-Quality and Uniform Graphene Films on Copper Foils. Science, 324(5932):1312 -- 1314, 2009. [15] Xuesong Li, Yanwu Zhu, Weiwei Cai, Mark Borysiak, Boyang Han, David Chen, Richard D. Piner, Luigi Colombo, and Rodney S. Ruoff. Transfer of large-area graphene films for high-performance transparent conductive electrodes. Nano Letters, 9(12):4359 -- 4363, 2009. PMID: 19845330. [16] Structure Probes Inc. P/N 4108PSN-BA. [17] John R. Vig. Uv/ozone cleaning of surfaces. Journal of Vacuum Science Technology A: Vacuum, Surfaces, and Films, 3(3):1027 -- 1034, May 1985. [18] D. Gabor. A new microscopic principle. Nature, 161(4098):777 -- 778, 1948. [19] P.Y. Huang, C.S. Ruiz-Vargas, A.M. van der Zande, W.S. Whitney, M.P. Levendorf, J.W. Kevek, S. Garg, J.S. Alden, C.J. Hustedt, Y. Zhu, et al. Grains and grain boundaries in single-layer graphene atomic patchwork quilts. Nature, 2011. [20] JC Meyer, AK Geim, MI Katsnelson, KS Novoselov, D. Obergfell, S. Roth, C. Girit, and A. Zettl. On the roughness of single-and bi-layer graphene membranes. Solid State Communications, 143(1-2):101 -- 109, 2007. [21] J.C. Meyer, AK Geim, MI Katsnelson, KS Novoselov, TJ Booth, and S. Roth. The structure of suspended graphene sheets. Nature, 446(7131):60 -- 63, 2007. [22] U. Bangert, MH Gass, AL Bleloch, RR Nair, and AK Geim. Manifes- tation of ripples in free-standing graphene in lattice images obtained in an aberration-corrected scanning transmission electron microscope. physica status solidi (a), 206(6):1117 -- 1122, 2009. [23] A. Fasolino, JH Los, and MI Katsnelson. Intrinsic ripples in graphene. Nature Materials, 6(11):858 -- 861, 2007. [24] VB Shenoy, CD Reddy, A. Ramasubramaniam, and YW Zhang. Edge- stress-induced warping of graphene sheets and nanoribbons. Physical review letters, 101(24):245501, 2008. 9 [25] W. Bao, F. Miao, Z. Chen, H. Zhang, W. Jang, C. Dames, and C.N. Lau. Controlled ripple texturing of suspended graphene and ultrathin graphite membranes. Nature Nanotechnology, 4(9):562 -- 566, 2009. [26] R.C. Thompson-Flagg, M.J.B. Moura, and M. Marder. Rippling of graphene. EPL (Europhysics Letters), 85:46002, 2009. [27] A. Jablonski, F. Salvat, and CJ Powell. NIST Electron Elastic- Scattering Cross-Section Database -- Version 3.1. National Institute of Standards and Technology, Gaithersburg MD). http://www. nist. gov/srd/nist64. htm, 2003. [28] C. Martin, ET Arakawa, and JC Callcottand. Low energy electron attenuation length studies in thin amorphous carbon films. Journal of Electron Spectroscopy and Related Phenomena, 35(2):307 -- 317, 1985. [29] A. Jablonski and CJ Powell. Relationships between electron inelas- tic mean free paths, effective attenuation lengths, and mean escape depths. Journal of Electron Spectroscopy and Related Phenomena, 100(1-3):137 -- 160, 1999. [30] JCH Spence, W. Qian, and MP Silverman. Electron source brightness and degeneracy from Fresnel fringes in field emission point projection microscopy. Journal of Vacuum Science & Technology A: Vacuum, Surfaces, and Films, 12(2):542 -- 547, 2009. [31] B. Cho, T. Ichimura, R. Shimizu, and C. Oshima. Quantitative evalu- ation of spatial coherence of the electron beam from low temperature field emitters. Physical review letters, 92(24):246103, 2004. [32] C.C. Chang, H.S. Kuo, I.S. Hwang, and T.T. Tsong. A fully coherent electron beam from a noble-metal covered W (111) single-atom emitter. Nanotechnology, 20:115401, 2009. [33] GB Stevens. Resolving power of lens-less low energy electron point source microscopy. Journal of Microscopy, 235(1):9 -- 24, 2009. 10 Figure 1: A diagram of the PPM experimental setup A biased, sharp metallic nanotip is brought close (100-5000nm) to a grounded grid. Electrons field emitted from the tip are projected through the sample, towards an electron-imaging screen (8cm away), resulting in a magnified image on the screen. 11 Figure 2: The electrostatic potentials due to a grounded discrete nanoscale object (left) and the same object suspended on graphene, modeled here as a thin grounded plane (right). The nanotip appears in blue, the small object is represented as a yellow circle, and graphene as a yellow horizontal line. The distance between tip apex and graphene is 200nm and the potential difference between them is 100V. In the first case, electrons emitted from the tip will have their paths distorted by the spatial inhomogeneity due to the field around the object while the grounded plane provides a flat anode mitigating distortions. 12 Figure 3: PPM Images of graphene a) An image of a portion of the graphene coated silicon nitride grid. The grid is perforated by 2um holes on a 4um pitch. The majority of the holes are covered entirely with graphene (as in top left), some are partially covered with graphene (top right) a few are totally uncovered (bottom middle). Note the straight lines crisscrossing the image, these are thought to be grain boundaries and/or wrinkles in the graphene. The lines are evidently decorated by leftover contaminants. b) A zoomed in portion of the partially covered hole from (a). The lines are clearly visible. The uncovered portion is in the top left of the image. Note the diagonal lines and what are evidently folded back portions along the hole. Also of note is the faceted nature of the edges of the hole. c) A zoomed in portion of the area indicated by the arrow in (b) These objects are small enough that they only partially scatter the electron beam. The interference pattern between the scattered and unscattered portion of the beam forms a hologram. d) Many highly visible fringes appear along the edge of the graphene sheet as we zoom in further. Inset is a profile along yellow line. Also, the interference due the diffraction around the contaminants along the lines become more visible. The voltage between the sample and the tip, along with the emission current is displayed in the bottom left corner of each image. 13 1µm -205V, 9pA 40 nm -124V, 30pA 200nm -175V, 21pA a) b) 40 nm -124V, 24pA d) c) Figure 4: A further zoomed-in portion of the partially covered hole. The graphene sheet appears to have cleaved along a grain boundary or fold, forming an angle of 120 degrees. Also evident is the fine structure to the contamination and the fringes visible from diffraction by contaminants. The graphene itself is rippled. To quantify the ripples, an average RDF of the graphene covered area and vacuum area are plotted together. The RDF of over the graphene is peaked around 13nm, while the RDF over vacuum shows no such structure. 14 50 nm -140V 21pA 0510152025Radius [nm]105110115120125RDFGrapheneVacuumAverage RDF over disks of radius 27nm Figure 5: The effects of contamination on PPM of graphene The intensity of the electron beam through the graphene sheet is very sensitive to the cleaning process. In each image the graphene is on the left and vacuum region is on the right. The graphene was prepared by annealing in UHV at a) 300oC for 40 minutes (the hexagonal pattern is the structure of the detector) b) 420oC for 40 minutes c) 300oC for 90 minutes and d) above 400oC for 8 hours. The sample in c) first underwent UV/Ozone treatment 15 a) b) d) Tip Voltage: -140V Tip Voltage: -168V Tip Voltage: -176V c) Tip Voltage: -112V Figure 6: STEM images of graphene a) a typical graphene covered hole in the SiN membrane. The same lines seen in PPM are visible here. The contamination induced by the high energy electron beam creates the many dark squares scattered around the image. No such contamination is visible in PPM. b) A magnified image of the portion in the white square in a). A line of discrete particles (most likelyleftover Cu nanoparticles) decorate what is likely a grain boundary. 16 a) b) 

1304.6647

1

1304

Topological effects and particle-physics analogies beyond the massless Dirac-Weyl fermion in graphene nanorings

Armchair and zigzag edge terminations in planar hexagonal and trigonal graphene nanorings are shown to underlie one-dimensional topological states associated with distinctive energy gaps and patterns (e.g., linear dispersion of the energy of an hexagonal ring with an armchair termination versus parabolic dispersion for a zigzag terminated one) in the bands of the tight-binding spectra as a function of the magnetic field. A relativistic Dirac-Kronig-Penney model analysis of the tight-binding Aharonov-Bohm behavior reveals that the graphene quasiparticle in an armchair hexagonal ring is a condensed-matter realization of an ultrarelativistic fermion with a position-dependent mass term, akin to the zero-energy fermionic solitons with fractional charge familiar from quantum field theory and from the theory of polyacetylene. The topological origins of the above behavior are highlighted by contrasting it with the case of a trigonal armchair ring, where we find that the quasiparticle excitations behave as familiar Dirac fermions with a constant mass. Furthermore, the spectra of a zigzag hexagonal ring correspond to the low-kinetic-energy nonrelativistic regime of a leptonlike massive fermion. A onedimensional relativistic Lagrangian formalism coupling a fermionic and a scalar bosonic field via a Yukawa interaction, in conjunction with the breaking of the Z2 reflectional symmetry of the scalar field, is shown to unify the above dissimilar behaviors.

cond-mat.mes-hall

cond-mat

Topological effects and particle-physics analogies beyond the massless Dirac-Weyl fermion in graphene nanorings Igor Romanovsky,∗ Constantine Yannouleas,† and Uzi Landman‡ School of Physics, Georgia Institute of Technology, Atlanta, Georgia 30332-0430 (Dated: 30 December 2012; Phys. Rev. B 87, 165431 (2013)) Armchair and zigzag edge terminations in planar hexagonal and trigonal graphene nanorings are shown to underlie one-dimensional topological states associated with distinctive energy gaps and patterns (e.g., linear dispersion of the energy of an hexagonal ring with an armchair termination versus parabolic dispersion for a zigzag terminated one) in the bands of the tight-binding spectra as a function of the magnetic field. A relativistic Dirac-Kronig-Penney model analysis of the tight- binding Aharonov-Bohm behavior reveals that the graphene quasiparticle in an armchair hexagonal ring is a condensed-matter realization of an ultrarelativistic fermion with a position-dependent mass term, akin to the zero-energy fermionic solitons with fractional charge familiar from quantum field theory and from the theory of polyacetylene. The topological origins of the above behavior are highlighted by contrasting it with the case of a trigonal armchair ring, where we find that the quasiparticle excitations behave as familiar Dirac fermions with a constant mass. Furthermore, the spectra of a zigzag hexagonal ring correspond to the low-kinetic-energy nonrelativistic regime of a leptonlike massive fermion. A onedimensional relativistic Lagrangian formalism coupling a fermionic and a scalar bosonic field via a Yukawa interaction, in conjunction with the breaking of the Z2 reflectional symmetry of the scalar field, is shown to unify the above dissimilar behaviors. PACS numbers: 73.22.Pr, 73.21.-b, 11.10.-z, 73.23.Ra I. INTRODUCTION Since its inception, relativistic quantum mechanics has been associated mainly with the fields of particle and high-energy physics.1 -- 4 Recently, however, a table- top version of relativistic quantum physics emerged, fol- lowing the experimental isolation of graphene, which is a single-layer, planar honeycomb lattice of carbon atoms.5 Indeed, the two-dimensional (2D) quasiparticle excita- tions of neutral graphene near the Fermi level behave6,7 as massless neutrinolike fermions described by the cel- ebrated Dirac-Weyl8 (DW) equation. The scientific and technological potential for exploiting charge carriers and quasiparticles with relativistic behavior in tunable condensed-matter and atomic-physics systems is attract- ing much attention.9 -- 13 In this context, an important question, as yet only partly explored, remains whether quasi-onedimensional (1D) graphene systems support ex- clusively DW massless or constant-mass Dirac fermions, or they can induce relativistic-quantum-field (RQF) be- haviors that require consideration of position-dependent mass terms, reflecting generalized underlying bosonic scalar fields. In this paper, we show that planar graphene nanor- ings do indeed exhibit a rich variety of physics, ranging from sophisticated RQF regimes to more familiar cases of constant-mass fermions (both in the relativistic and non- relativistic regimes). The emergence of these physical regimes depends on the specific combination of topologi- cal factors associated with modifications of the graphene lattice, such as the type of edge termination (i.e., arm- chair or zigzag) and the shape (i.e., hexagonal or trigonal) of the graphene ring. To this end, we investigate the properties of the cor- responding tight-binding7,14,15 (TB) spectra and of the associated Aharonov-Bohm16 (the AB effect, which is a hallmark topological effect in condensed matter systems) oscillations of the magnetization, as a function of the magnetic field B. We then analyze the spectra and AB characteristics with the help of a Dirac-Kronig-Penney (DKP) superlattice model,2,17 in the spirit of the vir- tual "band-structure" model for the nonrelativistic AB effect.18 We find that the relativistic behavior in arm- chair rings requires a profound modification of the 1D Dirac equation1 through the introduction of a position- dependent mass term, in analogy with the fractional- charge, zero-energy topological modes in quantum field theory and in the theory of trans-polyacetylene.19 -- 22 In contrast, the zigzag-ring spectra may correspond to the low-energy nonrelativistic regime of a leptonlike23 mas- sive particle, heavier than the electron. Planar graphene rings (and the associated AB spectra) have been recently investigated by a number of groups us- ing tight-binding15,24 -- 28 methods (for polygonal shapes with armchair or zigzag terminations), as well as con- tinuum DW24,29,30 equations supplemented with infinite- mass boundary conditions (for idealized circular shapes). These earlier studies did not address the question of pos- sible analogies to 1D quantum field theoretical models and particle physics. We stress that a prerequisite to raising and answering this question is the introduction by us of the virtual DKP superlattice model for the AB effect. For a review on recent experimental studies of the Aharonov-Bohm effect in graphene nanorings, see Ref. 31. In addition to planar graphene rings, graphene nanoribbons (GNRs) are another class of related quasi- 1D systems. GNRs have attracted substantially more attention than graphene rings and their study gave rise to a vast body of theoretical32 -- 35 and experimental36 lit- erature. For graphene nanoribbons, it was found that a gap ∆0 may open at the Fermi energy, leading to an apparent analogy with the constant-mass Dirac fermion [see Ref. 32(c)]. As elaborated below, for an armchair nanoribbon (aGNR), this gap arises from the topology of the armchair edge which mimicks the dimerized do- mains (i.e., formation of Kekul´e unequal carbon bonds) in trans-polyacetylene.19 -- 22 Then in analogy with the scalar Z2 kink-soliton associated with the Peierls tran- sition in trans-polyacetylene (or equivalently with the Z2 kink-soliton used in the Jackiw-Rebbi fermionic RQF model37) the qualitative features of the (fermionic) AB spectra of armchair graphene rings can be understood as resulting from an alternation (or lack of it) of two degen- erate dimerized domains associated with the arms of the graphene ring. We stress that the effective dimerization in the armchair GNRs and armchair graphene nanor- ings has a topological origin imposed by the presence of the armchair edges, while the dimerization in trans- polyacetylene is due to the Peierls instability.21,38 These two different underlying processes, however, lead to sim- ilar results that are characterized by the breaking of the 1D Z2 reflectional symmetry (see in particular Secs. IV A and V below). Our findings of quasiparticles in graphene with gen- eral position-dependent, and/or constant (rest), masses (unlike the massless neutrinolike quasiparticle in 2D graphene) is particularly interesting in light of increas- ing current interest in mass acquisition mechanisms, e.g., the Higgs mechanism in elementary particles39 -- 41 and condensed-matter physics.42,43 The predicted unprecedented emergent unfolding of fundamentally distinct physical regimes, namely com- plex quantum-field theoretical ones versus nonrelativis- tic constant-mass ones, depending solely on the mate- rials' shape and edge termination is to date unique to graphene. It will be of great interest to test signa- tures of such regime-crossover experimentally for specifi- cally prepared graphene systems with atomic precision,36 as well as to explore possible occurrence of such topological-in-origin physical behavior in other designer- Dirac-fermion artificial systems9,12 or nanopatterned ar- tificial graphene.44 The plan of the paper is as follows. Sec. II describes the Aharonov-Bohm tight-binding spectra for three characteristic planar graphene nanor- ings, i.e., an armchair hexagonal ring, an armchair trig- onal ring, and a zigzag hexagonal ring. Sec. III introduces the theoretical aspects of a rel- ativistic 1D Dirac-Kronig-Penney model, based on the generalized Dirac equation. The DKP model describes the virtual superlattice associated with the Aharonov- Bohm effect. Sec. IV presents the DKP interpretation of the tight- binding spectra calculated in Sec. II. The quasiparticle excitations in graphene nanorings are shown to exhibit 2 behavior associated with quantum field theoretical mod- els for elementary particles beyond the massless Dirac- Weyl fermion. Sec. V discusses the full relativistic quantum field La- grangian formalism that underlies the DKP interpreta- tion elaborated in Sec. IV. The Lagrangian formalism shows that the physics of quasiparticle excitations in pla- nar graphene nanorings relates to mass acquisition and formation of fermionic solitons. Finally, Sec. VI presents our conclusions. II. TIGHT-BINDING CALCULATIONS In this section, we will describe the TB spectra and corresponding AB magnetizations (as a function of the magnetic flux) for three characteristic cases of planar graphene rings, and specifically for a hexagonal arm- chair ring, a trigonal armchair ring, and a hexagonal zigzag ring (all three of similar dimensions). We note that the arms of the armchair rings studied here corre- spond to the class of perfect armchair nanoribbons re- ferred to as metallic;32(b),33 -- 35 they exhibit a vanishing energy gap, ∆0 = 0, between the valence and conduction bands in tight-binding and continuum DW calculations. However, any perturbation (including the incorporation into a nanoring structure) may result32(b) in the open- ing of a gap ∆0 > 0. The metallic aGNRs have a width corresponding to NW = 3l − 1, l = 1, 2, 3, . . . carbon atoms. Counting along a zigzag line in the middle of the arm (away from the corners), the hexagonal and trigonal armchair rings studied in this paper45 have NW = 14. To determine the single-particle spectrum [the energy in the tight-binding calculations for the levels εi(B)] graphene nanorings, we use the hamiltonian HTB = − (cid:88) † tijc i cj + h.c., (1) with <> indicating summation over the nearest-neighbor sites i, j. The hopping matrix element tij = tij exp ds · A(r) , (2) <i,j> (cid:18) ie (cid:90) rj ¯hc ri (cid:19) where ri and rj are the positions of the carbon atoms i and j, respectively, and A is the vector potential associ- ated with the applied constant magnetic field B applied perpendicular to the plane of the nanoring. The AB magnetization of the graphene ring is given by M (Φ) = −S dEtot dΦ , where the total energy Etot(Φ) = occ(cid:88) i,σ εi(Φ) (3) (4) 3 FIG. 1. (Color online) (a) Part of the hexagonal graphene ring (2718 carbon atoms) with armchair edges. 1 and 0 denote the short and long carbon dimers, respectively. Lengths in units of the graphene lattice constant a0 = 0.246 nm. (b) TB single-particle spectrum as a function of the magnetic flux (magnetic field). Energies in units of the TB hopping-parameter t = 2.7 eV. The dashed black line denotes the Fermi level for N = 14 electrons. The arrows highlight the band gaps. (c) Magnification of the TB spectrum around the δ1 band gap. δ1 ∼ 80 K, and thus it is easily detectable experimentally. the three cases studied here. These differences fall into two categories, namely, (A) same edge termination but different shape and (B) same shape but different edge termination. A. Hexagonal versus trigonal armchair rings The shape of the hexagonal graphene ring with arm- chair edge terminations considered here, as well as the corresponding TB results regarding the single-particle spectrum, the total energy, and the magnetization are displayed in Fig. 1 and Fig. 2, as a function of Φ/Φ0. The shape of the trigonal graphene ring with armchair edge terminations considered here, as well as the correspond- ing TB results regarding the single-particle spectrum are displayed in Fig. 3. Both TB energy spectra in Fig. 1(b) (armchair hexagon) and Fig. 3(c) (armchair triangle) are organized in braid bands separated by energy gaps. They exhibit, however, two main differences. The first concerns the composition of the braid bands, with the hexagonal ring exhibiting six-membered bands while the trigonal ring having three-membered bands. The sixfold and threefold groupings are a reflection of the Z6 and Z3 point-group symmetry of the hexagonal and trigonal shapes, respec- tively; these symmetries are fully taken into account by the DKP modeling in Sec. III. The second important difference between the TB en- ergy spectra in Fig. 1(b) and Fig. 3(c) concerns the num- ber and nature of energy gaps. Specifically, two regular superlattice gaps δ1 and δ2 are present in both cases; note the similar energy scale [a magnification of the re- gion around the δ1 gap is shown in Fig. 1(c)]. However, while a mass gap ∆0 is well developed for the trigonal ring [Fig. 3(c)], no corresponding ∆0 gap is present in the spectrum of the hexagonal ring, where the ε = 0 FIG. 2. (Color online) (a) TB total energies (sum over single- particle energies including spin) for N = 14 quasiparticles. (b) Correposnding TB magnetization (in units of the Bohr magneton). Energies in units of the TB hopping-parameter t = 2.7 eV. is given by the sum over all occupied single-particle en- ergies; the index σ runs over spins. Φ = BS is the mag- netic flux through the area S of the graphene ring and Φ0 = hc/e is the flux quantum. The diagonalization of the TB hamiltonian [Eq. (1)] is implemented with the use of the sparse-matrix solver ARPACK.46 In calculating Etot [see Eq. (4)], only the single-particle TB energies with εi(B) > 0 are considered.24 The TB results exhibit significant differences between 4 FIG. 3. (Color online) (a) Part of the trigonal graphene ring (2142 carbon atoms) with armchair edges. (b) Magnification of the corner of the trigonal ring shown in (a). 1 and 0 denote the short and long carbon dimers, respectively. Lengths in units of the graphene lattice constant a0 = 0.246 nm. (c) TB single-particle spectrum as a function of the magnetic flux (magnetic field). Energies in units of the TB hopping- parameter t = 2.7 eV. The two arrows denoted δ1 and δ2 highlight band gaps. The arrow denoted by ∆0 indicates the opening of a gap at the Fermi level (ε = 0) associated with generation of a rest mass M. Note that the ∆0 gap is absent in the TB spectrum of the hexagonal graphene ring in Fig. 1(b). horizontal axis dissects (splits in half) the corresponding sixfold braid band [Fig. 1(b)]. As we will show in Sec. IV, the gap ∆0, for the case of the armchair triangle, is con- sistent with the physics of a massive (but still relativistic) Dirac fermion, while the dissecting of the ε = 0 sixfold band, in the case of the armchair hexagon, is consistent with the formation of a fermionic soliton37 built on a scalar Z2 kink soliton (precisely, a train of six fermionic solitons attached to successive Z2 kink/antikink solitons; see also Sec. V). B. Armchair versus zigzag hexagonal rings As aforementioned, the TB results exhibit also signifi- cant differences between the cases of hexagonal rings with FIG. 4. (Color online) (a) Part of the hexagonal graphene ring (2688 carbon atoms) with zigzag edges. 1 and 0 denote the short and long carbon dimers, respectively. Lengths in units of the graphene lattice constant a0 = 0.246 nm. (b) TB single- particle spectrum as a function of the magnetic flux (magnetic field). Energies in units of the TB hopping-parameter t = 2.7 eV. The dashed black line denotes the Fermi level for N = 20 quasiparticles. The arrows highlight the band gaps. armchair and zigzag terminations. One such difference concerns the B-dependence of the single-particle energies ε(Φ), which is piecewise linear for the armchair case [Fig. 1(b)], but piecewise parabolic for the zigzag case [Fig. 4(b)]; this maintains also in the total energies Etot(Φ) [Fig. 2(a) and Fig. 5(a)]. For the AB magnetizations [see Eq. (3)], this results in a characteristically different pro- file for the AB oscillations of M (Φ): step-staggeredlike in the armchair case [Fig. 2(b)] and sawtoothlike [Fig. 5(b)] in the zigzag case. The paraboliclike B-dependence in the zigzag edge case15 is reminiscent of the spectra of a nonrelativistic ideal metal ring.47 In contrast (see be- low), the linear B-dependence (in conjunction with the other features of the hexagonal armchair spectrum) can be associated with the fully relativistic regime of Dirac 5 corners of the polygons providing a generalized analog to the "scatterers" of Ref. 18. Specifically, we consider a 1D relativistic Dirac-Kronig-Penney model with unit cells built out of square potential barriers (developed in Ref. 2 in the context of the physics of quarks). The DKP model considered here is based on the 1D generalized Dirac equation, which has the form: [E − V (x)]IΨ + i¯hvF α ∂Ψ ∂x − βφ(x)Ψ = 0, (5) with vF being the Fermi velocity of graphene, which re- places the speed of light c; vF /c ≈ 1/300. V (x) is an electrostatic potential and φ(x) is a bosonic position- dependent scalar field. We note that Ref. 2 uses φ(x) ≡ Mv2 F + Vs(x), with M denoting the rest mass of a Dirac fermion (including the massless case) and the term Vs(x) being referred to as the Lorentz scalar potential. Omit- ting the last term on the left of Eq. (5) reduces this equa- tion to the massless Dirac-Weyl8 one that underlies the majority of studies in planar graphene. The fermion field Ψ is a twodimensional vector Ψ = , (6) (cid:19) (cid:18) ψu ψl where the subscripts u and l stand for the upper and lower component, respectively. The 2 × 2 Dirac matrices α and β can be49 any two of the three Pauli matrices (cid:19) (cid:18) 0 1 1 0 (cid:18) 0 −i (cid:19) i 0 (cid:19) (cid:18) 1 0 0 −1 σ1 = ; σ2 = ; σ3 = . (7) For example in the Dirac representation, one has αD = σ1 and βD = σ3. I is the 2 × 2 identity matrix. We stress that the energy spectra of the DKP model are in- dependent of the specific representation used for α and β. Below we will often use the notation m(x), instead of φ(x), to stress the fact that φ(x) can be considered as a position-dependent mass term. i i i , m(n) i In the DKP modeling of the TB results, we assign to the nth side (n = 1, . . . , Ns) of the polygonal ring a number (J) of square potential steps (regions) denoted as (V (n) ), i = 1, . . . , J. We note again that the electrostatic potentials V (n) enter the 1D Dirac equation [Eq. (5] through the energy term (E − V )IΨ, while the mass terms m(n) replace the scalar potential in the term βφ(x)Ψ; as a result these two potentials lead to different physical behavior in the relativistic regime. The building block of the DKP model is a 2×2 wave- function matrix Ω formed by the components of two inde- pendent 2×1 spinor solutions of the onedimensional first- order Dirac equation. Ω plays2 the role of the Wronskian matrix W50 used in the second-order nonrelativistic KP model; it is defined as follows at a point x of the unit cell (here we use the Dirac representation): e−iKx ΛeiKx −Λe−iKx (cid:18) eiKx ΩK(x) = (cid:19) (8) , FIG. 5. (Color online) (a) TB total energies (sum over single- particle energies including spin) for N = 20 quasiparticles. (b) Correposnding TB magnetization (in units of the Bohr magneton). Energies in units of the TB hopping-parameter t = 2.7 eV. fermions with position-dependent masses.19,20 Both the armchair and zigzag single-particle spectra for hexagonal are organized in six-member braid bands separated by energy gaps. This sixfold grouping is a re- flection of the Z6 point-group symmetry of the hexagonal rings. The energy gaps in the zigzag case are compara- ble to the width of the braid bands, and both the gaps and the widths of the bands increase with higher energy [see Fig. 4(b)]; this is consistent with a nonrelativistic Kronig-Penney model.14 In Fig. 4(b), there are three en- ergy gaps labeled as δ0, δ1, and δ2. Unlike the relativistic regime, in the nonrelativistic limit a gap around ε = 0 is unrelated to the particle mass, and for this reason we use the symbol δ0 (with a lower-case δ) instead of ∆0 as was the case in Fig. 3(c). We note that in the nonrelativistic regime the effective mass of the quasiparticle excitations is proportional to the inverse of the second derivative of the (approximately) parabolic spectra; see, e.g., Ref. 48. III. DIRAC-KRONIG-PENNEY SUPERLATTICE As was shown for a semiconductor ring using a non- relativistic superlattice approach, the AB single-particle spectrum exhibits energy gaps demarcating Bloch bands when a scatterer is placed on the ring.18 In this con- text, the energy gaps that appear in Fig. 1(b), Fig. 3(c), and Fig. 4(b) indicate that the AB effect in polygonal graphene rings should be analyzed and modeled with the help of 1D Kronig-Penney-type superlattices, with the 6 FIG. 6. (Color online) Schematic representation of the (V (n) ), i = 1, . . . , 3 and n = 1, . . . , Ns square-step parameters entering in the DKP calculations: (a) in Sec. IV A. (b) in Sec. IV B. (c) in Sec. IV C. Each side of the polygon is divided in three lengths L1, L2, and L3. Boxes outside a polygon (online yellow) represent values m0 > 0. Boxes inside a polygon (online red) represent values −m0 < 0. The nonzero values for the V (n) in (c) are portrayed by thick lines (online brown) in the interior of the schematic hexagon. Zero values of parameters are not highlighted. and V (n) 1 , m(n) i 3 i where K 2 = (E − V )2 − m2v4 F ¯h2v2 F , Λ = ¯hvF K E − V + mv2 F . (9) We note again that, unlike the case of the original Dirac equation,1 m here is not a constant, but it may take dif- ferent values from one region to the next. The trans- fer matrix for a given region (extending between two matching points x1 and x2) is the product MK(x1, x2) = ΩK(x2)Ω−1 K (x1); this latter matrix depends only on the width x2 − x1 of the region, and not separately on x1 or x2. The relativistic M matrices defined here correspond to those considered51 in the case of a nonrelativistic su- perlattice in Ref. 14. The transfer matrix corresponding to the nth side of the hexagon is the product tn = MK(xi, xi+1), x1 = 0, xJ+1 = L, (10) (cid:89) i=1,J Ns(cid:89) with L being the (common) length on the hexagon side. The transfer matrix associated with the complete unit cell (around the ring) is the product T = tn, (11) n=1 where Ns is the number of sides of the polygonal shape considered (Ns = 3 and Ns = 6 for a triangle and hexagon, respectively). Following Ref. 18, we consider the supperlattice gen- erated from the virtual periodic translation of the unit cell as a result of the application of a magnetic field B perpendicular to the ring. Then the AB energy spectra are given as solutions of the dispersion equation where we have explicitly denoted the dependence of the r.h.s. on the energy E. The presence (η = 1/2) or ab- sence (η = 0) of an additional flux-shift in the relativis- tic or nonrelativistic case, respectively, follows through a comparison of the patterns of AB oscillations of a Dirac/Schrodinger electron in the limiting case of an ideal metallic circular ring.47,52 IV. DKP INTERPRETATION OF TIGHT-BINDING CASE STUDIES In this section, we demonstrate that our DKP model- ing can capture the essential physics underlying the TB spectra in Fig. 1(b), Fig. 3(c), and Fig. 4(b). In this re- spect, it also provides a framework for unifying the broad variety of behaviors of the TB spectra of graphene nanor- ings. A schematic representation of the parameter sets used in our DKP simulations is given in Fig. 6. A. Armchair hexagonal ring and position-dependent-mass relativistic regime First we attempt a solution corresponding to the gen- eralized Dirac equation (5) with φ(x) = 0. In Fig. 7(a), we display the DKP spectra of a massless excitation (i.e., m(n) i = 0 for all i and n = 1, . . . , 6). This massless DKP spectrum does not exhibit any gaps and it is strictly lin- ear and periodic (with period Φ0) as a function of Φ; this correlates with the behavior of a free massless fermion, as in the case of 2D graphene. We note that this gapless spectrum remains unchanged even when we consider in addition electrostatic potential steps, V (n) > 0, a fact i that is a reflection of Klein tunneling.53,54 cos [2π(Φ/Φ0 + η)] = Tr[T(E)]/2, (12) However, the TB spectrum in Fig. 1(b) exhibits energy 7 FIG. 7. (Color online) Spectra from the DKP model (relativistic regime) corresponding to the schematic case (hexagon) in Fig. 6(a); see text and Eqs. (13-15) for the full set of parameters employed. (a) m0 = 0 and any Vi. Note the absence of band gaps due to the Klein paradox. (b) m0 = 0.01t/v2 F . Note the similarity of the spectrum with that of the armchair graphene ring in Fig. 1(b). (c) m0 = 0.30t/v2 F . The horizontal lines result from suppression of the AB oscillations due to strong localization. The dashed line indicates the energy zero. gaps (denoted as δ1 and δ2, and highlighted by arrows), which require consideration of potential barriers in the DKP modeling. In the spirit of earlier investigations of real-space superlattices in 2D graphene,55 -- 57 we consider i = M > 0, and alternating first a constant mass m(n) V (n) i = V0 > 0 and V (n) i = 0 steps in consecutive regions (see Sec. III). However, calculations with this choice show an opening of an energy gap at ε = 0, a fact that con- flicts with Fig. 1(b); in addition, it does not preserve the particle-hole symmetry of the TB spectra. A crucial feature of the TB armchair spectra in Fig. 1(b) is the presence of zero-energy states (at half-integer fluxes). In order to capture this feature, and in light of our failed choices (see above), we attempt next to use a non-vanishing position-dependent scalar field φ(x) [de- noted also as m(x)] in Eq. (5), and set the electrostatic potential V (x) = 0. Recalling certain key elements in the theory of trans-polyacetylene pertaining to zero-energy solitonic modes,19,20 we employ in our DKP transfer- matrix solution of Eq. (5) a scalar potential φ(x) of the form m(x) = −m(−x). Consequently, we divide each side of the hexagon in three parts (J = 3) of length 1 = a, L(n) L(n) 2 = b, L(n) 3 = a, and assign values 1 = V (n) V (n) 2 = V (n) 3 = 0, and m(n) 1 = m(n) 3 = 0, m(n) 2 = (−1)nm0. (13) (14) (15) Note that the index n = 1, . . . , 6 here is numbering the sides of the hexagon, and thus overall the position- dependent mass term in our model is antisymmetric around each corner of the haxagonal ring. A schematic representation of the above parameters [Eq. (13)− Eq. (15)] is given in Fig. 6(a). Fig. 7(b) displays the DKP spectrum calculated with the dispersion equation (12) using the above parame- F , and a = 8a0, b = 15a0.58 ter set with m0 = 0.01t/v2 One observes that, in addition to the piecewise linear B-dependence, the DKP spectrum faithfully reproduces the two other central features of the TB spectrum [Fig. 1(b)]: (i) the zero-energy states at half-integer values of Φ/Φ0 and (ii) the opening at higher energies of energy gaps demarcating emerging sixfold braid bands. The behavior of each arm of the armchair hexagonal ring as a domain similar to the dimerized domains of the trans-polyacetylene has a deeper physical reason, which can be revealed if one considers each arm of the hexagon as a perturbed armchair graphene nanoribbon. Indeed analytic expressions for the energy dispersion of the aG- NRs have been recently derived [see Refs. 32(b) and 35]; they have the form (16) E(k) = ±t1eika + t2e−ikb, √ 3) and b = a0/ √ with k being the wave vector along the direction of the edge. a = a0/(2 3, with a0 = 0.246 nm being the lattice constant of graphene. t1 = −2t cos[pπ/(NW + 1)] + δt1, p = 1, 2, . . . ,NW and t2 = −t + δt2, with δt1, δt2 denoting the perturbation away from a perfect aGNR. The dispersion equation (16) is similar to the tight-binding one describing a onedimen- sional chain of carbon atoms with bonds (hopping ma- trix elements) of alternating strength t1 and t2 (Kekul´e structure). The spectrum E(k) exhibits a mass gap ∆0 = t1 − t2 at k = π/(a + b). This behavior is analogous to that of the linear-chain lattice TB model for trans-polyacetylene; see Eq. (2.1) in Ref. 21. In particular, when t1 (cid:54)= t2, Eq. (16) describes a single dimerized domain breaking the 1D reflectional symme- try; when t1 = t2 (metallic aGNR), it describes a sym- metric chain of atoms, which preserves the reflectional symmetry. We note that the factor underlying the for- mation of dimerized domains in trans-polyacetylene is the Peierls instability incorporated in the Su-Schrieffer- Heeger model.20,21 The corresponding factor in armchair graphene nanoribbons and rings is topological in nature, i.e., it is a reflection of the lattice distortions of graphene due to the edge termination and the shape. Further insight can be gained through the observa- tion that the armchair edge by itself reflects the car- bon dimerization. Indeed it exhibits shorter dimers (denoted by 1) alternating with longer ones (denoted by 0); see Fig. 1(a). Thus each arm of the graphene ring corresponds to one of two equivalent domains . . . , 1, 0, 1, 0, . . . or . . . , 0, 1, 0, 1, . . .. Following this no- tation and going around a given corner, one gets symbol- ically . . . , 0, 1, 0, 1, 1, 0, 1, 0, . . ., i.e., each corner (denoted by an underline) acts as a domain wall separating two alternative domains. In an infinite polyacetylene chain and at the position of the domain wall, a strongly localized fermionic soli- ton develops having a fractional charge ±1/2.19,20 How- ever, the graphene nanoring is a finite system and overall it can carry only integer charges.59,60 Furthermore, the TB results, and their DKP analog in Fig. 7(b), indicate a moderate extent of localization at the corners, result- ing from a non-negligible tunneling between the corners. Strong solitonlike localization (exhibiting a 1/6 fractional charge at each corner) can be achieved for larger values of m0. Indeed a large m0 can localize a massless fermion, as is known from the (theoretical) trapping methodology in 2D graphene referred to as the infinite-mass bound- ary condition (introduced in Ref. 61 in the context of trapping neutrinos), as well as from the localization of massless quarks discussed in Ref. 2. The DKP spec- trum for a large value m0 = 0.30t/v2 F is displayed in Fig. 7(c). It is seen that now the single-particle energies falling within the gap −0.3t < E < 0.3t form horizontal straight lines because the corresponding AB oscillations have been suppressed due to vanishing of the tunneling between the fermionic solitons at the corners; because of the localization no magnetic flux is trapped by the wave function on the hexagonal ring. It is of interest to note that such a train configuration of fermionic solitons in an hexagonal ring may be referred to as a fractional Wigner crystallite. We note that besides the zero-energy fermionic soliton discussed in the context of the states of polyacetylene,19,20 Fig. 7(c) indicates the emergence of two polaroniclike states22,62 with energies ≈ ±0.16t falling within the gap. B. Armchair trigonal ring and constant-mass relativistic regime The results of our DKP calculations [associated with Eq. (5)] for the armchair trigonal ring are shown in Fig. 8 8. Fig. 8(a) displays the DKP spectra of massless excita- tions (i.e., m(n) i = 0 for all i and n = 1, 2, 3). As was the case with the hexagonal ring [Fig. 7(a)], the massless DKP spectrum in Fig. 8(a) does not exhibit any energy gaps and it is strictly linear and periodic (with period Φ0) as a function of Φ (or equivalently the magnetic field B). We again note that this gapless spectrum remains unchanged even when we consider in addition electro- static potential steps, V (n) i > 0, a fact that is a reflection of Klein tunneling.53,54 However, the TB spectrum in Fig. 3(c) exhibits en- ergy gaps (denoted as ∆0, δ1 and δ2, and highlighted by arrows), which require inclusion of potential barriers in the DKP modeling. Following the analogy with the trans-polyacetylene, it is apparent that the opening of the ∆0 gap indicates the absence of domain alternation. Namely, the armchair trigonal ring represents a realiza- tion of a single domain extending along the full length of the triangle. Thus the corners of the triangle act a scatterers instead of domain walls as in the case of the hexagonal ring (see Sec. IV A). Indeed, following the no- tation of 1 and 0 introduced above for the short and long dimers [Fig. 1(a)], and going around the corner of the trigonal ring in Fig. 3(a) [or Fig. 3(b)], one gets the se- quence . . . , 0, 0, 0, 1, 1, 1, 0, 0, 0, . . ., which is in agreement with the presence of the same domain on both sides of the π/3 corner. Consequently, in the DKP modeling we keep the same parametrization as in Eqs. (13) and (14), but we replace Eq. (15) by m(n) 2 = m0 = M, n = 1, 2, 3, (17) 1 = m(n) 3 = 0, m(n) that is, the mass parameters are the same (not alternat- ing) on the three arms of the trigonal ring [see schematic representation in Fig. 6(b)]; recall that M denotes the fermion rest mass [see discussion below Eq. (5)]. Fig. 8(b) [see also a magnification in Fig. 8(c)] dis- plays the DKP spectrum calculated from the dispersion equation (12) using the above parametrization [see Eqs. (13), (14), and (17)] with M = 0.02t/v2 F , and a = 19a0, b = 10a0.58 One sees that the DKP spectrum reproduces the two central features of the TB spectrum in Fig. 3(c): (i) the gap ∆0 around ε = 0 and (ii) the threefold braid bands which are separated by the gaps δ1 and δ2 at higher energies. C. Zigzag hexagonal ring and constant-mass nonrelativistic limit As aforementioned, the TB spectrum of the hexagonal graphene ring with zigzag edges [Fig. 4(b)] shows trends associated with a large-mass fermion in the nonrelativis- tic regime: (i) almost-perfect parabolic B-dependence and (ii) energy gaps δ0, δ1 and δ2 that are as large as the width of the braid bands. We remind that, for a massless excitation (ultrarelativistic limit), the spectra 9 FIG. 8. (Color online) Spectra from the DKP model (relativistic regime) corresponding to the schematic case (triangle) in Fig. 6(b); see text and Eq. (17) for the full set of parameters employed. Panels (a) and (b,c) correspond to different choices of m0 [denoted also as M here; see Eq. (17)]. (a) m0 = 0 and any Vi. Note the absence of band gaps due to the Klein paradox. (b,c) m0 = 0.02t/v2 F , and in (c) we display a magnification of the region around E = 0. Note the similarity of the spectrum with that of the armchair graphene ring in Fig. 3(c). The dashed line indicates the energy zero. going around the 2π/3 corner of the zigzag ring in Fig. 4(a), one further gets the sequence . . . , 0, 0, 0, 1, 0, 0, 0, . . . This, in addition to the anticipation of a nonrelativistic regime, suggests that the entire hexagonal zigzag ring constitutes a single domain with embedded (corner) im- purities (electrostatic potential scatterers). In light of the above, using a large rest mass M in Eq. (9) [nonrelativistic limit, see Eqs. (4.27) and (4.28) in Ref. 2], we were able to reproduce in the DKP model [see Fig. 9] the overall trends of the TB spectrum [Fig. 4(b)] for the zigzag ring. The DKP parameters used58 were as follows, Lengths: FIG. 9. (Color online) Spectra from the DKP model (nonrela- tivistic regime) corresponding to the schematic case (hexagon) in Fig. 6(c); see text and Eqs. (18-20) for the full set of pa- rameters employed. The energies do not contain the rest- F , where E is the value obtained from solution of the DKP model and M mass contribution,63 that is, (cid:101)E(t) = E − Mv2 is determined by fitting (cid:101)E to the TB results [Fig. 4(b)]. The rest mass determined to yield the best fit shown here, is given by M = 42.06t/v2 F . This mass is heavier than that of the electron (2.10t/v2 F ), suggesting an analogy with leptons. are strictly linear and exhibit no gaps [see Figs. 7(a) and 8(a)]; consequently small or moderate mass terms (rel- ativistic regime) will result in smaller energy gaps com- pared to the width of the braid bands [see Fig. 7(b) and Figs. 8(b) and 8(c)]. The sharply different physics (nonrelativistic versus relativistic behavior) underlying the TB spectra of the zigzag and armchair rings originates from the different edge topology. Following the notation of 1 and 0 intro- duced above for the short and long dimers [Fig. 1(a)], and L(n) 1 = L(n) 3 = a = 1.5a0, L(n) 2 = b = 28a0. Rest mass: M = m(n) 1 = m(n) 2 = m(n) 3 = 42.06t/v2 F . Potential barrier step at each corner: V = V (n) 1 = V (n) 3 = 80 × 10−5t, with V (n) 2 = 0. (18) (19) (20) In the evaluation of the nonrelativistic limit of the A schematic representation of these parameters is given in Fig. 6(c). DKP model, one often sets E = Mv2 itive energies [see Eq. (4.27) in Ref. 2]. The quantity calculated from the DKP model in this limit is E, while in comparing with the TB results [Fig. 4(b)] we plot in F + (cid:101)E for the pos- Fig. 9 (cid:101)E vs. Φ. The value of M is determined by finding the best fit to the TB results.63 It is expected that fur- ther improvement in the agreement between the TB and DKP approaches can be achieved by employing smooth (rather than square-shaped) profiles for the potential bar- riers. The constant mass, M, found by us here is twenty times larger than the rest mass of the electron, indicat- ing an analogy with electronlike leptons,23 rather than the electron itself. We note that the large rest mass M found in this sec- tion is unrelated to the energy gap δ0 [see Fig. 4(b)]; F /δ0 ∼ 105). Indeed in the nonrelativistic regime, (Mv2 the effective mass is related to the second derivative of the band spectra.48 Our findings concerning the zigzag graphene ring are in agreement with the analysis of Ref. 48 regarding the TB spectra of (infinite) symmetric poly- acene, which is a single chain of fused benzene rings and can be considered as the thinnest possible zigzag graphene nanoribbon. V. THE UNDERLYING RELATIVISTIC QUANTUM FIELD LAGRANGIAN In the above (see Sec. II), we identified characteris- tic patterns of tight-binding spectra in planar graphene nanorings and we described their dependence on the edge termination (armchair versus zigzag) and the ring shape (hexagonal versus trigonal). Subsequently, we presented a unified interpretation of the TB results using a Dirac- Kronig-Penney model (Sec. III), built upon the gener- alized 1D Dirac equation [Eq. (5)]. A central finding of our relativistic DKP analysis were the close analogies (found in Secs. IV A and IV B) between the behavior of armchair hexagonal and trigonal graphene nanorings and the physics of trans-polyacetylene.19 -- 21 A natural next step towards a deeper understanding of the connection of our findings to relativistic quan- tum field theory is the elucidation of the underlying La- grangian formalism. Motivated by the theory of trans- polyacetylene,20 we write a total Lagrangian density L = Lf + Lφ, (21) which is the sum of (partial) Lagrangian densities for the fermion field Ψ [Eq. (6)] and the bosonic scalar field φ(x). We note that the scalar field φ(x) was denoted earlier also as m(x) and was referred to as a position-dependent mass; see Sec. III and Sec. IV A. The stationary generalized Dirac equation [Eq. (5)] can be derived from the fermionic Lagrangian density Ψ − φΨ†βΨ, Ψ − i¯hvF Ψ†α Lf = −i¯hΨ† ∂ ∂t ∂ ∂x (22) where we have neglected contributions from the electro- static potentials (see discussion in Sec. IV A). In Eq. (22), the last term LY = −φΨ†βΨ, (23) which depends on both the fermion Ψ and scalar φ fields, has the form of a Yukawa coupling. LY is the potential agent for rest-mass acquisition by the originally massless fermion [described by the first two terms in the right hand side of Eq. (22)]. The Yukawa interaction is also used in the Standard Model64 -- 66 to describe the coupling between the Higgs field and the massless quark and lepton fields (i.e., the 10 fundamental fermion particles). Through spontaneous symmetry breaking67 -- 70 of the Higgs field [which is a complex SU (2) doublet of four real scalar fields φ], these fermions acquire a mass proportional to the vacuum ex- pectation value of the Higgs field.71 -- 73 The essential observation that we make here is that, although, due to the 1D character of the graphene rings, the 1D Lagrangian in Eq. (21) does not possess the full richness of the Lagrangian of the Higgs sector in the Standard Model, both share the central aspect of sym- metry breaking and mass acquisition by a fermion via a Yukawatype interaction. We turn next to the task of constructing the La- grangian part Lφ for the scalar field φ, which is of the general form74,75 (in 1 + 1 dimensions, i.e., time plus one space dimension) Lφ = − 1 2 ( ∂φ ∂x )2 − V (φ), (24) and which preserves the reflectional Z2 symmetry, i.e., it is invariant under φ → −φ. Note that we are interested in the adiabatic approximation, and thus we omit the time dependent terms in Lφ. The emergence of a constant mass for the armchair trigonal ring (Sec. IV B) could simply be accounted for by considering a constant value of φ = φ0 = M in the fermion Lagrangian Lf ; then one needs to pay no further consideration to the bosonic Lφ. However, a more gen- eral position-dependent field, φ(x) = m(x), was found to be essential in our DKP-model analysis of the armchair hexagonal ring (Sec. IV A). In this case, φ(x) alternates between two unequal values ±φ0, with φ0 = m0 [see Fig. 6(a)]. This indicates breaking of the Z2 symmetry of the solutions to the equation of motion derived from the Lagrangian in Eq. (24). An expression for V (φ) which re- produces qualitatively the above behavior (including the trigonal ring case) is the socalled φ4, which corresponds to a quartic double well potential in φ, i.e., V (φ) = (φ2 − ζ 2)2, ξ 4 (25) where ξ and ζ are parameters. With the potential in Eq. (25), the bosonic sector has the field equation (see Ch. 2.3 in Ref. 74 and Ch. 1.1 in Ref. 75) − ∂2φ ∂x2 + ξ(φ2 − ζ 2)φ = 0. (26) Two solutions of Eq. (26) are φ(x) = ±φ0 = ±ζ; these solutions break the symmetry since φ0 (cid:54)= 0. Using these solutions in the Dirac Eq. (5), one obtains the standard constant-mass Dirac equation1 for the fermionic field. This case corresponds to the behavior of the armchair trigonal ring (Sec. IV B), as well as to that of the zigzag hexagonal ring (Sec. IV C). In addition, however, Eq. (26) has nonlinear solutions that interpolate between the locations φ0 and −φ0 of the VI. CONCLUSIONS 11 The paper investigated the different behavior of the Aharonov-Bohm spectra and magnetic-field induced os- cillations for three characteristic cases of planar graphene nanorings, i.e., an hexagonal ring with armchair edge terminations, a trigonal ring with armchair edge termi- nations, and an hexagonal ring with zigzag edge termi- nations. The tight-binding results (Sec. II) were an- alyzed with the help of a 1D relativistic Dirac-Kronig- Penney model2 (Sec. III), which accounts for the vir- tual superlattice associated18 with the applied magnetic field. This analysis revealed unexpected topological ef- fects and condensed-matter analogies with elementary particle physics. In particular, the behavior found by us for the armchair hexagonal ring (Sec. IV A) is reminiscent of the extreme relativistic regime describing zero-energy fermionic soli- tons with fractional charge in quantum field theory37 and in the theory of trans-polyacetylene.19 -- 22 This regime re- sults from a consideration of a modified (generalized) Dirac equation with a position-dependent mass term (or equivalently a position-dependent scalar bosonic field). In contrast, the quasiparticle excitations in the arm- chair trigonal ring (Sec. IV B) behave as relativistic Dirac fermions having a constant mass. A unification of these two dissimilar behaviors was presented in Sec. V by in- troducing the underlying relativistic Lagrangian formal- ism for a fermionic and a scalar bosonic fields coupled via a Yukawa interaction. The Yukawa term in conjunc- tion with the breaking of the Z2 reflectional symmetry of the scalar field may result in two outcomes, i.e., for- mation of a fermionic soliton (armchair hexagonal ring) or mass generation (armchair trigonal ring). The pro- foundly differing behaviors found by us for the armchair hexagonal and trigonal rings (with both sharing similar spatial dimensions), are manifestations of the quantum topological nature of this behavior, as distinguished from "quantum size effects" which are length-scale dependent phenomena, originating from spatial confinement of the electrons (quasiparticles, in general).76 -- 79 The behavior of the zigzag hexagonal ring resembles the low-kinetic-energy nonrelativistic regime of a lepton- like fermion having a rest mass larger than that of the electron (Sec. IV C). This behavior contrasts with the relativistic ones found for the aforementioned armchair rings, thus highlighting the compounded topological and edge -termination effects. These findings80,81 highlight the potential of graphene nanosystems for providing a bridge between condensed- matter and particle physics, well beyond the paradigm of the massless neutrinolike fermion familiar from the 2D graphene sheet. Furthermore beyond the realm of graphene proper, where atomically precise narrow nanoribbons have already been synthesized,36 we antic- ipate that our theoretical predictions could be tested using an ever expanding class of designer-Dirac-fermion manmade systems, such as optical lattices comprising ul- FIG. 10. (Color online) Dashed line: The scalar Z2 kink soliton [Eq. (27)]. Solid line: The particle density (unnor- malized) of the corresponding enslaved fermionic soliton [Eq. (29)]. The values ξ = 1 and ζ = 1 were used. The domain wall is located at x = 0. two minima of the V (φ) potential. One of these nonlinear solutions is called the Z2 kink soliton and the other the Z2 antikink soliton. The kink soliton is given by φk(x) = ζ tanh ζx , (27) (cid:33) (cid:32)(cid:114) ξ 2 and the antikink soliton has the form ¯φk(x) = −φk(x). (28) We note that φk(±∞) = ±ζ and ¯φk(±∞) = ∓ζ, while φk(0) = ¯φk(0) = 0. Using these kink or antikink scalar fields, the corre- sponding generalized Dirac equation [Eq. (5)] possesses fermionic soliton20 solutions of the form (here α = σ2, β = σ1) (cid:18) exp(cid:0)−(cid:82) x 0 φk(x(cid:48))dx(cid:48)(cid:1) (cid:19) 0 ΨS(x) ∝ . (29) The importance of these fermionic solitons lies in the fact that they have strictly zero energies;20 thus they fall into the particle/antiparticle (valence/conductance band) energy gap. Furthermore, they are localized at x = 0, which is the domain wall between x > 0 (φ0) and x < 0 (−φ0). The enslavement of the fermionic soliton ΨS(x) by the scalar potential of the kink soliton φk(x) is evident from Eq. (29). This is also reflected by the localization of ΨS(x) on the domain wall x = 0 (see Fig. 10). It is apparent that such zero-mode fermionic solitons ΨS(x) underlie qualitatively the behavior of the fermion excitations in the armchair hexagonal graphene nanoring (Sec. IV A), with the corners of the ring behaving as do- main walls and the m(x) stepwise function in the DKP model [φ(x) in Eq. (5)] mimicking an alternation of Z2 kink and antikink scalar solitons [Eqs. (27) and (28)]; for a quantum-field-theory description of a train of alternat- ing kinks and antikinks, see Ch. 1.7 in Ref. 75. tracold atoms,10,13 or "molecular"12 and nanopatterned artificial graphene.44 86ER45234. ACKNOWLEDGMENTS This work was supported by the Office of Basic En- ergy Sciences of the US D.O.E. under contract FG05- 12 ∗ Igor.Romanovsky@physics.gatech.edu † Constantine.Yannouleas@physics.gatech.edu ‡ Uzi.Landman@physics.gatech.edu 1 P. A. M. Dirac, Proc. Roy. Soc. London, Ser. A 117, 610 (1928). 235404 (2007). 25 D. A. Bahamon, A. L. C. Pereira, and P. A. Schulz, Phys. Rev. B 79, 125414 (2009). 26 T. Luo, A. Iyengar, A. H. Fertig, and L. Brey, Phys. Rev. B 80, 165310 (2009). 2 B. H. J. McKellar and G. J. Stephenson, Jr., Phys. Rev. 27 H. A. Fertig and L. Brey, Phil. Trans. R. Soc. A 368, 5483 C 35, 2262 (1987). (2010). 3 J. D. Bjorken and S. D. Drell, Relativistic quantum me- 28 M. M. Ma and J. W. Ding, Solid State Commun. 150, 1196 chanics (New York, McGraw-Hill, 1964). (2010). 4 D. Griffiths, Introduction to Elementary Particles (Wiley- 29 D. S. L. Abergel, V. M. Apalkov, and T. Chakraborty, VCH, Weinheim, 2008). 5 K. S. Novoselov, A. K. Geim, S. V. Morozov, D. Jiang, Y. Zhang, S. V. Dubonos, I. V. Grigorieva, and A. A. Firsov, Science 306, 666 (2004). 6 P. R. Wallace, Phys. Rev. 71, 622 (1947). 7 A. H. Castro Neto, F. Guinea, N. M. R. Peres, K. S. Novoselov, and A. K. Geim, Rev. Mod. Phys. 81, 109 (2009). 8 H. Weyl, Z. Phys. 56, 330 (1929) [English translation: Sur- veys High Energ. Phys. 5, 261 (1986)]. 9 S.-L. Zhu, B. Wang, and L.-M. Duan, Phys. Rev. Lett. 98, 260402 (2007). 10 J. Ruostekoski, J. Javanainen, and G. V. Dunne, Phys. Rev. A 77, 013603 (2008). 11 Y.L. Chen et al., Science 329, 659 (2010). 12 K. K. Gomes, W. Mar, W. Ko, F. Guinea, H. C. Manoha- ran, Nature 483, 306 (2012). 13 D.-W. Zhang, L.-B. Shao, Z.-Y. Xue, H. Yan, Z. D. Wang, S.-L. Zhu, arXiv:1206.5614. 14 I. Romanovsky, C. Yannouleas, and U. Landman, Phys. Rev. B 86, 165440 (2012). Phys. Rev. B 78, 193405 (2008). 30 M. Zarenia, J. M. Pereira, A. Chaves, F. M. Peeters, and G. A. Farias, Phys. Rev. B 81, 045431 (2010). 31 J. Schelter, P. Recher, and B. Trauzettel, Sol. State Com- mun. 152, 1411 (2012). 32 For TB and related studies, see for example (a) K. Nakada, M. Fujita, G. Dresselhaus, and M. S. Dresselhaus, Phys. Rev. B 54, 17954 (1996); (b) H. Zheng, Z. F. Wang, T. Luo,Q. W. Shi, and J. Chen, Phys. Rev. B 75, 165414 (2007). (c) Alexander Onipko, Phys. Rev. B 78, 245412 (2008). 33 For density-functional-theory studies, see for example Y.- W. Son, M. L. Cohen, and S. G. Louie, Phys. Rev. Lett. 97, 216803 (2006). 34 For continuum DW studies, see for example (a) L. Brey and H. A. Fertig, Phys. Rev. B 73, 235411 (2006); (b) L. Brey and H. A. Fertig, Phys. Rev. B 73, 195408 (2006). 35 For a recent theoretical review, see K. Wakabayashi, K.- I. Sasaki, T. Nakanishi, and T. Enoki, Sci. Technol. Adv. Mater. 11, 054504 (2010). 36 See for example J.-M. Cai, P. Ruffieux, R. Jaafar et al., 15 I. Romanovsky, C. Yannouleas, and U. Landman, Phys. Nature 466, 470 (2010). Rev. B 85, 165434 (2012). 16 Y. Aharonov and D. Bohm, Phys. Rev. 115, 485 (1959). 17 For the original nonrelativistic model, see R. de L. Kronig and W. G. Penney, Proc. R. Soc. London, Ser. A 130, 499 (1931). 18 M. Buttiker, Y. Imry, and R. Landauer, Phys. Lett. A 96, 365 (1983). 19 R. Jackiw, Phys. Scr. T146, 014005 (2012). 20 R. Jackiw and J. R. Schrieffer, Nucl. Phys. B 190, 253 (1981). 21 A. J. Heeger, S. Kivelson, J. R. Schrieffer, and W.-P. Su, Rev. Mod. Phys. 60, 781 (1988). 22 D. K. Campbell, Synth. Met. 125, 117 (2001). 23 F. Halzen and A. D. Martin, Quarks and leptons: an in- troductory course in modern particle physics (New York, Wiley, 1984). 24 P. Recher, B. Trauzettel, A. Rycerz, Y. M. Blanter, C. W. J. Beenakker, and A. F. Morpurgo, Phys. Rev. B 76, 37 R. Jackiw and C. Rebbi, Phys. Rev. D 13, 3398 (1976). 38 Carbon nanotubes (CNTs) are yet another extensively studied 1D carbon allotrope, which also exhibit a ∆0 gap at the Fermi level [see reviews: T. Ando, J. Phys. Soc. Jpn. 74, 777 (2005); J.-Ch. Charlier, X. Blase, and S. Roche, Rev. Mod. Phys. 79, 677 (2007)]. The ∆0 gap in CNTs, however, is due to a finite quantum-size effect [see Sec. 3.2 in the T. Ando review], namely to a quantization of the free-electron motion along the circumference of the nan- otube. CNTs do not seem to be susceptible to a spon- taneous Peierls transition either [R. Saito, M. Fujita, G. Dresselhaus, and M. S. Dresselhaus, Appl. Phys. Lett. 60, 2204 (1992)]. However, this subject remains still open to date. A most recent study suggests that a Peierls transi- tion may be possible in narrow CNTs; see G. Dumont, P. Boulanger, M. Cot´e, and M. Ernzerhof, Phys. Rev. B 82, 035419 (2010). 39 F. Englert and R. Brout, Phys. Rev. Lett. 13, 321 (1964). 40 P. W. Higgs, Phys. Rev. Lett. 13, 508 (1964). 41 G. S. Guralnik, C. R. Hagen, and T. W. B. Kibble, Phys. Rev. Lett. 13, 585 (1964). 42 P. W. Anderson, Phys. Rev. 130, 439 (1962). 43 T. Sato, K. Segawa, K. Kosaka, S. Souma, K. Nakayama, K. Eto, T. Minami, Y. Ando, and T. Takahashi, Nature Phys. 7, 840 (2011). 44 M. Gibertini, A. Singha, V. Pellegrini et al., Phys. Rev. B 79, 241406 (2009). 45 Study cases of armchair graphene rings with arms corre- sponding to the other two classes of ideal aGNRs (namely with NW = 3l − 2 and NW = 3l, l = 1, 2, 3, . . .) will be presented in a future publication [I. Romanovsky, C. Yan- nouleas, and U. Landman, in preparation]. 46 R. B. Lehoucq, D. C. Sorensen, and C. Yang, ARPACK Users' Guide: Solution of Large-Scale Eigenvalue Prob- lems with Implicitly Restarted Arnoldi Methods (SIAM, Philadelphia, 1998). 47 H.-F. Cheung, Y. Gefen, E. K. Riedel, and W.-H. Shih, Phys. Rev. B 37, 6050 (1988). 48 S. Kivelson and O. L. Chapman, Phys. Rev. B 28, 7236 (1983). 49 See, e.g., Appendix A in H. Nitta, T. Kudo, and H. Mi- nowa, Am. J. Phys. 67, 966 (1999). 50 See, e.g., Sec. II in Ref. 2 and A. S´anchez, E. Maci´a, and F. Dom´ınguez-Adame, Phys. Rev. B 49, 147 (1994). 51 See also R. Gilmore, Elementary Quantum Mechanics in One Dimension (The Johns Hopkins University Press, Bal- timore, 2004), Chaps. 37 and 38. 52 I.I. Cotaescu and E. Papp, J. Phys.: Condens. Matter 19, 242206 (2007). 53 O. Klein, Z. Phys. 53, 157 (1929). 54 M. I. Katsnelson, K. S. Novoselov, and A. K. Geim, Nat. Phys. 2, 620 (2006). 55 C.-H. Park, L. Yang, Y.-W. Son, M. L. Cohen, and S. G. Louie, Phys. Rev. Lett. 101, 126804 (2008). 56 M. Barbier, F. M. Peeters, P. Vasilopoulos, and J. M. Pereira, Jr., Phys. Rev. B 77, 115446 (2008). 57 D. P. Arovas, L. Brey, H. A. Fertig, E.-A. Kim, and K. Ziegler, New J. Phys. 12, 123020 (2010). 58 The total length L = 2a + b equals approximately that of the side of the corresponding graphene ring. 59 R. Rajaraman and J. S. Bell, Phys. Lett. B 116, 151 (1982). 60 S. Kivelson and J. R. Schrieffer, Phys. Rev. B 25, 6447 (1982). 61 M. V. Berry and R. J. Mondragon, Proc. R. Soc. London, Ser. A 412, 53 (1987). 62 D. K. Campbell, A. R. Bishop, and K. Fesser, Phys. Rev. B 26, 6862 (1982). 63 In the nonrelativistic limit, for negative energies (repre- senting holes with −e charge), one sets E = −Mv2 and uses a negative potential step −V . The resulting spec- F + (cid:101)E trum is mirror-symmetric about the horizontal axis (cid:101)E = 0; note the we mentioned here holes rather than antiparticles. 64 S. L. Glashow, Nucl. Phys. 22, 579 (1961). 65 S. Weinberg, Phys. Rev. Lett. 19, 1264 (1967). 66 A. Salam, Elementary Particle Physics: Relativistic Groups and Analyticity, Eighth Nobel Symposium Stock- holm, edited by N. Svartholm (Almquvist and Wiksell, Stockholm, 1968) p 367. 67 Symmetry breaking in elementary particle physics was introduced in Ref. 68. For the importance of symmetry 13 breaking in bulk condensed-matter systems, see Ref. 69. For symmetry breaking and the subsequent step of sym- metry restoration in finite systems, see Ref. 70. 68 Y. Nambu, Phys. Rev. Lett. 4, 380 (1960); Y. Nambu and G. Jona-Lasinio, Phys. Rev. 124, 246 (1961); Phys. Rev. 122, 345 (1961). 69 P. W. Anderson, Science 177 393 (1972); Basic Notions of Condensed Matter Physics (Addison-Wesley, Reading, 1984). 70 C. Yannouleas and U. Landman, Rep. Prog. Phys. 70, 2067 (2007). 71 For a pedagogical introduction, proceeding step-by-step from the simplest example to the full Higgs mechanism of mass generation in the Standard Model, see Refs. 72 and 73. 72 V. A. Bednyakov, N. D. Giokaris, and A. V. Bednyakov, Phys. Part. Nucl. 39, 13 (2008); arXiv:hep-ph/0703280. 73 D. McMahon, Quantum Field Theory DeMystified (McGraw-Hill, New York, 2008). 74 R. Rajaraman, Solitons and Instantons: An Introduc- tion to Solitons and Instantons in Quantum Field Theory (North-Holland, Amsterdam, 1982). 75 T. Vachaspati, Kinks and Domain Walls: An Introduction to Classical and Quantum Solitons (Cambridge University Press, Cambridge, 2006). 76 For a review of quantum size effects of electrons confined in ultrathin metallic films see: M. C. Tringides, M. Jalo- chowski, and E. Bauer, Phys. Today 60, 50 (2007). 77 For a review of quantum size effects before the discovery [Ref. 78(a)] of electronic shell effects in clusters, see: W. P. Halperin, Rev. Mod. Phys. 58, 533 (1986). 78 For electronic shell effects in clusters, see (a) W. D. Knight, K. Clemenger, W. A. De Heer, W. A. Saunders, M. Y. Chou, and M. L. Cohen, Phys. Rev. Lett. 52, 2141 (1984); (b) W. A. De Heer, Rev. Mod. Phys. 65, 611 (1993); (c) C. Yannouleas and U. Landman, in Large Clusters of Atoms and Molecules, edited by T. P. Martin (Kluwer, Dordrecht, 1996) p. 131. 79 For electronic shell effects in 2D constrictions and in 3D quantum wires, see: (a) B. J. van Wees, H. van Houten, C. W. J. Beenakker, J. G. Williamson, L. P. Kouwenhoven, D. van der Marel, and C. T. Foxton, Phys. Rev. Lett. 60, 646 (1988); (b) J. I. Pascual, J. Mendez, J. Gomez-Herrero, A. M. Baro, N. Garcia, U. Landman, W. D. Luedtke, E. N. Bogachek, and H.-P. Cheng, Science 267, 1783 (1995); (c) C. Yannouleas, E. N. Bogachek, and U. Landman, Phys. Rev. B 57, 4872 (1998); (d) N. Agrait, A. L. Yeyati, and J. M. van Ruitenbeek, Phys. Reps. 377, 81 (2003). 80 For simplicity, in describing the quantum-field-theory analogies in Section V, the scalar field φ(x) = m(x) associ- ated with the armchair trigonal case [Fig. 6(b)] was treated as a constant mass along the full perimeter of the triangle. Although this is a reasonable zero-order approximation, as it is evident from Fig. 6(b), the φ(x) field exhibits depres- sions (it attains zero values) at the corners of the triangle. Properly speaking, such a scalar-field dependence leads to states that are not identical with free Dirac particles. In the case of conducting polymers, the corresponding states are known as polarons [see Refs. 21, 22, and 62, and S. Brazovskii and N. N. Kirova, Zh. Eksp. Teor. Fiz. Pis'ma 33, 6 (1981) [JETP Lett. 33, 4 (1981)]. In particle physics, the similar states are referred to as fermion bags [see, e.g., D. K. Campbell and Y.-T. Liao, Phys. Rev. D 14, 2093 (1976); R. F. Dashen, B. Hasslacher, and A. Neveu, Phys. 14 Rev. D 12, 2443 (1975)], and they correspond to mass ac- quisition processes that are in principal more general than the Higgs mechanism [see R. MacKenzie and W. F. Palmer, Phys. Rev. D 42, 701 (1990)]. The polaron states provide also a connection [see Ref. 22] to the self-consistent Gross- Neveu model [D. J. Gross and A. Neveu, Phys. Rev. D 10, 3235 (1974)]. Such aspects, as they relate to the planar graphene rings, will be elaborated in a future publication [see Ref. 45]. 81 The general particle-physics and quantum-field theory analogies, discussed in this paper regarding the occur- rence or absence of an energy gap at the Fermi level of graphene-based nanostructures, may also provide a unifying theoretical background for the gap behavior in the moir´e superlattices formed in heterostructures cre- ated by placing a graphene monolayer on a hexagonal boron nitride (hBN) substrate. For the latest develop- ments in this area, see, e.g., the following eprints (no- ticed at the proof-reading stage of this paper): B. Hunt, J. D. Sanchez-Yamagishi, A. F. Young, K. Watanabe, T. Taniguchi, P. Moon, M. Koshino, P. Jarillo-Herrero, and R. C. Ashoori, arXiv:1303.6942 and J. R. Wallbank, A. A. Patel, M. Mucha-Kruczynski, A. K. Geim, and V. I. Fal'ko, arXiv:1211.4711. 

1409.4318

3

1409

Magnetic anisotropy of FePt nanoparticles

We carry out a systematic theoretical investigation of Magneto Crystalline Anisotropy (MCA) of L10 FePt clusters with alternating Fe and Pt planes along the (001) direction. We calculate the structural relaxation and magnetic moment of each cluster by using ab initio spin-polarized density functional theory (DFT), and the MCA with both spin-polarized DFT (including spin-orbit coupling self-consistently) and the torque method. We find that the MCA of any composite structure of a given size is enhanced with respect to that of the same-sized pure Pt or pure Fe cluster as well as to that of any pair of Fe and Pt atoms in bulk L10 FePt. This enhancement results from the hybridization we observe between the 3d orbital of the Fe atoms and the 5d orbital of their Pt neighbors. This hybridization, however, affects the electronic properties of the component atoms in significantly different ways. While it somewhat increases the spin moment of the Fe atoms, it has little effect on their orbital moment; at the same time, it greatly increases both the spin and orbital moment of the Pt atoms. Given the fact that the spin-orbit coupling (SOC) constant of Pt is about 7 times greater than that of Fe, this Fe-induced jump in the orbital moment of the Pt atoms produces the increase in MCA of the composite structures over that of their pure counterparts. That any composite structure exhibits higher MCA than bulk L10 FePt results from the lower coordination of Pt atoms in the cluster, whether Fe or Pt predominates within it. We also find that bipyramidal clusters whose central layer is Pt have higher MCA than their same-sized counterparts whose central layer is Fe. This results from the fact that Pt atoms in such configurations are coordinated with more Fe atoms than in the latter. By thus participating in more instances of hybridization, they contribute higher orbital moments to the overall MCA of the unit.

cond-mat.mes-hall

cond-mat

Magnetic anisotropy of FePt nanoparticles Alamgir Kabira , Jun Hub, Volodymyr Turkowskia, Ruqian Wu,b Robert Camleya,c and Talat S. Rahmana a Department of Physics, University of Central Florida, Orlando, FL 32816 b Department of Physics and Astronomy, University of California, Irvine, CA 92697 c Department of Physics, University of Colorado, Colorado Springs, CO 80918 ABSTRACT. We carry out a systematic theoretical investigation of Magneto Crystalline Anisotropy (MCA) of L10 FePt clusters with alternating Fe and Pt planes along the (001) direction. The clusters studied contain 30 - 484 atoms. We calculate the structural relaxation and magnetic moment of each cluster by using ab initio spin-polarized density functional theory (DFT), and the MCA with both the self-consistent direct method and the torque method. We find the two methods give equivalent results for all the structures examined. We find that bipyramidal clusters whose central layer is Pt have higher MCA than their same-sized counterparts whose central layer is Fe. This results from the fact that the Pt atoms in such configurations are coordinated with more Fe atoms than in the latter. By thus participating in more instances of hybridization, they contribute higher orbital moments to the overall MCA of the unit. Our findings suggest that by properly tailoring the structure, one can avoid encapsulating the FePt L10 nanoparticles, as has been proposed earlier to protect a high and stable magnetic anisotropy. Additionally, using a simple model to capture the thermal behavior, we predict that a five-layered nanoparticle with approximately 700 atoms can be expected to be useful in magnetic recording applications at room temperature. I. Introduction Understanding the physics of smaller structures can help in exploiting their useful properties. For example, the high surface-to-volume ratio and tailorable surface chemistry of metal nanoparticles have long been relied on in optimizing the activity and specificity of catalysts.1 And small metal-particle arrays have been used to build single-electron devices.2, 3 Recently demand has arisen for magnetic particles with high anisotropic energy necessary for energy-harvesting technologies4 as well as for ultra-high-density recording media.5 Satisfaction of this demand requires development of metal thin-film media with smaller particles, more tightly-sized distributions, and optimized compositions.6, 7 Since the mid-1930s Fe-Pt alloys of L10 phase have been known to exhibit high magnetocrystalline anisotropy.8 Since among the various ferromagnetic metals and alloys FePt alloys show large perpendicular MCA (on the order of meV/atom9) and since, in nanoscale particles, they do not exhibit the superparamagnetism often characteristic of such small clusters,10 it lends itself to magnetic applications requiring small-grained constructions. FePt 1 alloys also have advantages over rare-earth transition-metal-based compound with high MCA, such as Nd2Fe14B and SmCo5 in that they are very ductile and chemically inert.11 L10-based thin films and nanoparticles in general would seem to be promising candidates for ultra-high density magnetic storage media owing to their high corrosion resistance and excellent intrinsic magnetic properties.12 But, in contrast to the fine grain of the L10 FePt systems, other conventional magnetic materials (Fe, Co, Ni and their alloys) would, through thermal fluctuation, within a very short time become superparamagnetic, losing any stored information. And given their high cost, bulk FePt-based permanent magnets can be used only for some especially delicate applications, as in magnetic micro-electromechanical systems (MEMS),13 and in dentistry as attachment devices for retaining a dental cap in the cavity.14 The chemically ordered FePt L10 structure can be obtained by annealing from the fcc structure of FePt alloy or by deposition on substrate above the L10 ordering temperature.15, 16 At high temperature an fcc solid solution of Fe-Pt is observed in the A1 phase; below 1573K (and down to 973K) alloys with a nearly equal number of Fe and Pt atoms (35-55% Pt) show order-disorder transition and the L10 ordered phase begins to form.11 Though the L10 phase is typically obtained by heat treatment of A1 phase, it can also be produced by chemical synthesis of nanoparticles.12 Deposition of alternate Fe and Pt monolayers can reduce the onset temperature of L10 phase.17 Another way to obtain L10 phase experimentally is by annealing alternating multi-layers of Fe and Pt.11 Stable FePt L10 nanoparticles have been prepared experimentally18 both without any covering and with Al encapsulation. The L10 structure has alternating Fe and Pt planes along the (001) direction, which is also the easy axis of magnetization, abandoning the cubic symmetry of the fcc system. In this type of layered magnetic system the MCA is mainly due to the contribution from the Pt (5d element) having large spin-orbit coupling while the Fe (3d element) provides the exchange splitting of the Pt sub-lattice.19-21 It is well known that in the FePt system, the Pt atoms play an important role in magnetic anisotropy, because the hybridization of Fe orbitals cause spin polarization of Pt atoms, which in turn enhance the MCA owing to the relative strong SOC of the Pt atoms in comparison with that of the Fe atoms. There have been several theoretical studies on MCA of nanoparticles. Cyrille et al.22 calculated the size- and shape-dependent magnetic properties of L10 FePt clusters using a tight-binding approach. In their study the central plane of clusters is always Fe and they do not take into account the atomic relaxation of the clusters. Błonski and Hafner23 undertook ab initio density- functional calculations of the magnetic anisotropy of supported nanostructures. Fernandez- Seivane and Ferrer24 studied the correlation of the magnetic anisotropy with the geometric structure and magnetic ordering of small atomic clusters of sizes up to 7 atoms. Gruner et al. 18, 25 demonstrated that in cuboctahedral nanoparticles the high anisotropy of the layers increases as one moves towards the surface, and the anisotropy can be even enhanced by embedding the material in some suitable other metal (e.g., Au in the case of Pt-terminated structures). Various experimental studies, using X-ray Magnetic Circular Dichroism Spectroscopy (XMCD) or X-ray 2 Absorption spectra (XAS), have recently confirmed the enhancement of MCA in free or supported clusters.26-28 The earlier theoretical results suggest that in order to preserve the high values of the MCA, one needs to encapsulate the particles. 25 However issues coming from encapsulation such as charge transfer or structural integrity may adversely affect the MCA. In this work, we explore a possibility to tune the MCA by changing system geometry in such a way that the anisotropy mostly comes from the central part of the particle, which may help avoid the necessity of capping. We carry out a systematic theoretical investigation of the MCA of L10 FePt clusters consisting of alternating Fe and Pt planes along the (001) direction. The clusters studied have 1(2), 2(3), 3(4) and 4(5) layers of Fe(Pt) atoms – both with Pt outer layers and with Fe outer layers – of sizes 30, 38, 71, 79, 114, 132, 140, 230, 386 and 484 atoms, respectively. We also examine the electronic structural and magnetic properties (including the orbital moments) of each atom in each of these configurations. To calculate the structural relaxation and magnetic moments of the clusters we adopted an ab initio spin-polarized density functional theory (DFT) approach. To calculate the MCA we employed two methods: (i) the direct method where we take the difference in band energy for two orientations of the average magnetic moment and (ii) the torque method.29, 30 The latter method is simpler and computationally less demanding. In this work we show its validity even for systems at the nanoscale. We found that the MCA of layered L10 FePt clusters is enhanced over that of both bulk FePt and that of either a pure Fe or Pt cluster of comparable size, all with L10 geometry. Previous investigations attributed this enhancement is due to the hybridization 3d orbital of Fe atom with the 5d orbital of Pt atom. Our calculations indicate that this is so because this hybridization increases both spin and orbital moment of the Pt atoms. And given the large spin-orbit coupling constant of Pt it is this enhanced orbital moment of Pt that is responsible for the higher anisotropy of the system as a whole. We also found that when the central layer of the bipyramidal cluster is Pt, the cluster has higher MCA than a cluster of the same size but with Fe as the central layer, in contrast to the cuboctahedral cases, 18, 25 in which it is the surface layers that play crucial role in high MCA . This stems from the fact that when Pt atoms comprise the central layer they have more Fe atoms neighboring them, so that hybridization increases, lending them higher orbital moments than are possessed by Pt atoms in other layers. This center-of- system ‘concentration’ of the MCA makes it possible to preserve the anisotropy without having to resort to capping of the particles. 3 II. Computational Details To calculate the structural relaxation and spin and orbital magnetic moments we used the ab initio spin-polarized DFT approach implemented in the VASP code.31 In describing the electronic exchange and correlation effects we used the spin-polarized generalized-gradient approximation (GGA) with the Pardew and Wang (PW91) functional32 and the spin interpolation proposed by Vosko et al.33 In calculating the ionic relaxation, we employed the conjugate gradient algorithm. In describing the electron-ion interaction we used the projector augmented- wave (PAW)34 formalism. To calculate the strength of the spin-orbit coupling it is essential to take relativistic effects into account. We did so by choosing the relativistic version of PW91 as an input to the non-collinear mode framework implemented in VASP.35, 36 To construct the bulk FePt L10 structure we replaced every alternating layer of fcc Fe with Pt atoms in (001) planes. This ordering induces a contraction along the (001) direction of the fcc lattice, which reduces the ratio 𝑐∗𝑎∗� , where 𝑎∗ is the nearest-neighbor distance in (001) planes (related to the primitive cell parameter a as 𝑎=𝑎∗√2), from the fcc value (√2) to 1.363. Fig. 1 shows the relation between the lattice parameter for the pseudo-cell and the parameter for the primitive unit cell. Figure 1. The L1o crystal structure of bulk FePt (Fe atoms in dark (red) and Pt in light (gray). The dashed line represents the primitive cell. The parameters are related as follows: 𝒄=𝒄∗ and 𝒂=𝒂∗√𝟐 , where 𝒄∗ and 𝒂∗ are the lattice parameters of the primitive cell and c and a are the corresponding parameters of the pseudo-cell. 4 Lattice distortion brings about chemical re-ordering in the unit cell. For the lattice parameter we used the experimentally-derived data for bulk powder,37 (𝑎∗=2.72Å an 𝑐∗𝑎∗� =1.36), and then relaxed the structure using the DFT approach described above. For dimers and clusters, in relaxing the structure using the method described above, we set the vacuum space to 12 Å in all three directions, in order to prevent artificial electric field interactions between the images, and used only one (Gamma) point in the Brillouin zone. To obtain the relaxed geometry of a cluster of given size and shape, we first obtained the relaxed lattice parameter for bulk L10 FePt, then cut the bulk to the size and layered shape of the cluster in question to get its initial configuration, and finally relaxed that configuration. In calculating the MCA of L10 FePt nanoparticles we used both the direct method23 (i.e., including spin-orbit coupling self-consistently), and the torque method. 29, 30 In the direct method, once the structural relaxation for a given cluster is completed, we calculate the MCA by comparing the band energy between two magnetic orientations. The band energy of the system is calculated by using spin-polarized DFT, now taking SOC into account. (1) 𝑀𝑀𝑀=𝐸(↑)−𝐸(→)=∑ 𝑜𝑜𝑜′ 𝜀𝑖(↑) −∑ 𝑜𝑜𝑜′′ 𝜀𝑖(→) where 𝜀𝑖 is the band energy of the 𝑖𝑡ℎ state and the arrow in the parentheses denote the direction of magnetization. (For the surface of a film, these are usually the directions perpendicular and parallel to the film and for nanoparticles they are the directions along the z-axis and parallel to xy-plane, respectively). We aligned the magnetization towards two mutually perpendicular directions setting spin arrangement in the system to be non-collinear. The sums in equation (1) are usually large numbers (on the order of hundreds of eV), whereas the MCA is on the order of few meV. Since MCA is a small number coming from the difference of two large numbers, one needs to take great care, in determining the Kohn-Sham energy of the system, in selecting the convergence criterion for the calculation. In fact, for accurate integration, it is necessary to use a very fine k-point mesh in reciprocal space. And the fact that a very accurate convergence of energy is also required in the self-consistent cycle makes the calculations rather expensive. This method can give quite divergent results if one does not use sufficient number of integration points: for example Gay and Richter38 predict the easy axis of monolayer Fe to be perpendicular to the plane and found the anisotropic energy to be -0.4 meV/atom, whereas Karas et at.39 report the easy axis of the same system to be along the plane of the layer, and the value of anisotropy to be 3.4 meV/atom. Since we use a large supercell in the calculations, we do not need to include large number of k points, though we did carry out a set of calculations for 3x3x3 k-point mesh as a test for our 38-atom clusters, and the results are almost same as that obtained with the Γ point. Therefore, in other calculations we confined ourselves to using simply the Γ point. 5 An alternative way of calculating the MCA is the torque method29, 30 which is suitable for systems with uniaxial symmetry. The advantages of this method are that it does not require calculations of the total energy of the system with very high accuracy and that it is much faster because it does not require self-consistent calculations with SOC for two different directions of the magnetization. For our calculations using the torque method we employed the VASP post- processing package developed by Jun Hu et al.,40 based on the augmented-wave projection of the SOC operator41. We also calculated the MCA to see whether the results differed considerably from those obtained from the direct method. (As Figure 7 indicates, they generally did not.) III. Results and Discussion Magnetic anisotropy of the bulk system The lattice parameters of the relaxed structure of bulk L10 FePt do not change much from the alloy the Pt atoms possess a magnetic moment. The magnetic moment of Fe atoms increases substantially in the alloy from their values in bulk Fe. The enhancement of the Fe magnetic experimental ones (𝑎∗=2.72 Å and 𝑐∗𝑎∗� =1.36): the in-plane parameter 𝑎∗=2.74 Å, and the ratio value 𝑐∗𝑎∗� =1.37. We have obtained the following values for the magnetization: 2.85 𝜇𝐵 for the Fe atoms and 0.36 𝜇𝐵 for the Pt atoms, both in good agreement with experiment (Fe: 2.90 𝜇𝐵, Pt: 0.34 𝜇𝐵).37 It is worth mentioning that though bulk Pt is nonmagnetic, in the FePt moment is a consequence of the hybridization between the states of the 3d orbitals of the Fe atoms and not only the 5d but also s- and p-orbitals of their neighboring Pt atoms. As can be seen in Fig. 2, the hybridization causes a broadening of the d-bands for the majority spins, while magnetic moment (0.36 𝜇𝐵) of Pt atoms in FePt clusters comes mostly from the hybridization of its effect on the shift of the minority spins, though smaller, that shift moves the band across the Fermi level into the area unoccupied by any electron. Thus, it is natural to suppose that the finite the minority spin bands. 6  Figure 2. Spin- and orbital-projected densities of states for different cases: (a) bulk Fe, (b) bulk Pt, (c) Fe atom in the L10 FePt and (d) Pt atom in L10 FePt.(e) Spin-projected density of states for bulk L10 FePt The density of states (DOS) of bulk L10 FePt is plotted in Fig. 2, where we also present the projected DOS for the Fe and Pt atoms, which is similar to the one reported by C. Barreteau et al. 22 Our calculations show MCA value for the bulk to be 2.22 meV per FePt pair, which is also in agreement with other studies,2 encouraging confidence that our results in the nanocase are reliable as well. We attribute the large MCA to the large SO coupling of Pt atoms and to the increase in orbital moment owing to the strong hybridization of their 5d orbitals with the 3d orbitals of Fe. It is this hybridization that breaks the symmetry of free energy in the two perpendicular directions, resulting in a surplus of free energy in the direction of magnetization.19 Interestingly, Lyubina et al. found that the anisotropy of the disordered phase of FePt is an order of magnitude less than that of the L10 phase.11 Magnetic anisotropy of the dimer To gain more insight into the nature of MCA in the nanoparticles that are the chief object of our study, we also considered the case of pure Fe, pure Pt, and FePt dimers. For Fe2 the bond length is found to be 1.98 Å and that for Pt2 the bond length is 2.37 Å which are in good agreement with the bond length reported in the earlier work by Blonski et al.23 and the references therein. The experimental data is available only for the Fe dimer (bond length is 1.87 Å)42 which is also in good agreement with our calculations. For the FePt dimer we obtained a dimer bondlength of 7 2.18 Å when leaving spin-orbit coupling out of account, and 2.17 Å when taking SOC into account. We thus infer that SOC does not play an important role in determining the geometry of this system. This result is also in good agreement with the results of Ref.43 Fe2 turns out to exhibit very small anisotropy owing to the relatively small SOC of the Fe atom. The SOC of Fe atom is 81.6 meV and of the Pt atom is nearly 7 times greater 544 meV44. Contrary to the other two dimers, in Pt2 the spin moment and orbital moment significantly differ in the directions along the dimer and perpendicular to it. Therefore, it is the Pt atom’s high SOC and relatively large orbital moment in Pt2 that together account for the fact that the MCA of that dimer greatly exceeds those of the other two. The MCA of the mixed FePt system has a value - 10.37 meV - between those of the two monometallic cases - 0.07 and 55.94 .meV, respectively. Table 1 and Figure 3 help us understand why this is the case. In the FePt dimer the total magnetic moment of the Fe atom is 3.22 𝜇𝐵 and that of the Pt atom - 0.58 𝜇𝐵 . The large magnetic moment of Fe in this dimer can be attributed to the charge transfer from the 3d orbital of Fe to the 5d orbital of the Pt atom, which creates a polarization and an extra “hole charge” on the Fe atom.43 The MCA for FePt dimer is 10.37 meV (with the easy axis perpendicular to the dimer axis). This value is higher than the value for Fe2, because the values of both the spin and orbital momenta are larger in the FePt dimer. Our results for the spin and orbital momenta and the MCA energy along with corresponding available numerical results obtained by other methods are compiled in Table 1. Figure 3. Density of states of (a) each Fe atom in the Fe dimer, (b) of each Pt atom in the Pt dimer and (c) of the Fe and Pt atoms in the FePt dimer. The solid (red) line corresponds to Fe atoms and the doted(blue) to Pt atoms. The dashed green lines indicate the HOMO level of the dimers. In all three cases, the easy axis of magnetization coincides with the direction of highest orbital momentum, in agreement with Bruno45 (See Table 1). The negative MCA for the case of the FePt dimer means that the easy axis of magnetization is perpendicular to the dimer axis in this case. 8  E100-E001 0.07 0.3 32 0.5 55.94 46.3 220.0 75.00 -10.37 Table 1: Magnetic moments and MCA energy (in meV) of the Fe, Pt and FePt dimers. LDA stands for the local density approximation in DFT. Dimer Spin Moment (µB) Orbital Moment (µB) MCA Fe2 Pt2 FePt (001) direction 0.16 0.16 0.89 0.10 0.80 0.80 1.20 0.80 0.41 0.40 (100) direction (001) direction (100) direction This work GGA23 LDA46 GGA43 This work GGA23 LDA47 GGA43 This work GGA43 5.99 5.84 6.00 6.00 1.89 1.88 1.90 1.80 4.16 4.30 0.32 0.32 1.89 0.25 2.74 2.74 2.40 2.40 0.36 0.2 5.99 5.84 6.00 6.00 1.38 1.34 1.65 1.70 4.26 4.30 9  Figure 4. Projected DOS for the d orbitals of (a) the Fe atom in the Fe dimer, (b) the Pt atom in the Pt dimer, (c) the Fe atom in FePt dimer, and (d) the Pt atom in FePt dimer. The dashed lines highlight the HOMO level in each case. The projected DOS for the 𝑑 orbitals of the Fe and Pt atoms in the dimers are shown in Fig.4. As follows from this figure, the majority spin orbitals have zero occupancy at the HOMO level, but the minority spins are present there, indicating that the change of the orbital occupancy upon hybridization is defined by the charge transfer for the spin-down electrons only. We thus conclude that the anisotropy and the magnetization of the dimers are generated by the minority spins as well. 10  Figure 5. Orbital moment vs MCA of dimers. The arrows the direction of easy axes for magnetization – perpendicular to the dimer axis for FePt and parallel for the monometallic clusters. indicate To summarize, the MCA is proportional to the change of orbital momentum in two different directions; the comparatively small value of MCA for Fe2 is due to its small SOC constant; larger value of MCA for Pt2 is due both to its large SOC constant and to its relatively large orbital moment; while the intermediate value of MCA in the FePt dimer is due to mutual tempering of each atom of the other’s SOC and to the fact that its orbital moment falls between those of Fe2 and of Pt2 . Magnetic anisotropy of L10 FePt nanoclusters The high MCA of bulk L10 FePt alloy with equiatomic composition has stimulated researchers to inquire into the properties of its small particles. To be sure, there are some problems in maintaining the bulk geometry for the nanoparticles. For example, Muller et al. predicted theoretically that the L10 phase is not thermodynamically stable48, and another study, by Jarvi et al.49 showed that the structure may alter, disrupting the original atomic order. Moreover, several experimental studies support the existence of a minimum size limit below which the L10 order can no longer be achieved,50, 51 and other studies have shown the migration of Pt atoms towards the surface in smaller particles.52, 53 However, for more definitive answers concerning feasibility of exploiting FePt nanoparticles in the applications mentioned in the Introduction, detailed and systematic studies are in order. Those undertaken so far are suggestive. In particular, it is known that the magnetic properties of nanoparticles depend on the size, shape and the way of synthesizing.11 Gas-phase particles can in fact exhibit lower MCA than perfectly-ordered bulk L10 alloys, because (i) their internal structure may not be perfect L10, and individual particles can become multiply-intertwined, (ii) the Pt atoms may tend to migrate towards the surface,43,44 or (iii) an inhomogeneous alloying may be present from the beginning, as indicated by the EXAFS measurements by Antoniak.54 The enhancement of surface-to-volume ratio (and hence the size) of a nanoparticle plays a significant role in its MCA. For example, the crystal symmetry-dependent quenching of the orbital magnetic moments disappears for all surface atoms of nanoparticles, thereby enhancing 11 their orbital moments relative to those of bulk or core atoms.55 (Antioniak’s XMCD measurements have confirmed this for Fe surface atoms in L10 FePt nanoparticles.56) The cluster structures we studied – constructed as described in the section on Computational Details – are shown in the Fig. 6. Figure 6. Clusters of the twenty L10 FePt nanoparticles under comparative study here. Dark (red) and light (gray) balls represent Fe and Pt atoms, respectively. The clusters whose central layer is composed of Pt atoms are presented in (a), and the clusters with central layer composed of Fe atoms are presented in (b). The values of MCA for the above clusters obtained with both methods are presented in Fig. 7. 12  Figure 7: MCA of the clusters studied, calculated by different methods. Light (white striped) bars represent MCA energy according to the direct approach; dark (magenta crossed) bars indicate results under the torque approach. The results for MCA calculated with the direct and the torque methods generally agree well. This is encouraging, particularly for nanostructures, because the direct method in this case is much more computationally demanding. One definite conclusion supported by the calculations is that those nanoparticles with a larger number of Pt atoms have larger MCA than their counterparts with a larger number of Fe atoms. We also find that the atoms on the outside of the clusters have higher magnetic moments than do atoms inside. One example of such a situation is shown in Fig.8 for Fe72Pt68 and Fe68Pt72 clusters. The line through which we calculate the magnetic moment is presented by the arrow in Fig. 8(a). The line (red) with solid circles and the line (blue) with open circles in Fig. 8(b) represent magnetization for Fe (middle layer with Fe atoms, Fe72Pt68) and Pt (middle layer with Pt atoms, Fe68Pt72) atoms, respectively. It is clear from the figure that the outer atoms in the cluster have larger magnetization than the inner atoms. As we shall see, this comes from the fact that exterior atoms have fewer neighbors than do those inside the cluster. 13  Figure 8: Magnetic moment of atoms at different positions in central layer of the cluster. The arrows in (a) indicate the row in which we picked atoms for comparing their magnetization. (b) The magnetic moment of atoms at different positions. For this pair of clusters, there are six atoms along the line passing through the cluster’s center; the 3rd and 4th atoms are at the center of the cluster. Pt atoms in the FePt clusters exhibit magnetization, in contrast to atoms in bulk Pt (where they exhibit virtually none). Once again, the inside Pt atoms have smaller magnetization than do the Pt atoms on the outside. The contrast between Fe atoms in our FePt clusters and Fe atoms in bulk Fe is different but parallel: Atoms in bulk Fe do show magnetization, but Fe atoms in FePt clusters exhibit even higher magnetization. Here, too, the inside atoms exhibit less magnetization than those on the outside of the cluster. All four contrasts – between magnetization of atoms pure bulk and in composite clusters, and between that of interior and exterior atoms within clusters of the same size – are due to the orbital hybridization of Pt with Fe atoms. The lower number of neighbors for surface atoms explains the narrowing of the 3d orbital bands of the Fe and the 5d bands of Pt that is in turn responsible for enhancing the magnetization of the surface atoms. The projected DOS for the atoms at different positions in Fe20Pt18 and Fe18Pt20 clusters is presented in Fig. 9. 14 (a) 15 (b) Figure 9.Projected density of states for the atoms at different positions,(a) in Fe20Pt18 and (b) in Fe18Pt20 cluster. The DOS is more localized for an atom on the surface of the cluster than for one atom inside. For all cases the majority spin band is completely filled and magnetization is due to the contribution from the minority band only. The value of m in each graph is the magnetic moment of that particular atom. 16 2 -y As this figure makes clear, there is a significant difference in the PDOS for the Pt atoms within the interior and at the periphery of the Fe18Pt20 cluster. Remarkably, the main contribution to this 2 orbital in the case of the inside Pt atoms. difference comes from the less filled spin-down dx This leads to a significant difference in the orbital momenta of the inside and outside atoms and hence of the high MCA. On the other hand, in the case of “flipped” cluster Fe20Pt18 one gets less difference in the orbital occupancies for the inside and outside atoms for both cases of Fe and Pt. Thus, it is not surprising that we find that bi-pyramidal structures with large central Pt layer has significantly higher MCA per atom compared to the cuboctahedral ones.18, 25 For example, we get an MCA per atom value 1.14 meV for the Fe64Pt68 versus 0.86 meV reported in the papers above for the 147 atom FePt cluster reported in Ref [25]. This difference can be traced to the larger percentage of Pt atoms in the center for the bi-pyramidal case compared to the cuboctahedral structure. Table 2: The magnetic moments, orbital moments, and MCA of clusters with magnetization along two different directions. Magnetic Moment (µB) mx 46.74 62.33 68.16 72.76 117.18 134.18 133.57 137.26 175.97 213.24 214.84 232.88 233.90 242.69 mz 46.26 62.03 68.06 72.75 115.19 134.14 132.46 136.53 175.67 213.44 214.76 232.64 231.71 242.95 Fe12Pt18 Fe18Pt12 Fe18pt20 Fe20Pt18 Fe32Pt39 Fe39Pt32 Fe39Pt40 Fe40Pt39 Fe50Pt64 Fe64Pt50 Fe64Pt68 Fe68Pt64 Fe68Pt72 Fe72Pt68 Orbital Moment (µB) lz lx MCA (meV) MCA=(Ex-Ez) 2.037 2.54 2.97 3.09 4.46 5.22 5.17 5.63 6.91 7.91 8.24 9.11 10.06 10.51 84.60 47.60 94.37 20.01 154.28 34.15 81.05 71.82 170.11 33.33 155.90 75.06 142.80 73.65 2.283 2.14 3.31 3.30 5.86 5.39 5.74 6.13 7.95 8.52 9.64 10.08 10.59 10.15 17 Fe98Pt132 Fe132Pt98 Fe162Pt224 Fe224Pt162 Fe260Pt224 Fe224Pt260 350.02 452.21 583.97 722.66 870.00 773.98 350.78 451.98 583.97 722.07 870.21 773.94 16.90 15.86 28.11 27.47 36.10 34.581 14.57 13.42 23.63 26.32 33.38 30.38 349.38 340.99 659.77 320.27 454.44 725.53 As Table 2 reveals, the total magnetic moment does not significantly change as the direction of magnetization shifts from (001) to (100). The orbital moment changes in such a way that the easy axis of magnetization is always along the direction of the lower orbital moments. This finding contradicts Bruno’s prediction,45 according to which the easy axis of magnetization always coincides with the direction of the highest orbital moments. To be sure, for dimers (Table 1) the direction of easy magnetization does follow the direction of highest orbital moment of the system, but for larger clusters it behaves in the opposite fashion. One would expect the easy axis of magnetization of still larger clusters to exhibit the same sort of alignment of direction of easy axis of magnetization with the direction highest orbital moment, as in those clusters under study here.. The magnitude of difference in orbital moment between the directions of magnetization ∆l=lx-lz per atom also plays a key role in determining the overall MCA of clusters. As Table 2 shows, for the majority of clusters anisotropy increases with increase in the difference between orbital moment per number of atoms. And, between clusters of the same size, anisotropy is higher in the cluster those whose predominant constituent element has the higher SOC energy. In sum: for majority of clusters we studied, the contribution to MCA comes mostly from increase in the orbital moment of the system along with increase in SOC energy, though in general case the dependence of the MCA on the size and shape of the clusters is highly nontrivial and requires further detailed studies. We also studied the MCA of different layers within the clusters (Fig. 10). 18  Figure 10: MCA of different layers of atoms in the four clusters consisting of 38 atoms. Dark (red) corresponds to Fe and light (gray) to Pt atoms. holds for all sizes of clusters we study here. along with the greater SOC energy of Pt, thus increases the MCA of the central layer. These As Fig. 10 indicates, the central plane of a cluster of the same size has much higher anisotropy when it consists of Pt than when it consists of Fe atoms. Anisotropy is also higher when the central Pt layer adjoins a layer of Fe than when its neighboring layer is Pt. This turns out to be the case for all of the clusters we studied which can be explained as follows. When a Pt atom hybridizes with neighboring Fe atoms in FePt clusters, the orbital moment of Pt atoms increases. In the example above, the Pt atom in the Fe20Pt18 cluster shows an orbital moment of 0.11 𝜇𝐵 but in Fe18Pt20 its orbital moments increases to 0.145 𝜇𝐵. This increase in the orbital moments, values of the orbital moment for the Pt atoms are much higher than that of Pt in Pt38 (0.02 𝜇𝐵), Fe in Fe38 (0.06 𝜇𝐵), Fe in Fe20Pt18 (0.06 𝜇𝐵), and Fe in Fe18Pt20 (0.04 𝜇𝐵). The same pattern the portion of the 𝑑 bandwidth contributed by the SOC component of the Hamiltonian is much smaller than the 𝑑 bandwidth as a whole.48 We see from Fig. 9 that, since all the majority spin The difference in anisotropy for different layers of Fe/Pt atoms can be explained in simple terms of orbital occupancies as we show below from perturbation theory. This is appropriate because states are completely occupied, all the empty states belong to the minority spins only. Since there are no empty states available for occupation by the spin-up electrons, there are only two types of SO interactions in the systems: the coupling between occupied and unoccupied spin-down states and the coupling between states occupied by spin-up electrons with states unoccupied by spin- down electrons. The MCA energy can be calculated by using the following formula: 48 𝐸𝑋−𝐸𝑍~𝜉2∑ ��𝑜�𝑳�𝒛�𝑢��2−��𝑜�𝑳�𝒙�𝑢��2 𝜀𝑢−𝜀𝑜 𝑜,𝑢 (2) 19 momentum operators. We have calculated the orbital occupancy of both empty and filled states where 𝜉 is the spin-orbit coupling constant, 〈O and 〈u are the occupied and unoccupied minority spin states, respectively, and 𝑳�𝒛 and 𝑳�𝒙 are the z and x components of the angular of both spin arrangements and find that for the Fe atom in the 1st layer of Fe20Pt18 𝐸𝑥−𝐸𝑧= 0.24𝜉2, for the atoms in the 3rd layer (exterior) of Fe20Pt18 this value is 𝐸𝑥−𝐸𝑧=0.17𝜉2 and for the atoms in the 3rd layer(interior) of Fe20Pt18 this value is 𝐸𝑥−𝐸𝑧=0.08𝜉2. Thus, our perturbation theory estimation from the last paragraph gives a MCA per atom for the top Fe layer that is 2.5 times larger than the corresponding value for the third Fe layer. This result is in a good agreement with the DFT ratio for the MCA’s, ~3, for the corresponding layers (see Fig.10), suggesting the ability of DFT to describe correctly the MCA in these clusters in terms of explicit orbital occupancies. The discrepancy in results for the MCA may come from the simplicity of the estimation used. For greater accuracy of the perturbation theory calculations, one would need to take into account the differences in hybridization undergone by each individual atom. An example of the orbital occupancies for differently-situated atoms is given in the Table 3. Indeed, as it follows from this Table, the occupancies of individual d-orbitals are much less than one, which suggests strong hybridization of these orbitals. Table 3: Occupancy of d-orbitals of differently situated atoms in the 1st and 3rd Fe layers in the Fe18Pt20 cluster. 1st-layer atom (all exterior) 0.51 0.29 0.30 0.18 3rd-layer atom (exterior) 0.45 0.17 0.10 0.28 3rd-layer atom (interior) 0.45 0.25 0.16, 0.13 0.28 Projected d-orbitals 𝑑𝑥2−𝑦2 𝑑𝑥𝑦 𝑑𝑥𝑧, 𝑑𝑦𝑧 𝑑𝑧2 IV. Thermal Effects The calculations above are all appropriate for zero temperature. We now deal with two thermal effects. The first topic deals with the potential application for FePt nanoparticles in the magnetic storage of data. It is well known that many nano-sized particles are superparamagnetic because their total anisotropy energy is on the order of the thermal energy. A standard measure for the magnetic stability time of a small magnetic particle is given by an Arrhenius law57 involving the probability of climbing over an energy barrier. For magnetic storage applications, the ratio of the anisotropy energy to thermal energy is quite large. For T = 25 years, one finds 𝐾𝐾𝑘𝐵𝑇=𝑙𝑙�𝑇𝑇0� �≥41. Here K is the anisotropy per unit volume and V is the volume. We have used 𝑇0=10−9𝑠𝑠𝑐 as the value for a typical attempt time for the system to climb the anisotropy barrier. With our values for the largest anisotropy (KV is about 725 meV for the Fe224Pt260 cluster), we get (3) 20 𝐾𝐾𝑘𝐵𝑇= 28 (4) Clearly, the total anisotropy energy is still not large enough compared to the thermal energy. The required anisotropy energy for the particle is about 1060 meV to satisfy the condition in Eq (3). This means that one still needs larger particles if these elements are to work for magnetic storage. We can use our previous data to project values of MCA for larger particles. It follows from Fig. 7 that the MCA of clusters with the same number of layers scales nearly linearly with the number of Pt atoms in the cluster. As an example we plot the MCA as a function of the number of Pt atoms for 5 layer clusters with Pt as central layer in Fig. 11. Figure 11. MCA with respect to number of Pt atoms in the five layered L10 FePt cluster with Pt as central layer. From Fig. 11 we can predict (for this structure) that to get an MCA of 1060 meV per cluster one needs to have a cluster with approximately 350 Pt atoms. This would correspond to a total of about 270 Fe atoms. We note that other applications of FePt nanoparticles, such as contrast agents for MRI do not have such strict stability requirements and could be done with much smaller particles, indeed with superparamagnetic nanoparticles.58 The second topic in the thermal behavior of these clusters is to estimate how the magnetization would change in these samples as a function of temperature, M(T). Dealing with this problem 21 exactly is difficult, in part because there could be multiple exchange constants within the nanoparticle, and these are not known. Instead, we only want to provide a simple estimate of M(T) and see how this could vary depending on the size and shape of the nanoparticle, and depending on the position of the Fe atoms within the nanoparticle. (6) (7) (5) The thermal averaged magnitude of a spin in an effective magnetic field is given by For bulk Fe with a body centered cubic structure there are 8 nearest neighboring atoms of each �−�12𝑆�𝑐𝑐𝑐ℎ�𝑥2𝑆� 〈𝑆〉=𝑆𝐵𝑆(𝑥) where x is the ratio of the magnetic energy to the thermal energy 𝑥=𝑔𝜇𝐵𝑆𝐻𝑒𝑒𝑒 𝑘𝐵𝑇 Here 𝑔 is the gyromagnetic ratio, S is the spin, Heff is the effective magnetic field, and 𝜇𝐵 is the magnetic moment of the atom. The function 𝐵𝑠(𝑥) is the Brillouin function, and it is given by 𝐵𝑠(𝑥)=�(2𝑆+1) 2𝑆 �𝑐𝑐𝑐ℎ�(2𝑆+1)𝑥 2𝑆 Fe atom, so the effective field can be written as 𝐻𝑒𝑒𝑒=8𝐽〈𝑆〉 where, 𝐽 is the exchange constant values of 𝑥 (appropriate near T = Tc) the Brillouin function in equation (7) may be expanded to give 𝐵(𝑥)= �𝑆+13𝑆�𝑥 Substituting this in equation (5) we can get the relation between 𝐽 and 𝑇𝑜 8𝑔𝜇𝐵𝑆(𝑆+1) (9) 3𝑘𝑇𝑐 We can use the value of the critical temperature, Tc, to obtain the exchange constant. For small Using the value of Tc = 1043 K, appropriate for bulk Fe, we can get an estimate for the exchange constants between nearest neighbor Fe atoms. and <S> is the thermal averaged magnitude of the spin. 𝐽= (8) In the nanoparticle structures, the effective field acted on each atom is different, so one cannot do standard mean-field theory. Instead we use a simple self-consistent local-mean-field magnetic model.59 As an example, we show the key elements of the calculation for the Fe20Pt18 configuration. We label each Fe atom in our nanoparticles with a different number, as an example, as seen in Fig. 12. 22  Figure 12. Schematic representation of Fe20Pt18 cluster in our simple mean field model. effective field on spin at site labeled by 1 can be written as We calculate the effective field acting on each spin, as arising from the exchange coupling from nearby spins. We identify two types of coupling: 1) Coupling of nearest neighbors within a plane with exchange constant 𝐽 2) Coupling of nearest neighbors between planes, with exchange constant 𝐽𝑝 The effective field acting on each atom can be written in terms of 𝐽 and 𝐽𝑝. For example the 𝐻1=𝐽(〈𝑆2〉+〈𝑆3〉)+𝐽𝑝〈𝑆8〉 , here we assumed that the spins are all pointing in one direction and 〈𝑆𝑛〉 is the thermal averaged magnitude of the spin at site 𝑙. We can write the effective field equation for each of the site in 〈𝑆𝑛〉=𝑆𝑛𝐵(𝑥𝑛) where, 𝑥𝑛 in this case can be written as 𝑥𝑛=𝑔𝑛𝜇𝐵𝑆𝑛(〈𝑆𝑛〉𝐻𝑛+𝐻0) 𝐾𝑇 the cluster, and the thermal averaged magnitude at any site now can be found by using the expression, (12) (10) (11) 23 One then has a set of n coupled equations involving the variables <S1> to <Sn>. We solve this iteratively by picking a site n at random, calculating the effective field at that site and the resulting new value of <Sn>. Then a new spin is chosen and the process is repeated until the process converges to final values and all spins have thermal averaged magnitudes which are consistent with the values of the neighboring spins. We use the following parameters in our calculations µB = 9.27 x 10-21 erg-Gauss, Boltzmann’s A small external magnetic field Ho = 100 Oe is used to help orient the moments and speed convergence. To see how the coupling between different planes of Fe atoms affects the results, constant is k = 1.38 x 10-16 erg/Kelvin, g = 2, and J = 1.455 x 106 Oe (bulk values of 𝐽 ), S = 1. we use two different values for the perpendicular coupling 𝐽𝑝; 𝐽𝑝=𝐽 and 𝐽𝑝=0.5𝐽. critical temperature is substantially less than the critical temperature of bulk Fe (𝑇𝑜=1043𝐾). Our results for the Fe20Pt18 and Fe39Pt40 clusters are presented in Fig.13. For both cases the However there is still a substantial moment at room temperature. The reduction in Tc is due to the reduced coordination number of the Fe atoms in the FePt structure. As might be expected, this reduced coordination has a smaller effect for the larger structure because the percentage of atoms at the surfaces and edges is smaller. Indeed we find that the Fe39Pt40 has a higher Tc value. We also see that the larger perpendicular coupling case gives a higher critical temperature, as expected. This trend is consistent with experimental data showing that larger FePt nanoparticles have a higher Tc 60 The thermal average spin values at different sites of the cluster are show in Fig, 13(b), which were done for Jp = 0.5J. As can be easily seen by symmetry, there are only three unique types of site for Fe20Pt18 and six unique site for Fe39Pt40 cluster. A key result is that even for moderate temperatures, the thermal moments at the different sites throughout the cluster can be quite different. Indeed the thermal averaged values for spins at the outer surfaces and edges can be quite small compared to those in the center for some temperatures. 24  Figure 13. (a) Averaged spin with respect to temperature for Fe20Pt18 cluster(on the top) and for Fe39Pt40 cluster(on the bottom) (b) averaged spin for inequivalent sites in Fe20Pt18(on the top) and Fe39Pt40 cluster(on the bottom). The calculation in (b) is done with Jp = 0.5 J We note that these calculations neglect any magnetic moments in the Pt atoms, and assume that all the Fe exchange values are the same. This is clearly a simplification, but the results presented here, nonetheless, should give some idea of possible thermal behaviors. V. Conclusions We have systematically studied the magnetic properties of FePt L10 nanoparticles as a function of particle sizes (30, 38, 71, 79, 114, 132, 140, 230, 386 and 484 atoms) and compositions (i.e., consisting of pure Fe and Pt atoms and of alternating planes of Fe and Pt atoms). We find that nanoparticles have much higher magnetic moments than do bulk atoms. This is due to the fact that the MCA arises from the orbital moment coupled with the spin moment and that in the bulk the system orbital moments are almost quenched, whereas in small clusters the orbital moments of the system’s atoms are considerably enhanced. We propose that this explains why it is that (as earlier studies have shown) the hybridization of the 5d(Pt) orbitals with 3d(Fe) orbitals produces a high magnetic anisotropy for layered FePt nanoparticles. We also find that clusters 25 with Pt atoms as the central layer have much larger anisotropy than those in which the central layer consists of Fe atoms. The explanation for this is that the central layer has more atoms than other layers in the cluster, and when these atoms are of high orbital moment – Pt atoms hybridizing with the Fe atoms below and above – the system as a whole exhibits higher anisotropy than when is central layer consists of Fe atoms, whose orbital moment, in hybridizing with the Pt atoms above and below, is markedly lower. In contrast to the cuboctahedral case 18, 25 bi-pyramidal nanoparticles possess (similar magnitude) MCA mostly at the (large) central Pt layer. This fact may eliminate the need to cap them in order to preserve MCA. Our calculation show that five-layered nanoparticles with approximately 700 atoms can be expected to be useful in magnetic recording applications at room temperatures. Meanwhile, a deeper analysis of the electronic structure of these and other TM nano systems could contribute further to this end Generally speaking, the relation between the structure and MCA is not yet completely understood for FePt and other binary alloys. For example, as an alternative type of system, 147- atom cuboctahedral FePt clusters encapsulated into Cu-, Au-, and Al matrices were studied theoretically in Ref.25 where it was found, for example, that surface atoms have larger MCA. Another consideration is the particular role of electron-electron correlation in these systems. This was found to be important for small Fe61 and FePt62 clusters and invite further study. ACKNOWLEDGMENTS We would like to thank Lyman Baker for critical reading of the manuscript and numerous enlightening comments. The work was supported in part by DOE Grant DE-FG02-07ER46354. REC would like to thank UCF for partial sabbatical support. References A. T. Bell, Science 299, 1688 (2003). R. P. Andres, T. Bein, M. Dorogi, S. Feng, J. I. Henderson, C. P. Kubiak, W. Mahoney, R. G. Osifchin, and R. Reifenberger, Science 272, 1323 (1996). D. Davidović and M. Tinkham, Appl. Phys. Lett. 73, 3959 (1998). N. Jones, Nature 472, 22 (2011). J. Henderson, S. Shi, S. Cakmaktepe, and T. M. Crawford, Nanotechnology 23, 185304 (2012). T. Yogi, C. Tsang, T. A. Nguyen, K. Ju, G. L. Gorman, and G. Castillo, IEEE Trans. Magn. 26, 2271 (1990). J. S. Li, M. Mirzamaani, X. P. Bian, M. Doerner, S. L. Duan, K. Tang, M. Toney, T. Arnoldussen, and M. Madison, J Appl Phys 85, 4286 (1999). 26 1 2 3 4 5 6 7 L. Graf and A. Kussmann, Physik. Z 36, 544 (1935). C. Antoniak, et al., Phys. Rev. Lett. 97, 117201 (2006). R. P. Cowburn, J Appl Phys 93, 9310 (2003). J. Lyubina, B. Rellinghaus, O. Gutfleisch, and M. Albrecht, Handbook of Magnetic Materials, Vol. 19, edited by KHJ Buschow (Amsterdam: Elsevier, 2011). S. Sun, C. B. Murray, D. Weller, L. Folks, and A. Moser, Science 287, 1989 (2000). B. Azzerboni, G. Asti, and L. Pareti, Magnetic Nanostructures in Modern Technology: Spintronics, Magnetic MEMS and Recording (Springer, 2007). I. Watanabe, Y. Tanaka, E. Watanabe, and K. Hisatsune, J. Prosthet. Dent. 92, 278 (2004). S. Okamoto, N. Kikuchi, O. Kitakami, T. Miyazaki, Y. Shimada, and K. Fukamichi, Phys Rev B 66, 024413 (2002). J. U. Thiele, L. Folks, M. F. Toney, and D. K. Weller, J Appl Phys 84, 5686 (1998). T. Shima, T. Moriguchi, S. Mitani, and K. Takanashi, Appl. Phys. Lett. 80, 288 (2002). C. Antoniak, et al., Nature Commun. 2, 528 (2011). G. H. Daalderop, P. J. Kelly, and M. F. Schuurmans, Phys Rev B 44, 12054 (1991). I. V. Solovyev, P. H. Dederichs, and I. Mertig, Phys Rev B 52, 13419 (1995). A. Shick and O. Mryasov, Phys Rev B 67 (2003). C. Barreteau and D. Spanjaard, Journ. of Phys. Cond. Mat. 24, 406004 (2012). P. Blonski and J. Hafner, J. Phys. Cond. Matt. 21, 426001 (2009). L. Fernandez-Seivane and J. Ferrer, Phys. Rev. Lett. 99, 183401 (2007). M. E. Gruner, phys. Status Solidi A 210, 1282 (2013). P. Gambardella, et al., Science 300, 1130 (2003). S. Rohart, C. Raufast, L. Favre, E. Bernstein, E. Bonet, and V. Dupuis, Phys Rev B 74, 104408 (2006). J. Miyawaki, D. Matsumura, H. Abe, T. Ohtsuki, E. Sakai, K. Amemiya, and T. Ohta, Phys Rev B 80, 020408(R) (2009). X. Wang, R. Wu, D. Wang, and A. J. Freeman, Phys Rev B 54, 61 (1996). R. Wu and A. J. Freeman, Journ. of Magn. and Magn. Mat. 200, 498 (1999). G. Kresse and J. Hafner, Phys Rev B 47, 558 (1993). J. P. Perdew, J. A. Chevary, S. H. Vosko, K. A. Jackson, M. R. Pederson, D. J. Singh, and C. Fiolhais, Phys Rev B 46, 6671 (1992). S. H. Vosko, L. Wilk, and M. Nusair, Can. Jour. of Phys. 58, 1200 (1980). G. Kresse and D. Joubert, Phys Rev B 59, 1758 (1999). D. Hobbs, G. Kresse, and J. Hafner, Phys Rev B 62, 11556 (2000). M. Marsman and J. Hafner, Phys Rev B 66, 224409 (2002). J. Lyubina, I. Opahle, M. Richter, O. Gutfleisch, K.-H. Müller, L. Schultz, and O. Isnard, Appl. Phys. Lett. 89, 032506 (2006). J. G. Gay and R. Richter, Phys Rev Lett 56, 2728 (1986). J. N. W. Karas, and L. Fritsche, J. Chim. Phys. Phys.-Chim. Biol. 86, 861 (1989). J. Hu and R. Wu, Physical Review Letters 110, 097202 (2013). P. E. Blochl, Physical review. B, Condensed matter 50, 17953 (1994). M. P. A, Solid State Commun 35, 53 (1980). K. Boufala, L. Fernández-Seivane, J. Ferrer, and M. Samah, Journ. of Magn. and Magn. Mat. 322, 3428 (2010). A. R. Mackintosh and O. K. Andersen, Electrons at the Fermi Surface (Cambridge University Press, Cambridge, 1980). P. Bruno, Phys Rev B 39, 865 (1989). D. Fritsch, K. Koepernik, M. Richter, and H. Eschrig, J.Comput. Chem. 29, 2210 (2008). 27 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 L. Ferrnandez-Seivane and J. Ferrer, Phys. Rev. Lett. 99, 183401 (2007). M. Müller, P. Erhart, and K. Albe, Phys Rev B 76, 155412 (2007). T. T. Järvi, D. Pohl, K. Albe, B. Rellinghaus, L. Schultz, J. Fassbender, A. Kuronen, and K. Nordlund, EPL (Europhysics Letters) 85, 26001 (2009). T. Miyazaki, O. Kitakami, S. Okamoto, Y. Shimada, Z. Akase, Y. Murakami, D. Shindo, Y. K. Takahashi, and K. Hono, Phys Rev B 72, 144419 (2005). Y. K. Takahashi, T. Koyama, M. Ohnuma, T. Ohkubo, and K. Hono, J Appl Phys 95, 2690 (2004). R. M. Wang, O. Dmitrieva, M. Farle, G. Dumpich, H. Q. Ye, H. Poppa, R. Kilaas, and C. Kisielowski, Phys. Rev. Lett. 100, 017205 (2008). R. M. Wang, O. Dmitrieva, M. Farle, G. Dumpich, M. Acet, S. Mejia-Rosales, E. Perez-Tijerina, M. J. Yacaman, and C. Kisielowski, Journ. of Phys. Chem. C 113, 4395 (2009). C. Antoniak, et al., Journ. of Phys.-Cond. Mat. 21, 336002 (2009). O. Šipr, M. Košuth, and H. Ebert, Phys Rev B 70, 174423 (2004). C. Antoniak, M. Spasova, A. Trunova, K. Fauth, M. Farle, and H. Wende, Journ. of Phys.: Conference Series 190, 012118 (2009). L. Néel, Ann. Géophys 5, 99 (1949). S. Maenosono, T. Suzuki, and S. Saita, J Magn Magn Mater 320, 79 (2008). R. E. Camley and D. R. Tilley, Physical Review B 37, 3413 (1988). C.-b. Rong, D. Li, V. Nandwana, N. Poudyal, Y. Ding, Z. L. Wang, a. J. Hao Zeng, and Ping Liu, Adv. Mater. 18, 2984 (2006). V. Turkowski, A. Kabir, N. Nayyar, and T. S. Rahman, J. Phys. Cond. Mat. 22, 462202 (2010). V. Turkowski, A. Kabir, N. Nayyar, and T. S. Rahman, The Journ. of Chem. Phys. 136, 114108 (2012). 28 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 . 

1212.6129

1

1212

Ultrafast modulation of near-field heat transfer with tunable metamaterials

We propose a mechanism of active near-field heat transfer modulation relying on externally tunable metamaterials. A large modulation effect is observed and can be explained by the coupling of surface modes, which is dramatically varied in the presence of controllable magnetoelectric coupling in metamaterials. We finally discuss how a practical picosecond-scale thermal modulator can be made. This modulator allows manipulating nanoscale heat flux in an ultrafast and noncontact (by optical means) manner.

cond-mat.mes-hall

cond-mat

Ultrafast modulation of near-field heat transfer with tunable metamaterials Longji Cui, Yong Huang,a) Ju Wang, and Ke-Yong Zhu School of Aeronautic Science and Engineering, Beijing University of Aeronautics and Astronautics, Beijing 100191, China Abstract We propose a mechanism of active near-field heat transfer modulation relying on externally tunable metamaterials. A large modulation effect is observed and can be explained by the coupling of surface modes, which is dramatically varied in the presence of controllable magnetoelectric coupling in metamaterials. We finally discuss how a practical picosecond-scale thermal modulator can be made. This modulator allows manipulating nanoscale heat flux in an ultrafast and noncontact (by optical means) manner. 1 Radiative heat transfer between surfaces in close proximity has attracted much attention since it has been predicted that the heat flux can break the Planck’s blackbody radiation law.1 This effect, due to the tunneling of evanescent waves, is pronounced only in the near-field, i.e., at separation of surfaces smaller than the characteristic thermal wavelength. Particularly, enhancement that is several orders of magnitude over blackbody limit occurs if the heat transfer involves thermally excited surface modes,2-5 such as surface plasmon and phonon polaritons. Considerable theoretical effort has been devoted to understand the detailed mechanism of near-field heat transfer in a number of geometries6-12 and materials.13-19 Near-field thermal effect has been verified in recent experiments,20-23 and holds promise for applications such as imaging,24,25 energy conversion,26-28 and noncontact localized heating.29 Despite these progress, relatively little attention has been focused on how to actively control this transfer. This has deep implications for applications in photonic thermal circuits and thermal management in nanoelectronics. Since near-field heat flux is intimately related to optical properties of materials, one possibility is to modulate heat flow by using materials whose properties can be tuned by external stimulus. To this end, several schemes have been put forward by employing phase-change materials,30-32 graphene,33,34 anisotropic structures,35 as well as temperature dependent materials.36,37 However, limited by properties of the applied materials, existing schemes involve only tuning the dielectric response. The potential to control heat flux with tunable magnetic and magnetoelectric coupling response has not been explored yet. In this paper, we present an alternative scheme for controlling near-field heat flux. It shows that, by using the size and sign of magnetoelectric coupling in metamaterials as controllable parameters, 2 it is feasible to achieve a large heat flux modulation effect. Metamaterials are artificially structured materials possessing many unusual properties, such as negative refraction,38 magnetism up to optical frequencies,39 and giant magnetoelectric coupling (chiral metamaterials).40 These properties are gained from the unit-cells that replace atoms of natural materials as basic elements in light-matter interactions. Based on the flexibility of unit-cell design, tunable metamaterials are constructed by incorporating photo- or electric-active materials (e.g., semiconductors).41-43 This makes it possible to switch the electromagnetic responses through photoexcitation or external voltage. Specifically, recent progress44,45 has made metamaterials with tunable magnetoelectric coupling possible. More importantly, switching of different responses induced by photoexcitation can be achieved within very short time.41,46 Therefore, this metamaterial has great potential to make an ultrafast thermal modulator. Consider two chiral metamaterials (labeled 1 and 2), modeled as planar semi-infinite objects, separated by a vacuum gap of thickness d, and maintained in local thermodynamic equilibrium with temperature 1T and 2T . The radiative heat flux across the gap can be calculated by fluctuational electrodynamics introduced by Rytov.47 In this framework, both the far-field contribution and the near-field enhancement of heat transfer have been taken into account. In general, firstly the correlation function of stochastic current density is related to the system temperature by fluctuation dissipation theorem; then the fluctuational field is solved by Green’s dyadic technique; finally the heat flux is obtained from the statistically averaged Poynting vector. For the case of planar parallel bodies considered here, the net near-field heat flux q between 3 the two surfaces is given by19 ω ω d [ Θ T , ( 2 2π ∫ k ∫ 0 = × q ∞ < k ) − Θ ω T ( , 1 ) ] ⎧ ⎨ ⎩ ∫ k + > k 0 d k 2π 0 d k 2π ⋅ ⋅ * I R R D ⎡ k Tr ( - ) ⎣ 12 2 2 ⋅ ⋅ ⋅ * * I R R D ( - ) 1 12 1 ⎤ ⎦ , (1) ⎡ k Tr ( ⎣ ⋅ − * R R D ) 12 2 2 ⋅ ( ⋅ − * * R R D ) 1 12 1 − 2γ d e ⎤ ⎦ ⎫ ⎬ ⎭ where Θ ω T ( , ) = (cid:61) ω / (exp( ω (cid:61) / k T B − ) 1) is the average energy of Planck’s oscillator, 0k / cω= , k and = γ 2 k - k 0 2 are, respectively, parallel and normal component of vacuum wavevector. Tr is the trace operator, the symbol * denotes hermitian conjugation, and D 12 = ( ⋅ − I R R 1 e 2 − 2γ − 1 )d with R iR the reflection matrix defined as ω ω sp ss r r ⎤ ⎡ , k ) , k ) ( ( i i ⎥ ⎢ ω ω r r , k ) ( , k ) ( ⎢ ⎥ ⎣ ⎦ i i 1, 2) i ( = = pp ps i , (2) where ab ir gives the ratio of the reflected waves with b-polarization to the incident waves with a-polarization, and s (p) corresponds to transverse magnetic (electric) polarization. For isotropic media, the off-diagonal elements in Eq. (2) vanish and the diagonal elements reduce to the well-known Fresnel coefficient in terms of electric permittivity ε and magnetic permeability μ. However for chiral media, the constitutive relation has the form48 εε κ + E i H c / 0 and D = = B μμ κ − H i E c / 0 , where c = 1 / εμ 0 0 is the vacuum light speed, and κ is chirality parameter characterizing the magnetoelectric coupling response. In this case, both the diagonal and off-diagonal elements in the reflection matrices become non-zero due to magnetoelectric coupling effect. Assuming the incidence of the electromagnetic waves is from vacuum to chiral media, the reflection coefficients can be explicitly written as48 4 = − ss ir η η ξ ξ ηηξξ + − + 2 2 ( )( ) 2 ( + − + − 0 0 η η ξ ξ ηηξξ + + + 2 2 ) 2 ( )( ( + − + − 0 0 − 1) + , (3a) 1) = pp ir η η ξ ξ ηηξξ − − + 2 2 ( )( ) 2 ( + − + − 0 0 η η ξ ξ ηηξξ + + + 2 2 ) 2 ( )( ( + − + − 0 0 − + 1) 1) , (3b) sp r i = − ps r i = ηηξ ξ − i 2 ( ) + − 0 η η ξ ξ ηηξξ + + + 2 2 ) 2 ( )( ( + − + − 0 0 + , (3c) 1) where η μ ε = / 0 0 0 , η μμ εε = / i i i 0 0 , and ξ± = γ / ± γn ± in which ± = n εμ κ ± i i i and = γ ± 2 2 n k ± 0 − 2 k are, respectively, the refraction indices and normal wavevector components of the two circular polarized waves in chiral media. Now we proceed to demonstrate how near-field heat flux can be modulated by using the dynamic characteristic of magnetoelectric coupling in metamaterials. The constitutive parameters of chiral media with single resonance are written in a general form49 εω = − ) 1 ( ω Ω 2 ε ε , (4a) ω ω ω + ϒ − 2 2 i ε ε = − μω ) 1 ( Ω ω 2 μ + ϒ − ω ω 2 2 i μ , (4b) ω μ κω ( ) = ωω ±Ω κ κ ω ω + ϒ − 2 2 i κ , (4c) ω κ Where εω , μω , and κω denote the resonance frequencies of the dielectric, magnetic, and chirality responses; εϒ , μϒ , and κϒ are the damping factors; εΩ , μΩ , and κΩ account for the resonance strength of the permittivity, permeability, and chirality, respectively. For chiral metamaterials, the dielectric, magnetic, and chirality resonances generally have the same frequency, ω ω ω ω = = = . Moreover, the second law of thermodynamics requires that the κ μ ε 0 5 lossy part of the constitutive parameters must satisfy the inequality,50 κ 2 Im ( ) ≤ ε μ Im( ) Im( ) , therefore, there exists an upper limit for κΩ . For the convenience of numerical analysis, we assume the same damping factor, ϒ= ϒ = ϒ = ϒ , the limit value can thus be obtained by κ μ ε Ω = Ω Ω . substituting Eqs. (4a)-(4c) into the inequality, i.e., κ ε μ In what follows, two metamaterial designs with distinct tunability are considered: the design in Ref. 44 allowing switching chiral metamaterial between its initial handedness and the opposite handedness, and the design in Ref. 45 enabling continuous modulation of chirality. It should be noted that these metamaterials are designed specifically for electromagnetic responses in THz frequency region. Since the response frequency of a metamaterial scales inversely with its unit-cell geometrical dimension up to optical frequency (i.e., the size-scaling rule),51-53 it is expected that these designs can be miniaturized to possess responses in the infrared that is of interest for near-field radiative heat transfer. In principle, the size-scaling rule keeps valid as long as the metal in the unit-cell behaves as ideal metal. According to Ref. 53, the minimum dimension of the size-scaling regime for effective miniaturization of the unit-cell is approximately 55 nm. Here, it must be pointed out that since this minimum value is for a specific design of metamaterial (i.e., split ring resonator), a rigorous numerical simulation should be carried out to determine the separation at which the size-scaling rule breaks down for other metamaterial designs. We first consider a setup including two metamaterials with the constitutive parameters, 2ε ε= 1 , 2μ μ= 1 , 2κ κ= 1 (without control) and κ κ= − 1 2 (under control). To quantify the heat flux modulation induced by the chirality reversion, we define the relative difference of heat 6 transfer coefficients between [ ] = − q T / ( 2 T 1 ) the = two configurations as M ( − h O h C ) / h O , where h lim → = T T 2 T 1 , and the subscript O and C corresponds to the original and controlled state, respectively. Without loss of generality, the parameters are taken as ω = × 14 3 10 rad/s 0 , ϒ = 0.01ω 0 , 2εΩ = , μΩ = 0.5 , and T 2 − T 1 =(cid:19) T 2 300K . Figure 1(a) displays modulation M as a function of dimensionless separation distance d d λ≡ / 0 λ π ω ≡ ) /c (2 ( 0 0 is the resonance wavelength) for different sizes of chirality. It shows that modulation of heat transfer exists since 2κ κ= the heat transfer coefficient in the case of two identical metamaterials ( 1 ) is always larger than that in the case of opposite chirality ( κ κ= − 1 2 ). Moreover, the modulation favors configurations with strong chirality and the maximum modulation shown in Fig. 1(a) is found to be 0.7. In order to understand the physical origin of heat flux modulation, we examine the distribution of near-field heat flux in the (k , )ω plane, which is dictated by the integrand of Eq. (1). As shown in Fig. 1(b) and 1(c), for both configurations, heat flux is mainly transferred through channels in the vicinity of the surface mode resonance frequencies ( 1.08ω ω= 0 and 1.60ω ω= 0 for the system considered here). This indicates that when the two parallel surfaces are closely spaced in the near field, heat transfer is dominated by the coupling of surface modes at the resonance frequencies. Since the dispersion relation of the surface modes corresponds to the pole of the reflection coefficients in Eq. (3), the resonances can be found by evaluating the asymptotic behavior of these coefficients. Indeed, in the limit k k(cid:21) 0 , ssr ≈ ε μ κ σ − − + 2 [( ] / 1) 1)( , ppr ≈ ε μ κ σ + − − 2 1) [( ] / 1)( , and spr iκ σ ≈ − 2 / with 7 σ ε μ κ + − = + 2 1) ( 1)( . The surface mode resonances are therefore exhibited at frequencies where ε μ + + = 1) ( 1)( κ 2 . It should be noted here that the resonance frequencies are independent of the sign of κ. However, the coupling efficiency of the surface modes shows significant difference as sign of κ is reversed. For the configuration of opposite κ, we observe that the coupling of surface modes fades out at a cutoff value k c ≈ 40k 0 for 1.08ω ω= 0 , and k c ≈ 50k 0 for 1.60ω ω= 0 . In contrast, more coupling modes ( k c ≈ 50k 0 and 65k for two resonances, 0 respectively) and stronger coupling strength is found to contribute to heat transfer for the configuration of same handedness. It is clear from the above analysis that two discrete levels of near-field heat flux can be achieved by only switching the handedness of metamaterials, now we turn to investigate the heat flux modulation using dynamical metamaterials that allow tuning κ continuously. In Fig. 2(a), we plot the logarithmic contour of heat flux between two metamaterials with the constitutive ] [ Ω Ω ∈ − , 1, 1 κ κ 2 1 . It shows that at fixed separation distance parameters, , and 2ε ε= 1 , 2μ μ= 1 considered here ( d ≈ 0.01 ), different combinations of 1κΩ and 2κΩ lead to a wide range modulation of heat flux. The maximal heat flux occurs around Ω = Ω = κ κ 1 2 0.1 (P2 in Fig 2(a)) and is about 7% larger than the hear flux in the absence of magnetoelectric coupling in materials. It is also found that the maximal hear flux is more than one order larger than the minimal heat flux which occurs in the thermodynamic limit of the chiral media [P4, Ω = −Ω = 2 1.0 κ κ 1 ]. The pattern of heat flux modulation shown on the contour and the large heat transfer contrast can be attributed to the contribution of coupled surface modes in different configurations. We note that, 8 2κ κ= symmetrical or anti-symmetrical configurations ( 1 or κ κ= − ) tend to generate a large 1 2 heat flux between the surfaces. The underlying mechanism is that for such configurations, surface mode resonances are excited on both material surfaces at identical frequencies, and thus lead to more efficient coupling of surface modes at the resonances. On the contrary, for non-symmetrical configurations, surface modes on the two surfaces move out of resonance, resulting in weak coupling. As confirmed in Fig. 2(b), where the heat flux spectral variation for different points (from P1 to P4) is plotted, the heat transfer contrast in the figure is mainly due to the rapidly falling off of h at the surface mode resonance frequencies. In the following we discuss in more detail the possibility to make a practical heat flux modulator. Our theoretical predictions are based on effective medium approximation (EMA) that the metamaterial is regarded as homogeneous medium characterized by effective constitutive parameters shown in Eqs. (4a)-(4c). In fact, the EMA can be used as long as the wavelength of the important electromagnetic fluctuation, which is on the order of the separation distance between surfaces, is larger than the atomic dimension of the medium,54 i.e., d l> , where l is the unit-cell dimension. Moreover, one can also consider that the a nonlocal model is necessary if the parallel wave vectors of evanescent modes are on the order of / lπ . Since the exponential in the transmission coefficient in Eq. (1) sets a cut-off wave vector of evanescent modes contributing to near-field heat transfer which is ~1 / d , one finds that a local effective medium description is permissible for d l π> / . Hence, it can be seen that for the given unit-cell dimension considered here, the separation distances set at > d 60 nm can be viewed as appropriate. However, the EMA should be cautiously applied in small separation distances and a 9 precise minimum separation at which the EMA is invalid can only be obtained through numerical simulation for a concrete material. On the other hand, note that thermal modulation is pronounced at <(cid:4) d 0.01 λ , indicating 0 that the metamaterial should possess a large wavelength-to-structure ratio ( aλ > 0 / 100 ) to ensure a well defined EMA. In reality, some specific metamaterials with very compact inner structure (e.g., the Swiss roll design40 with 0 / aλ larger than a few hundred) can be good candidates for this purpose. However, it should also be noted that due to its compact structure, the suggested Swiss roll design cannot guarantee neither the electromagnetic response in the infrared nor a separation distance smaller than a hundred nanometers at which the EMA is valid. In reality, for practical near-field thermal application, the present scheme is still limited by current metamaterial design methods. In addition, the fabrication of complex metamaterials structures on micron/nanometer scale remains to be a challenging problem. To illustrate the potential of tunable metamaterial for fast heat flux modulation, we compare our scheme with the reported schemes using phase-change material30,31 and graphene-based structure.33 For photoexcitation controlled metamaterials, the tunability of the electromagnetic responses is accomplished by generating a high concentration of free charge carrier (and thus change the electric conductivity of the metamolecules) using an optical pulse. The pump fluence of the pulse enables to modulate the strength of the response. The switching time between different responses is determined by the carrier lifetime of the photo-active materials integrated into the metamolecules. It has been shown that by appropriately engineering the materials with short carrier lifetime, the switching time can be minimized to be as short as a few picoseconds.46 10 Thus, it is seen that an ultrafast thermal modulator can be made. Moreover, the heat flux control can be either continuous or discrete, depending on specific design of the metamaterials (as shown in our calculation). In comparison, heat flux modulation using the switching of two discrete states (crystal-amorphous or metal-insulator) of phase-change materials is on the order of a few nanoseconds, which is several orders of magnitude slower than the present scheme; instead the high carrier mobility of graphene that can be tuned by gating enables a high operating frequency up to a hundred gigahertz, comparable to the switching frequency of dynamic metamaterials. Furthermore, after a limited number of cycles, thermal modulator made with phase-change materials ceases to be reversible since the heating pulse causes incomplete state transition, and eventually leading to indistinguishable dielectric property between different phases.55 On the contrary, reversibility of the present scheme relies on the carrier recombination that can be completely recovered to the initial state transiently following the optical pulse. Therefore, this suggests a different mechanism for improving cycle endurance of thermal modulator. In conclusion we show that modulation of near-field heat transfer can be realized by means of metamaterials which can be externally tuned for obtaining desired electromagnetic responses. We show that, by using the size and sign of magnetoelectric coupling response as controllable parameters, a large heat flux modulation effect is feasible. It must be pointed out that our purpose of this Letter is to reveal the possibility of active nanoscale thermal modulation that can be operated ultrafast and by optical means, rather than give a complete account. Therefore, there exist plenty of other possibilities for obtaining noncontact and ultrafast thermal modulation based on the concept proposed here, such as using optically controlled semiconductors and even 11 conceiving of the schemes working effectively in the ultra-near-field regime where the EMA breaks down. Along with the present scheme, we believe novel metamaterials that are specifically designed for near-field thermal application are necessary. On the other hand, for simplicity we consider the simple parallel planes and ideal metamaterial responses, further practical realization of this scheme can be extend to more general configurations and complex metamaterial properties, which can benefit from the great flexibility in engineering geometrical and optical properties of metamaterials. 12 Acknowledgements Y. H. gratefully acknowledges support from the Fundamental Research Funds for the Central Universities (No.YWF-11-03-Q-028). 13 a)Electronic mail: huangy@buaa.edu.cn References 1D. Polder and M. V. Hove, Phys. Rev. B 4, 3303 (1971). 2K. Joulain, J.-P. Mulet, F. Marquier, R. Carminati, and J.-J. Greffet, Surf. Sci. Rep. 57, 59 (2005). 3A. I. Volokitin and B. N. J. Persson, Rev. Mod. Phys. 79, 1291 (2007). 4S.-A. Biehs, E. Rousseau, and J.-J. Greffet, Phys. Rev. Lett. 105, 234301 (2010). 5P. Ben-Abdallah and K. Joulain, Phys. Rev. B 82, 121419 (2010). 6G. Domingues, S. Volz, K. Joulain, and J.-J. Greffet, Phys. Rev. Lett. 94, 085901 (2005). 7S.-A. Biehs, Eur. Phys. J. B 58, 423 (2007). 8A. Narayanaswamy and G. Chen, Phys. Rev. B 77, 075125 (2008). 9P. Ben-Abdallah, K. Joulain, J. Drevillon, and G. Domingues, J. Appl. Phys. 106, 044306 (2009). 10M. Krüger, Th. Emig, and M. Kardar, Phys. Rev. Lett. 106, 210404 (2011). 11C. Otey and S. Fan, Phys. Rev. B 84, 245431 (2011). 12P. Ben-Abdallah, S.-A. Biehs, and K. Joulain, Phys. Rev. Lett. 107, 114301 (2011). 13C. J. Fu and Z. M. Zhang, Int. J. Heat Mass Transfer 49, 1703 (2006). 14P. Ben-Abdallah, K. Joulain, and A. Pryamikov, Appl. Phys. Lett. 96, 143117 (2010). 15K. Joulain, J. Drevillon, and P. Ben-Abdallah, Phys. Rev. B 81, 165119 (2010). 16S.-A. Biehs, P. Ben-Abdallah, F. S. S. Rosa, K. Joulain, and J.-J. Greffet, Opt. Express, 19, 14 A1088 (2011). 17S.-A. Biehs, M. Tschikin, and P. Ben-Abdallah, Phys. Rev. Lett 109, 104301 (2012). 18Y. Guo, C. L. Cortes, S. Molesky, and Z. Jacob, Appl. Phys. Lett. 101, 131106 (2012). 19L. Cui, Y. Huang, and J. Wang, J. Appl. Phys. 112, 084309 (2012). 20A. Narayanaswamy, S. Shen, and G. Chen, Phys. Rev. B 78, 115303 (2008). 21E. Rousseau, A. Siria, G. Jourdan, S. Volz, F. Comin, J. Chevrier, and J.-J. Greffet, Nat. Photonics 3, 514 (2009). 22R. S. Ottens, V. Quetschke, S. Wise, A. A. Alemi, R. Lundock, G. Mueller, D. H. Reitze, D. B. Tanner, and B. F. Whiting, Phys. Rev. Lett. 107, 014301 (2011). 23S. Shen, A. Mavrokefalos, P. Sambegoro, and G. Chen, Appl. Phys. Lett. 100, 233114 (2012). 24Y. De Wilde, F. Formanek, R. Carminati, B. Gralak, P.-A. Lemoine, K. Joulain, J.-P. Mulet, Y. Chen, and J.-J. Greffet, Nature 444, 740 (2006). 25A. Kittel, U. F. Wischnath, J. Welker, O. Huth, F. Rüting, and S.-A. Biehs, Appl. Phys. Lett. 93, 193109 (2008). 26A. Narayanaswamy and G. Chen, Appl. Phys. Lett. 82, 3544 (2003). 27K. Park, S. Basu, W. P. King, and Z. M. Zhang, J. Quant. Spectrosc. Radiat. Transf. 109, 305 (2008). 28M. Francoeur, R. Vaillon, and M. P. Mengüc¸, IEEE Trans. Energy Convers. 26(2), 686 (2011). 29B. J. Lee, Y.-B. Chen, and Z.M. Zhang, J. Quant. Spectrosc. Radiat. Transf. 109, 608 (2008). 30P. J. van Zwol, K. Joulain, P. Ben Abdallah, J. J. Greffet, and J. Chevrier, Phys. Rev. B 83, 201404 (2011). 15 31P. J. van Zwol, K. Joulain, P. Ben-Abdallah, and J. Chevrier, Phys. Rev. B 84, 161413 (2011). 32P. J. van Zwol, L. Ranno, and J. Chevrier, Phys. Rev. Lett. 108, 234301 (2012). 33V. B. Svetovoy, P. J. van Zwol, and J. Chevrier, Phys. Rev. B 85, 155418 (2012). 34O. Ilic, M. Jablan, J. D. Joannopoulos, I. Celanovic, H. Buljan, and M. Soljacic, Phys. Rev. B 85, 155422 (2012). 35S.-A. Biehs, P. Ben-Abdallah, and F. S. S. Rosa, Appl. Phys. Lett. 98, 243102 (2011). 36C. R. Otey, W. T. Lau, and S. Fan, Phys. Rev. Lett. 104, 154301 (2010). 37S. Basu and M. Francoeur, Appl. Phys. Lett. 98, 113106 (2011). 38R. A. Shelby, D. R. Smith, and S. Schultz, Science 292, 77 (2001). 39C. Enkrich, M. Wegener, S. Linden, S. Burger, L. Zschiedrich, F. Schmidt, J. F. Zhou, Th. Koschny, and C. M. Soukoulis, Phys. Rev. Lett. 95, 203901 (2005). 40J. B. Pendry, Science 306, 1353 (2004). 41W. J. Padilla, A. J. Taylor, C. Highstrete, M. Lee, and R. D. Averitt, Phys. Rev. Lett. 96, 107401 (2006). 42H.-T. Chen, W. J. Padilla, J. M. O. Zide, A. C. Gossard, A. J. Taylor, and R. D. Averitt, Nature 444, 597 (2006). 43I. V. Shadrivov, P. V. Kapitanova, S. I. Maslovski, and Y. S. Kivshar, Phys. Rev. Lett. 109, 083902 (2012). 44S. Zhang, J. Zhou, Y.-S. Park, J. Rho, R. Singh, S. Nam, A. K. Azad, H.-T. Chen, X. Yin, A. J. Taylor, and X. Zhang, Nature Commun. 3, 942 (2012). 45J. Zhou, D. R. Chowdhury, R. Zhao, A. K. Azad, H.-T. Chen, C. M. Soukoulis, A. J. Taylor, 16 and J. F. O’Hara, Phys. Rev. B 86, 035448 (2012). 46H.-T. Chen, W. J. Padilla, J. M. O. Zide, S. R. Bank, A. C. Gossard, A. J. Taylor, and R. D. Averitt, Opt. Lett. 32, 1620 (2007). 47S. M. Rytov, Y. A. Kravtsov, and V. I. Tatarskii, Principles of Statistical Radiophysics (Springer-Verlag, Berlin, 1989), Vol. 3. 48A. Lakhtakia, Beltrami Fields in Chiral Media (World Scientific, Singapore, 1994). 49R. Zhao, T. Koschny, and C. M. Soukoulis, Opt. Express 18, 14553 (2010). 50I. V. Lindell, A. H. Shivola, S. A. Tretyakov, and A. J. Viitanen, Electromagnetic Waves in Chiral and Bi-Isotropic Media (Artech House, Boston, 1994). 51J. B. Pendry, A. J. Holden, W. J. Stewart, and I. Youngs, Phys. Rev. Lett. 76, 4773 (1996). 52J. Pendry, A. J. Holden, D. J. Robbins, and W. J. Stewart, IEEE Trans. Microwave Theory Tech. 47, 2075 (1999). 53J. Zhou, Th. Koschny, M. Kafesaki, E. N. Economou, J. B. Pendry, and C. M. Soukoulis, Phys. Rev. Lett. 395, 223902 (2005). 54E. M. Lifshitz and L. P. Pitaevskii, Statistical Physics, Part 2 (Pergmon Press, Oxford, 1980). 55M. H. R. Lankhorst, B.W. S.M.M.Ketelaars, and R.A.M.Wolters, Nat. Mater. 4, 347 (2005). 17 Figure Caption FIG. 1. (Color online) (a) Near-field heat flux modulation effect due to chirality reversion, defined as relative difference of heat transfer coefficients between two chiral metamaterials with 2κ κ= same handedness ( 1 κ κ= − ) and opposite handedness ( 1 2 ), as a function of separation distance. The contour plot (a.u.) of the integrand of Eq. (1) in the (k , )ω plane is displayed for (b) the configuration of same handedness, and (c) the configuration of opposite handedness, at κΩ = 0.9 and d λ = / 0 0.01 . FIG. 2. (Color online) (a) Logarithmic contour plot of heat flux between two tunable ]1, 1− metamaterials with κΩ individually varying in the interval [ 2κΩ are marked from P1 to P4 (P1 is the case with no 1κΩ and combinations of d λ = / 0 at 0.01 . Different magnetoelectric coupling; P2 and P4 denotes the maximal and minimal heat flux respectively). (b) Spectral variation of near-field heat transfer coefficient for P1 to P4 marked in (a). 18 

1502.06005

2

1502

Quasi-ballistic phonon transport effects on the determination of the mean-free path accumulation function for the effective thermal conductivity

The mean-free path (MFP) accumulation function for the effective thermal conductivity, introduced by Dames and Chen is a compact, universal and highly useful summary of the effect of ballistic thermal transport on the effective thermal conductivity measured by various experiments. The frequency domain thermoreflectance (FDTR) experiment is especially well suited to its determination. Extraction of the accumulation function from this experiment uses the thermal penetration depth (TPD) of phonons at each frequency as a cut-off for classifying phonons as ballistic or diffusive at that frequency. In this paper, we show that using the TPD as a cut-off is arbitrary and prone to serious error. We report on a new technique to deduce the MFP accumulation function from the FDTR experiment by numerical solution of an enhanced Fourier law.

cond-mat.mes-hall

cond-mat

Quasi-ballistic phonon transport effects on the determination of the mean-free path accumulation function for the effective thermal conductivity Ashok T. Ramu† and John E. Bowers University of California Santa Barbara †Corresponding author: ashok.ramu@gmail.com The mean-free path (MFP) accumulation function for the effective thermal conductivity, introduced by Dames and Chen is a compact, universal and highly useful summary of the effect of ballistic thermal transport on the effective thermal conductivity measured by various experiments. The frequency domain thermoreflectance (FDTR) experiment is especially well suited to its determination. Extraction of the accumulation function from this experiment uses the thermal penetration depth (TPD) of phonons at each frequency as a cut-off for classifying phonons as ballistic or diffusive at that frequency. In this paper, we show that using the TPD as a cut-off is arbitrary and prone to serious error. We report on a new technique to deduce the MFP accumulation function from the FDTR experiment by numerical solution of an enhanced Fourier law. 1. Introduction The mean-free path accumulation function (MFPAF) introduced by Dames and Chen [1] is a powerful tool in studying ballistic phonon transport. The MFPAF at a given mean-free path Λ is defined as the effective thermal conductivity (𝜅𝑒𝑓𝑓) contributed by all phonons with mean-free paths less than or equal to Λ. The utility of the MFPAF lies in that it explains within a unified framework[2] diverse experiments, like the transient gratings[3], time-domain thermoreflectance (TDTR)[4][5] and frequency domain thermoreflectance (FDTR)[6] experiments, that probe heat transport on length scales comparable to the phonon mean-free path. For bulk materials, measurements of the MFPAF have been conducted for crystalline[7] and amorphous[8] materials. For nanostructured materials, Yang and Dames [9] have given a relationship to the MFPAF of bulk materials. The MFPAF of nanostructured materials has been determined by measuring length- dependent conductivity in nanowires[10]. The MFPAF has been applied to the determination of thermal properties of nanostructured materials from bulk properties [11]. The concept of MFPAF has been applied to the study of thermal interfaces as well [12]. The thermoreflectance class of measurements was invented by Paddock and Easley [13]. The FDTR experiment is especially well suited to the determination of the MFPAF. In the FDTR experiment, a sinusoidally modulated “pump” laser beam impinges on the sample, and its modulation frequency is varied across a wide range. The resulting surface temperature is monitored by means of a weak “probe” beam reflected by a thin metal transducer layer that coats the sample, and whose reflectivity is sensitive to the temperature fluctuations caused by the pump. As the modulation frequency ν increases, the thermal penetration depth (TPD), which is the depth at which the temperature oscillation amplitude equals 1/e times the surface value, falls as √1/ν in the weakly quasi-ballistic limit of phonon transport [20]. This limit is valid only when the modulation frequency is much lesser than the inverse lifetime of the dominant heat carriers. The TPD is a measure of how deeply the thermal wave penetrates the sample. Phonons with mean-free paths much greater than the TPD are supposed to fail to equilibrate with the 1 lattice within a depth equal to the TPD, and are therefore presumed lost to the measurement. Thus the “effective” thermal conductivity (𝜅𝑒𝑓𝑓) decreases from its bulk value as the modulation frequency increases [6]. Throughout this paper we focus on silicon bulk substrates with a fixed and somewhat arbitrary, but reasonable boundary thermal conductance of 200 MW/m2 -K between the metal transducer and the silicon. This value is in concord with the fitting of Regner et al. to the high-frequency (~100 MHz) FDTR data [6]. The results of this paper, especially the newly extracted MFPAF (Fig. 6), are particularly robust to boundary thermal conductances between 50-1000 MW/m2-K. Silicon has a broad MFPAF with phonons of almost 3 orders of magnitudes of MFPs contributing non-trivially to thermal transport [14]. In this paper, we show that the choice of the TPD as the cutoff between ballistic and diffusive modes is intrinsically arbitrary and leads to significant error in the extracted MFPAF, especially in the low MFP regime. Arbitrariness may be due to two reasons: the choice of the 1/e depth as opposed to, say, the 1/e2 depth is one source. This is just a matter of choice and we will not focus on this here. The other and more serious source of error lies in that, as will be shown below, phonons with MFP equal to the TPD contribute significantly (roughly 42% at 88 MHz in silicon at 300 K) to the net heat-flux. Clearly, ignoring phonon modes with this MFP as fully ballistic (instead of quasi-ballistic) leads to erroneous conclusions about the MFPAF. This paper is organized as follows: First, we briefly introduce the theoretical model we use to derive our results and state its main features. Next, we describe the methodology we use to derive the cut-off for the MFP of quasi-ballistic phonons. In doing so, we introduce expressly for the first time the concept of a frequency-dependent MFP accumulation function. We follow by examining in detail the effect of cutting off the diffuse modes at the TPD (hereafter called the TPD-cutoff model). We then apply our theoretical model and show that it yields cutoffs far from the TPD, giving us a very different MFPAF. We conclude by summarizing our findings. 2. Theoretical model Our theoretical analysis is based on the enhanced Fourier law (EFL) [15] introduced by one of the authors. The basis of our model of thermal transport is the “two-fluid” assumption[16], wherein the phonon spectrum is divided into two parts- one a high-heat-capacity, high-frequency (HF) part that is in quasi- thermal equilibrium with a well-defined local temperature, and the other, a low-frequency (LF), low-heat- capacity part that is farther out of equilibrium. The LF modes do not interact with each other due to the small phase-space for such scattering[17], but can exchange energy with the HF modes. This model accounts for the effect of low-frequency phonon modes of long mean-free path, that propagate concomitantly to the dominant high-frequency modes. The cutoff for classification of modes as HF and LF will be the subject of much discussion in this paper. The theory of the EFL is based on spherical harmonic expansions of the phonon distribution functions, wherein the high-frequency mode distribution function is truncated at the first order in the expansion, while the low-frequency mode distribution function, which is farther out of thermal equilibrium, is truncated at the second order. This procedure has the advantage that successive terms of the spherical harmonic expansion may be directly related to quantities of physical interest like the energy-density and heat-flux. The EFL, which may be viewed as a non-local refinement of the Fourier law, is written as 2 𝜕𝑞 𝜕𝑥 = −𝐶𝑣 𝜕𝑇 𝜕𝑡 + 𝑆𝐻𝐹(𝑥, 𝑡) 𝑞 = 3 5 (𝛬𝐿𝐹)2 𝜕2𝑞 𝜕𝑥2 + 3 5 𝜅𝐻𝐹(𝛬𝐿𝐹)2 𝜕3𝑇 𝜕𝑥3 − 𝜅 𝜕𝑇 𝜕𝑥 (1) (2) Here, the net heat-flux 𝑞= 𝑞𝐿𝐹+𝑞𝐻𝐹,𝑞𝐿𝐹= the LF-mode contribution to the heat-flux, 𝑞𝐻𝐹= the HF-mode contribution to the heat-flux; 𝐶𝑣 is the volumetric heat capacity; 𝑆𝐻𝐹(𝑥, 𝑡)= external heat source term; 𝑇 = local temperature of HF modes; 𝜅 is the net bulk thermal conductivity; 𝜅 = 𝜅𝐿𝐹 + 𝜅𝐻𝐹, 𝜅𝐻𝐹= the contribution of HF modes to the bulk thermal conductivity, 𝜅𝐿𝐹= the contribution of LF modes to the bulk thermal conductivity; and 𝛬𝐿𝐹= the MFP of the LF modes =𝑣𝜏 where 𝑣 is the group-velocity magnitude of all LF modes and 𝜏 = LF mode lifetime. We assume that each and every LF mode has the same lifetime 𝜏, as well as the same group-velocity magnitude 𝑣. The frequency of an LF mode of wave-vector 𝒌, denoted by 𝜔(𝑘) is permitted to vary with 𝑘. We also assume isotropic phonon dispersion. For future reference, we state equations for the LF mode and HF mode heat-fluxes 𝑞𝐻𝐹 and 𝑞𝐿𝐹separately: 𝑞𝐿𝐹 = 3 5 (𝛬𝐿𝐹)2 𝜕2𝑞𝐿𝐹 𝜕𝑥2 − 𝜅𝐿𝐹 𝜕𝑇 𝜕𝑥 𝑞𝐻𝐹 = −𝜅𝐻𝐹 𝜕𝑇 𝜕𝑥 (3) (4) It has been shown in an earlier article that truncation of the LF mode distribution function at the second order in the spherical harmonic approximation is a good approximation[19] and that an excellent match with the transient gratings experiment in silicon at 300 K is obtained with an LF phonon MFP of Si (400 nm) that agrees well with Ref. [20], thus validating our model. These equations are derived from the model assumptions in Appendix A. We adapt the EFL to the conditions of the FDTR experiment by assigning a time-dependence of the form eiωt to all variables, where ω is the modulation (angular) frequency, and then by working entirely in the Fourier domain. In that case, for the substrate, where there is no heat source, substituting Eq. (1) into the derivative (with respect to x) of Eq. (2), we obtain a fourth-order linear homogeneous differential equation for (complex) T. Its four characteristic roots were found numerically using MATLAB®. Assuming an infinite substrate with all heat-fluxes and temperatures set to 0 at the bottom, we may drop the two exponentially growing solutions. Thus there remain two constants of integration. These constants are determined using continuity of heat-flux between the transducer and substrate: we set 𝑞𝐿𝐹=0 and 𝑞𝐻𝐹 = 𝑄 (Eqs. (3), (4)) at the silicon surface, where 𝑄 is the net heat-flux emanating from the transducer; thus the surface temperature of silicon in the absence of an interface may be found. The thermal interface is accounted for by a temperature drop ΔT=Q/G where Q is the heat-flux and G is the boundary thermal conductance. Finally 𝑄 is related to the input from the “pump” laser by solving the Fourier law in the tranducer film, with heat-capacity 1.6X106 W/m3-K, thermal conductivity 100 W/m- K and thickness 100 nm as appropriate for an Au transducer [6]. The phase of the temperature is of interest since it is robust to fluctuations in laser power[21], and is calculated as the arctangent of the ratio of the imaginary to the real part of the complex temperature. The only subtlety here is that we assume all of the heat emitted by the “pump” laser into the transducer transfers exclusively to the HF modes of silicon. This is because heat from the laser scatters into final states in silicon in proportion to their density-of-state, according to the Fermi golden rule [22]. Since the density- 3 of-state for acoustic phonons varies approximately as the square of the frequency[24], HF modes are excited much more efficiently than LF modes by external heat sources. 3. Methodology We consider in Fig. 1 the transducer phase as a function of logarithmic frequency. This function is so designed as to replicate, for hypothetically large “pump” laser spot diameters (≥ 100 micrometer), the Fig. 4 of Regner et al. [6] for the MFPAF of silicon at room temperature, using the TPD-cutoff model. In other words, the phase of Fig. 1 at each frequency is fitted to the usual Fourier law with an effective conductivity 𝜅𝑒𝑓𝑓, and the corresponding frequency 𝜔 is translated into an MFP by setting the MFP equal to the TPD, √2𝜅𝑒𝑓𝑓/𝜔𝐶𝑣. Here 𝐶𝑣 is the volumetric heat capacity of silicon. We thus recover the MFPAF, Fig. 4 of Regner et al., from our assumed phase data. We choose not to use the experimental phase data of Regner et al. because their small spot size (~ 3.4 microns) would cause significant radial ballistic effects[22] which would complicate the subsequent analysis. Fig. 4 of Ref. [6] is replotted in Fig. 6 (red solid curve). Our main point in this paper is that when we analyze the same phase data (Fig. 1) using the enhanced Fourier law [15] instead, we recover an accumulation function very different from Fig. 4 of Regner et al. (Ref. [6]). To this end, at each frequency, we find the MFP ΛLF of the LF modes by iteratively varying it, using the MATLAB® routine “fsolve” until the heat-flux in the HF modes equals the heat-flux in the LF modes in magnitude (this criterion is discussed later). The other parameters are determined as follows: in Eqs. (1) and (2), at each frequency, we take 𝜅𝐻𝐹 as 𝜅𝑒𝑓𝑓 at that frequency, 𝜅𝑒𝑓𝑓 being as before the effective conducitivity according to the usual Fourier law. This is from the definition of the MFPAF and means that we leave y-coordinates of the MFPAF unchanged. Also, 𝜅𝐿𝐹 = (𝜅𝑏𝑢𝑙𝑘 − 𝜅𝑒𝑓𝑓) where 𝜅𝑏𝑢𝑙𝑘 is the bulk conductivity of silicon, constant at 143 W/m-K. This is from the definition of 𝜅𝐿𝐹 in the EFL. The laser power input into the transducer is assumed to be 106/t W/m3 where t is the transducer thickness in m. In order to differentiate our methodology from that of the TPD-cutoff model, we introduce here the concept of a “modulation frequency-dependent MFP accumulation function”. We note that at each frequency, the TPD-cutoff model implicitly uses this concept; its frequency dependent MFPAF consists of simply rejecting the contributions of all modes with MFP ≥ TPD. Our model of the frequency dependent MFPAF whereas consists of a cut-off MFP ΛLF, not pre-set at any value, but solved for from the detailed thermal properties of the substrate. We assume that modes with MFP equal to the cut-off MFP contribute quasi-ballistically and constitute the 𝑞𝐿𝐹 of Eq. (3). Fig. 2 shows a schematic of the difference between the two models. While assigning the same MFP (ΛLF) to all modes above the cutoff might seem like an oversimplification considering the broad MFP distribution of silicon, we note that for a specific frequency of modulation, most of the phonons with MFPs above the cut-off ΛLF are lost to the measurement, and hence their contribution to the bulk conductivity becomes irrelevant. At any rate, our model is superior to the TPD- cutoff model of Koh and Cahill [4], that entirely ignores quasi-ballistic conduction by phonons with MFPs at and above the TPD. 4 We set the criterion for the cut-off as the mean-free path at which the HF-mode heat-flux contribution (magnitude of 𝑞𝐻𝐹 of Eq. (4)) drops to the point where it equals the LF-mode contribution (magnitude of 𝑞𝐿𝐹 of Eq. (3)). At all frequencies, it is observed that this criterion ensures that the LF-mode heat-flux from phonons of MFP equal to this cut-off, and therefore from phonons of MFP greater than this cut-off, is negligible compared to the net heat-flux. Fig. 1: Phase of the transducer temperature oscillation as a function of (linear) frequency. The phase function is designed to replicate the MFP data of Fig. 4 of Regner et al. [6], shown in this paper as the red solid curve of Fig. 6. 5 Fig. 2: A schematic representation of the modulation frequency-dependent MFPAF at a modulation frequency of 88 MHz for (a) our EFL-based model, and (b) the TPD-cutoff model. Our model accounts for quasi-ballistic transport above the MFP cutoffs (abscissae of the vertical lines). 6 4. Results and discussion We plot in Fig. 3 the heat-fluxes vs depth under the TPD-cutoff model. It is seen that LF-channel phonons with an MFP of 376 nm contribute 42% of the net heat-flux at 88 MHz, when measured at a depth of 376 nm (equal to the TPD at that frequency) from the surface. Also, the LF-mode heat-flux at that depth is about 25% of the surface heat-flux (8X105 W/m2). In Fig. 4, using the EFL-based model of this work, with the cut-off ΛLF set at 1.47 micrometers this contribution is reduced to 6% of the net heat flux at the same depth (376 nm). Now we may safely say that phonons with MFP greater than the cutoff contribute negligibly to the measurement. Fig. 5 shows the cut-off MFP vs frequency for both this work and the TPD-cutoff model. It is seen that our cut-off is dramatically larger than the cut-off based on the TPD. Also, our model is not a simple scaling of the TPD-cutoff by a constant factor. At low frequencies, our cut-off somewhat resembles 8 times the TPD, but even this connection breaks down at higher frequencies by being a factor of 2 too great. We further note that the quasi-ballistic heat-flux 𝑞𝐿𝐹 is rather insensitive to the mean-free path cut-off ΛLF at low- frequencies. Thus it may be seen that a slightly different cut-off criterion might give quite different values of the MFP cut-off. This reflects, in our opinion, a fundamental inability to determine the contribution of high MFP phonons to the accumulation function from FDTR measurements. This is because they carry too little heat, e.g. 5% of the bulk value at our lowest frequency (0.4 MHz). This low value is in turn because of the low heat-capacity of these modes [24]. Fig. 6 compares the MFPAF derived from our model to that from the TPD-cutoff model. Our model clearly shifts the MFPAF derived from the TPD cut-off model to the right. This is to be expected, since quasi- ballistic phonon heat-fluxes which decay slower than diffusive fluxes are accounted for in our model. We note here that unlike Regner et al.[6] whose MFPAF extraction is based on the TPD model, our MFPAF is inconsistent with ab-initio calculations of Esfarjani et al. [14]. Further exploration of this discrepancy is strongly indicated. It must be emphasized that Fig. 6 should not be viewed as the “correct” and “wrong” MFPAF for silicon – it simply indicates that the MFPAF as given by the TPD-cutoff model is inconsistent with the experiment from which it is extracted. 7  Fig. 3: Phonons with MFPs above the TPD contribute as much as 42% to the net heat-flux at 88 MHz. Their neglect results in serious error in the MFPAF. Dashed line shows “effective” Fourier law result, which is less than the EFL net heat-flux because it neglects quasi-ballistic phonon transport. Fig. 4: 𝛬𝐿𝐹 in Eq. (2) is varied iteratively until the LF mode heat-flux equals the HF-mode heat-flux, and the resulting value (1.465 microns, as shown) is taken as the actual cutoff. Then we find that only 6% of the net heat-flux at a depth of 376 nm (see Fig. 1) comes from LF modes. Also, HF mode heat-flux agrees well with “effective” Fourier law (dashed line). 8 Fig. 5: The cutoff MFP is 4X TBD at high frequencies, and rises to about 9X TPD at low frequencies. Green curve shows fit of EFL-based cut-off with 8X TPD; error is large at high frequencies (by a factor of 2). Fig. 6: The mean-free path accumulation function; red solid curve is after Regner et al. [6]. Our values (blue solid curve) are shifted considerably to the right due to contributions from quasi-ballistic LF-mode phonons. 5. Conclusions We have utilized the enhanced Fourier law proposed earlier by one of the authors to develop a novel framework for evaluating the mean-free path accumulation function for effective thermal conductivity from Fourier domain thermoreflectance experiments. We have seen that cutting off the quasi-ballistic modes at the thermal penetration depth at each frequency neglects their important contribution to the heat-flux, especially at high frequencies. Our more rigorous criterion has resulted in mean-free path cutoffs much larger than the thermal penetration depth. This has changed drastically the accumulation function, a highly important quantity for thermal transport theory and measurements. The current limitation of our framework is that it is in one dimension, while practical FDTR experiments require a two- dimensional enhanced Fourier law for their analyses due to small pump-beam spot sizes (3-15 microns). 9 Appendix A We derive Eqs. (1)-(4) from the model assumptions stated in Sec. 2 (Theoretical Model). The derivation follows [15] closely. We denote the distribution function for LF modes as 𝑔(𝑥, 𝒌), where all spatial variation is assumed to be along the x-direction, and 𝒌 is the phonon mode wave-vector of magnitude 𝑘 and making an angle 𝜃 with the x-axis. We expand the distribution function in terms of spherical harmonics 𝑃𝑙(𝑐𝑜𝑠𝜃). Since spherical harmonics are eigen-functions of the Legendre differential equation which is a Sturm-Liouville equation, they form an orthogonal basis for expanding angle-dependent azimuthally symmetric functions [26]: 𝑔(𝑥, 𝒌) = ∑ 𝑔𝑙(𝑥, 𝑘)𝑃𝑙(𝑐𝑜𝑠𝜃) ∞ 𝑙=0 (5) The steady-state linearized BTE for the LF modes is given by 𝑣𝑐𝑜𝑠𝜃 𝜕𝑔(𝑥,𝒌) 𝜕𝑥 = − 𝑔(𝑥,𝒌)−𝑓𝐸𝑞(𝑥,𝑘,𝑇) 𝜏 (6) We begin with the observation that, owing to the orthogonality of the spherical harmonics, the x- component of the LF heat-flux is determined solely by the first spherical harmonic 𝑔1: 𝑞𝐿𝐹 = 2𝜋 ∑ ∫ 𝑘 𝜋 𝜃=0 ℏ𝜔𝑔(𝑥, 𝒌)𝑣𝑐𝑜𝑠𝜃𝑠𝑖𝑛𝜃𝑑𝜃 = 4𝜋 3 ∑ ℏ𝜔𝑣𝑔1 𝑘 (7) 1 Therefore we seek a differential equation for 𝑔1. Here and henceforth, ∑ 𝐼(𝑘) is shorthand for (2𝜋)3 ∫ 𝑑𝑘𝐼(𝑘)𝑘2 where 𝐼(𝑘) is any function of k, and the integral is over all LF mode wave-vector magnitudes. Substituting Eq. (1) into the Boltzmann transport equation, Eq. (2), multiplying successively by 𝑃𝑙′(𝑐𝑜𝑠𝜃)𝑠𝑖𝑛𝜃 for 𝑙′ = 0,1,2, … and integrating over 𝜃, we arrive at a hierarchy of coupled equations for the 𝑔𝑙s, [27], the first three of which are 𝑘 1 3 2 5 3 7 𝑣 𝜕𝑔1 𝜕𝑥 + 𝑔0−𝑓𝐸𝑞(𝑇) 𝜏 = 0 𝑣 𝜕𝑔2 𝜕𝑥 + 𝑣 𝜕𝑔0 𝜕𝑥 + 𝑔1 𝜏 = 0 𝑣 𝜕𝑔3 𝜕𝑥 + 2 3 𝑣 𝜕𝑔1 𝜕𝑥 + 𝑔2 𝜏 = 0 (8a) (8b) (8c) We truncate the hierarchy by setting 𝑔3 = 0; other truncations are possible [27]. Substituting Eq. (8c) into Eq. (8b) to eliminate 𝑔2, and the result into Eq. (8a) to eliminate 𝑔0, we arrive at an equation expressed solely in terms of 𝑔1: − 3 5 (𝑣𝜏)2 𝜕2𝑔1 𝜕𝑥2 + 𝑣𝜏 𝜕𝑓𝐸𝑞(𝑇) 𝜕𝑥 + 𝑔1 = 0 (9) = 𝜕𝑓𝐸𝑞(𝑇) 𝜕𝑇 𝑑𝑇 𝑑𝑥 . Multiplying Eq. (9) by 𝑓𝐸𝑞(𝑇) depends on x only through T, enabling the replacement 4𝜋 3 ℏ𝜔𝑣 and summing over all k, 𝜕𝑓𝐸𝑞(𝑇) 𝜕𝑥 − 3 5 (𝑣𝜏)2 𝜕2𝑞𝐿𝐹 𝜕𝑥2 + 1 3 𝐶𝐿𝐹𝑣2𝜏 𝜕𝑇 𝜕𝑥 + 𝑞𝐿𝐹 = 0 (10) 10 Here 𝐶𝐿𝐹 = 4𝜋 𝜕 𝜕𝑇 thermodynamic property. ∑ ℏ𝜔𝑓𝐸𝑞(𝑇) 𝑘 is the volumetric heat-capacity of the LF modes, an equilibrium Defining 𝜅𝐿𝐹 = 𝐶𝐿𝐹𝑣2𝜏 as the thermal conductivity of the LF modes, in analogy with the expression of the kinetic theory [24], and 𝛬𝐿𝐹 = 𝑣𝜏, the MFP of low-frequency phonons, Eq. (10) reads 1 3 𝑞𝐿𝐹 = 3 5 (𝛬𝐿𝐹)2 𝜕2𝑞𝐿𝐹 𝜕𝑥2 − 𝜅𝐿𝐹 𝜕𝑇 𝜕𝑥 (11) A similar analysis is carried out for the high-frequency distribution, ℎ(𝑥, 𝒌) = ∑ ℎ𝑙(𝑥, 𝑘)𝑃𝑙(𝑐𝑜𝑠𝜃) ∞ 𝑙=0 (12) with two distinctions. First, we truncate the SHE at the first order, ℎ(𝑥, 𝒌) = ℎ0 + ℎ1𝑐𝑜𝑠𝜃. Second, we assume quasi-Bose statistics for the symmetric part of this distribution, that is, ℎ0 = 𝑓𝐸𝑞(𝑇). The result of this analysis is 𝑞𝐻𝐹 = −𝜅𝐻𝐹 𝜕𝑇 𝜕𝑥 (13) Adding Eq. (11) and Eq. (13), and defining the total thermal conductivity 𝜅 = 𝜅𝐻𝐹 + 𝜅𝐿𝐹, 𝑞 = 𝑞𝐿𝐹 + 𝑞𝐻𝐹 = 3 5 (𝛬𝐿𝐹)2 𝜕2𝑞𝐿𝐹 (14) 𝜕𝑇 𝜕𝑥 𝜕𝑥2 − 𝜅 (𝛬𝐿𝐹)2 𝜕2𝑞𝐻𝐹 3 Adding and subtracting 5 for the heat-flux: 𝜕𝑥2 where 𝑞𝐻𝐹 is known from Eq. (13), we finally arrive at an equation 𝑞 = 3 5 (𝛬𝐿𝐹)2 𝜕2𝑞 𝜕𝑥2 + 3 5 𝜅𝐻𝐹(𝛬𝐿𝐹)2 𝜕3𝑇 𝜕𝑥3 − 𝜅 𝜕𝑇 𝜕𝑥 (15) This is to be combined with energy conservation for the total energy density E, in the presence of an external source of heat 𝑆𝐻𝐹(𝑥, 𝑡) that couples only to the HF modes 𝜕𝑞 𝜕𝑥 = − 𝜕𝐸 𝜕𝑡 + 𝑆𝐻𝐹(𝑥, 𝑡) (16) We now make the first approximation beyond the truncation of the SHEs, namely that 𝐸~𝐶𝑇 where the total heat capacity 𝐶 equals 𝐶𝐻𝐹, the heat capacity of the HF modes only. In other words, we assume that the heat-capacity of the LF modes satisfies 𝐶𝐿𝐹 ≪ 𝐶𝐻𝐹. This assumption is usually included as part of the two-fluid model [16] as described in Sec. 2, and its plausibility may be roughly justified by noting that in a simple Debye model, the density-of-state, and hence the heat capacity per unit energy, varies as the square of the frequency [24]. The thermal conductivities of the two channels however may still be comparable because of their additional dependence on the respective MFPs. Although Eq. (16) is a statement of overall energy conservation, we exclude the coupling of external heat sources to the LF modes in order to maintain consistency with the omission of source terms from Eq. (6) for the LF modes. Thus we arrive at our result, the enhanced Fourier law, which comprises the set of two coupled equations: 𝜕𝑞 𝜕𝑥 = −𝐶 𝜕𝑇 𝜕𝑡 + 𝑆𝐻𝐹(𝑥, 𝑡) (17a) 11 𝑞 = 3 5 (𝛬𝐿𝐹)2 𝜕2𝑞 𝜕𝑥2 + 3 5 𝜅𝐻𝐹(𝛬𝐿𝐹)2 𝜕3𝑇 𝜕𝑥3 − 𝜅 𝜕𝑇 𝜕𝑥 (17b) The basic approximation of truncation of the spherical harmonic expansion at the second order needs more justification for quasi-ballistic phonon modes. For this, we refer the reader to [25], where a generalized form of the enhanced Fourier law is compared against a closed form solution of the BTE for the transient grating experiment, and nearly perfect agreement is obtained over the entire range of values of the experimental length-scale. Acknowledgments: We are grateful to Dr. Alexei Maznev (Massachusetts Institute of Technology) for several helpful discussions. We also wish to thank Professor Jonathan Malen (Carnegie Mellon University) for giving us access to raw data from the paper, Regner et al., which was used to construct Figs. 1 and 6. This work was funded by the National Science Foundation (NSF) under contract number CMMI-1363207. References 1. C. Dames, G. Chen, ''Thermal conductivity of nanostructured thermoelectric materials, Thermoelectrics 2. Handbook: Macro to Nano, Chapter 42, CRC Press, ed. D. Rowe, 2005 Justin P. Freedman, Jacob H. Leach, Edward A. Preble, Zlatko Sitar, Robert F. Davis and Jonathan A. Malen, Universal phonon mean free path spectra in crystalline semiconductors at high temperature, Sci. Reports 3, 2963 (2013) 3. Maznev, A. A., Jeremy A. Johnson, and Keith A. Nelson. "Onset of nondiffusive phonon transport in transient thermal grating decay." Physical Review B 84, no. 19 (2011): 195206. 4. Y. K. Koh and D. G. Cahill, Frequency dependence of the thermal conductivity of semiconductor alloys, Phys. Rev. B 76, 075207, 2007 5. A. J. Minnich, J. A. Johnson, A. J. Schmidt, K. Esfarjani, M. S. Dresselhaus, K. A. Nelson, and G. Chen, Thermal Conductivity Spectroscopy Technique to Measure Phonon Mean Free Paths, PRL 107, 095901, 2011 6. Keith T. Regner, Daniel P. Sellan, Zonghui Su, Cristina H. Amon, Alan J.H. McGaughey, Jonathan A. Malen, Broadband phonon mean free path contributions to thermal conductivity measured using frequency domain thermoreflectance. Nature Comm. 4, 1640 (2013) 7. K. T. Regner, S. Majumdar, and J. A. Malen, Instrumentation of broadband frequency domain thermoreflectance for measuring thermal conductivity accumulation functions, Review of Scientific Instruments 84, 064901 (2013); Jason M. Larkin and Alan J. H. McGaughey, Thermal conductivity accumulation in amorphous silica and amorphous silicon, Phys. Rev. B 89, 144303 (2014) 8. 9. F. Yang and C. Dames, Mean free path spectra as a tool to understand thermal conductivity in bulk and nanostructures, Phys. Rev. B 87, 035437 (2013) 10. Hang Zhang, Chengyun Hua, Ding Ding, Austin J. Minnich. Length Dependent Thermal Conductivity Measurements Yield Phonon Mean Free Path Spectra in Nanostructures, arXiv:1410.6233v1 [cond- mat.mes-hall] 2014 11. G. Romano and J. C. Grossman, Multiscale Phonon Conduction in Nanostructured Materials Predicted by Bulk Thermal Conductivity Accumulation Function, arXiv:1312.7849v3 (2014) 12. Paddock, Carolyn A., and Gary L. Eesley. "Transient thermoreflectance from thin metal films." Journal of Applied Physics 60, no. 1 (1986): 285-290. 13. Ramez Cheaito, John T. Gaskins, Matthew E. Caplan, Brian F. Donovan, Brian M. Foley, Ashutosh Giri, John C. Duda, Chester J. Szwejkowski, Costel Constantin, Harlan J. Brown-Shaklee, Jon F. Ihlefeld, and Patrick E. Hopkins, Thermal boundary conductance accumulation and transmission: Measurements and theory. Phys. Rev. B 91, 035432 (2015) interfacial phonon 12 14. Keivan Esfarjani, Gang Chen, and Harold T. Stokes, Heat transport in silicon from first-principles calculations, Phys. Rev. B 84, 085204, 2011 15. A. T. Ramu, “An enhanced Fourier law derivable from the Boltzmann transport equation and a sample application in determining the mean-free path of nondiffusive phonon modes”, J. Appl. Phys. 116, 093501 (2014) 16. B. H. Armstrong, "Two-fluid theory of thermal conductivity of dielectric crystals", Physical Review B 23, no. 2 (1981): 883. 17. A. A. Maznev, J. A. Johnson, and K. A. Nelson. "Onset of nondiffusive phonon transport in transient thermal grating decay." Physical Review B 84, no. 19 (2011): 195206 18. Bjorn Vermeersch, Jesus Carrete, Natalio Mingo, Ali Shakouri. Superdiffusive heat conduction in semiconductor alloys -- I. Theoretical foundations. arXiv:1406.7341v2 19. A. T. Ramu and Y. Ma, Validation of a unified nondiffusive-diffusive phonon transport model for nanoscale heat transfer simulations, Proceedings of the ASME 2014 Internat. Mech. Eng. Cong. and Exposition, November 14-20, 2014, Montreal, Quebec, Canada 20. C. Hua, and A. J. Minnich, "Transport regimes in quasiballistic heat conduction", Physical Review B 89, no. 9, 094302 (2014). 21. N. Taketoshi, T. Baba, E. Schaub, and A. Ono, Homodyne Detection Technique using spontaneously generated reference signal in picosecond thermoreflectance meaurements, Rev. Sci. Instrum 74, 12, 2003 22. H. Kroemer, Quantum Mechanics for Engineering, Materials Science and Applied Physics (Prentice-Hall, 1994). 23. A. J. Schmidt, X. Chen, and G. Chen, Pulse accumulation, radial heat conduction, and anisotropic thermal conductivity in pump-probe transient thermoreflectance, Rev. Sci. Instrum. 79, 114902 (2008) 24. J. M. Ziman, “Electrons and phonons: The theory of transport phenomena in solids.” © Oxford University Press, pg. 259 (1960) 25. A. T. Ramu and J. E. Bowers, “A generalized enhanced Fourier law and underlying connections to major frameworks for quasi-ballistic phonon transport”, arXiv preprint arXiv: 1506.00668v1 (2015) 26. J. D. Jackson, “Classical electrodynamics”, John Wiley & Sons, © 2003: pp. 96 27. G. A. Baraff, "Maximum anisotropy approximation for calculating electron distributions; application to high field transport in semiconductors." Physical Review 133, no. 1A (1964): A26. 13 

1709.05829

1

1709

Topological properties and edge states in a driven modified dimerized chain

We investigate topological phases induced by a driven electric field coupled to a dimer chain (a model for poly-acetylene) at high frequency regime. It is shown how the topological invariant of the system can be controlled by the field amplitude. Furthermore, in the presence of a time-periodic electric field, the effect of the next-nearest neighbor hopping amplitudes on topological properties is studied. Breaking of the inversion symmetry causes to remove the degeneracy of zero edge states. The fractional Zak phase which is now measurable by ultra-cold atoms in one dimensional optical lattice is also calculated. For calculating modified band structure, we also develop a general Floquet-Bloch approach for systems under application of a potential with lattice and time translation invariance.

cond-mat.mes-hall

cond-mat

a Topological properties and edge states in a driven modified dimerized chain Fatemeh Askari Shahid, Hosein Cheraghchi∗ School of Physics, Damghan University, 36716-41167, Damghan, Iran (Dated: March 13, 2018) We investigate topological phases induced by a driven electric field coupled to a dimer chain (a model for polyacetylene) at high frequency regime. It is shown how the topological invariant of the system can be controlled by the field amplitude. Furthermore, in the presence of a time-periodic electric field, the effect of the next-nearest neighbour hopping amplitudes on topological properties is studied. Breaking of the inversion symmetry causes to remove the degeneracy of zero edge states. The fractional Zak phase which is now measurable by ultra-cold atoms in one dimensional optical lattice is also calculated. For calculating modified band structure, we also develop a general Floquet-Bloch approach for systems under application of a potential with lattice and time translation invariance. PACS numbers: Keywords: Topological invariant, SSH model, Zak phase I. INTRODUCTION Additional to the topological insulators originating from spin-orbit interaction1,2, recently, the attention has been paid to inducing non-trivial topological phases by application of time-periodic perturbations3–5. Those topological phases which are sometimes disappeared in the undriven trivial systems6. By using recent advances in experiment, engineering of quantum properties is man- ifested such that a topological band structure is achiev- able to create by application of an external irradiation7,8. Furthermore, a quantum Hall state can be realized in a time-periodic perturbation3,9. Motivated by the new possibility in the measurement of topological invariants10,11, one dimensional systems with non-trivial properties have recently attracted much attention to study6,12,13. Ultra-cold atoms trapped by optical lattices have prepared an appropriate substrate for realizing topological invariants and measuring the Zak phase as the parameter for characterizing topologi- cal properties of 1D systems10. The simplest one dimen- sional model which exhibits topological band structure is so called Su-Schrieffer-Heeger (SSH) model which be- longs to the BDI symmetry class; a model for describ- ing polyacetylene14. This model can be mapped into many other systems such as; graphene nanoribbons15, sp-orbital optical ladder systems16, off-diagonal bichro- matic optical lattices17. Such a dimerized chain has a topological non-trivial phase depending on the ratio of the intera-cell to inter-cell hopping parameters18. How- ever, illumination of an ac electric field can displace the boundary of trivial to non-trivial phases and induce some extra edge states which are controlled by the renormal- ized parameters arising from ac field19. On the other hand, the extended SSH system with considering next- nearest neighbour hopping amplitude shows Haldane's phase diagram when hopping amplitudes are cyclically modulated by an additional parameter12. However, for manifesting such a system In this work, for calculating the band structure of a quantum lattice in the presence of time-periodic per- turbation, we present a general and straightforward ap- proach based on the Floquet-Bloch (FB) states for a lattice with basis in its unit cell. In different driving regimes, photon-assisted quasi-energy spectrum and FB states are derived for a modified driven SSH system in the presence of next-nearest neighbour (NNN) couplings. The NNN couplings break the inversion symmetry which results in a break at the zero mode's degeneracy. At high frequency regime, the quasi-energy spectrum and its topological invariant (the Zak phase) are investigated to characterize the edge state dependency on the ac field parameters. Motivated by the direct measurement of the Zak phase in onsite-staggered ultra-cold atoms trapped in 1D optical lattice10, the Zak phase variation inducing by the NNN hopping parameter is studied. This paper is organized as the following: In section II, the FB band theory is presented in a general form. As an application, we introduce modified driven SSH Hamilto- nian in section III. The FB band structure for different cases will be presented in section IV for high frequency regime. The next section is addressed to the Zak phase calculation and the edge states tracing. Conclusion sec- tion summarizes our results. II. FLOQUET-BLOCH BAND THEORY ON A CRYSTAL WITH BASIS In a periodically driven quantum lattice, Hamiltonian is periodic in time and space H(~r, t) = H(~r + ~R, t) = H(~r, t + T ), where ~R and T = 2π/ω are the lattice vec- tor and period of time. The time-dependent Schrodinger equation is as the following form. HF (t)Ψ(~r, t) = 0 (1) where Floquet Hamiltonian is defined as HF (t) = H(t) − ∂ . Here H(t) is the time-periodic Hamiltonian of a i ∂t lattice and hence wavefunction Ψ must be a Bloch type both in time and space. As a result, the wavefunction in a one period evolution changes by a pure phase factor leading to invariance of the observable value < ΨAΨ >. Therefore, we have Ψ(~r, t) = e−iǫtψ(~r, t), where ψ as a FB state is a periodic function of time and obeys the Bloch theorem in space as well. Here ǫ is called the quasi- energy of FB states. Replacing this type of wavefunction in Eq.1 transforms Schrodinger equation to an eigenvalue equation form. HF (t)ψ(~r, t) = ǫψ(~r, t) (2) Remembering the tight-binding method, we can in prin- ciple construct the crystal states consisting of a linear combination of localized atomic orbitals20. Therefore, one can correspondingly construct the FB sum of ~k vector as the following: ψλ,α,n(~k, ~r, t) = N − D 2 X~R ei~k. ~R−inωtφα,n(~r − ~R−~τλ) (3) where φα,n(~r − ~R−~τλ) is the α 'th Floquet atomic orbital centered in the reference unit cell for the atom in a po- sition ~τλ in the unit cell (for λ 'th basis in the unit cell). Here n, D and N are the quantum number characterizing Floquet bands, the system dimension and the number of unit cell in the crystal, respectively. φ ~R,~τ ,α,n > is a pe- riodic state in space. The above FB sum preserves both the Bloch theorem and time invariant of the FB state. T ~R′ψλα,n(~k, ~r, t) > = ei~k. ~R′ TT ψλα,n(~k, ~r, t) > = ψλα,n(~k, ~r, t) > ψλα,n(~k, ~r, t) >, (4) (5) where T ~R′ and TT are translational operators in space and time. The FB sum defined in Eq.(3) is used as the basis function for constructing a general solution for the crystal wave function of ~k in the form Φ(~k, ~r, t) = Xλ,α,n Cλ,α,n(~k)ψλ,α,n(~k, ~r, t). (6) where the coefficients Cλ,α,n(~k) are determined by vari- ational method. By using the above expansion of crys- tal wave function on FB sums, one can calculate quasi- energies and eigenfunctions of Floquet Hamiltonian by application of the variational principle. The quasi-energy spectrum and crystal eigenfunctions are obtained from the following determinant. [HF (~k)]n,m λα,λ′α′ − ǫδn,mδλα,λ′α′ = 0 The matrix elements of Floquet Hamiltonian in the basis of FB sums (Eq.(3)) are written in the following form. λα,λ′α′ [HF (~k)]n,m = ≪ ψλ,α,n(~k, ~r, t)HF (t)ψλ′,α′,m(~k, ~r, t) ≫ = Gn,m λα,λ′α′ (~k) − nωδn,mδλ,λ′ δα,α′ (7) 2 FIG. 1: A schematic view of driven modified SSH model where intera-cell and inter-cell hoppings are γ ′ and γ, respectively. Hoppings between atoms in A and B sublattices are γa and γb, respectively. where Gn,m λα,λ′α′(~k) = X~R′′ e−i~k. ~R′′h^H m−ni ~R′′, ~τλ,α;0,~τ ′ λ′ ,α′ (8) ≪ .. ≫= 1/T R T In the above calculation, the time averaging is denoted to 0 dt < .. >. Furthermore, we use orthog- onality condition of the Floquet atomic orbitals as the following ≪ φ ~R,~τ ,α,nφ ~R′,~τ ′,α′,m ≫= δn,mδ ~R, ~R′ δ~τ ,~τ ′δα,α′ . Here, Fourier transformation of the time-periodic Hamil- tonian is defined as ^H m−n = 0 H(t)ei(m−n)ωtdt. The written as expectation value of eH on the Floquet atomic orbitals is (cid:16) ^H m−n(cid:17) ~R′′,~τ ,α;0,~τ ′,α′ =< φ ~R′′,~τ ,α,n^H m−nφ0,~τ ′,α′,m > (9) 1 T R T where "0" denotes to the reference unit cell. Conse- quently, to have the spectrum of driven systems it is enough to calculate matrix elements of FB Hamiltonian G as represented in Eq.8. Diagonalizing the Floquet Hamiltonain results in the quasi-energies of driven sys- tem. III. DRIVEN MODIFIED SSH MODEL A well-known simplest model for describing topological insulator properties is SSH model. This model is a spin- less fermion hopping in one-dimensional lattice with stag- gered hopping amplitudes. In the Driven modified SSH Hamiltonian, nearest-neighbour(NN) and next-nearest- neighbour(NNN) hoppings are both time dependent in- ducing by an ac electric field. H(t) = Xi [(γ′(t)c† a,icb,i + γ(t)c† b,i−1ca,i + C.C.) + (γa(t)c† a,i−1ca,i + γb(t)c† b,i−1cb,i) + C.C.)](10) where hopping parameters depend on time as the fol- ~A(t).d~l]. Φ0 is the quan- lowing γi,j(t) = γ0 tum of magnetic flux. The vector potential is originated Φ0 R ~ri i,j exp[ 2π ~rj 10 (a) 5 0 −5 −10 −1 10 (c) 5 0 −5 10 (b) 5 0 −5 n=2 n=1 n=0 n=−1 n=−2 −0.5 0 0.5 −10 −1 1 −0.5 0 0.5 1 10 (d) 5 0 −5 ) 0 γ ( ε ) 0 γ ( ε −10 −1 −0.5 0 k(π/a0) 0.5 −10 −1 1 −0.5 0.5 1 0 k(π/a0) FIG. 2: The quasi-energy spectrum of the driven modified SSH model for different cases at high frequency regime ω = 3: a) considering only nearest neighbour (NN) hoppings with γa = γb = 0, b) by adding next nearest neighbour (NNN) hoppings to the Hamiltonian γa = 0.2, γb = 0.1. Here λ = γ ′/γ is lesser than the unity (λ = 0.3). The quasi-energy spectrum in a frequencies comparable with the NN hoppings ω = 1.2 and λ = 1.2 for the two cases: c) NN approximation, d) in the presence of NNN hoppings with γa = 0.3, γb = 0.1. In all cases, the amplitude of the ac electric field is considered to be E0 = 0.1. All energies are scaled to γ = 1. from a time-periodic electric field written in a Weyl gauge ~E = −∂ ~A/∂t. Here we neglect magnetic component of illuminated laser. The vector potential results in a phase shift in the hopping amplitude connecting site i to j. As depicted in Fig.1, hoppings at zero field, γ0 ij, are equal to γ′ for intra-cell hopping, γ for inter-cell hoppings, γa for hopping between atoms in A sublattice, and γb for hopping between B atoms. The vector potential for monochromatic waves with linear polarization is defined as A(t) = A0 sin(ωt), where A0 = E0/ω is related to the amplitude of the electric field. FB Hamiltonian can be derived using formula 8 for arbitrary n, m. In the SSH model, the number of basis in the unit cell and number of orbitals on each site are as λ = 2, α = 1, respectively. Gn,m(k) =  where 3 OBC 1.5 1 0.5 0 PBC ) 0 γ ( ε −0.5 −1 −1.5 0 2.5 2 A 0 1 3 3 4 3.5 4 0.5 ) x ( u 0 −0.5 0 0.5 ) x ( u 0 −0.5 0 50 site position 100 50 site position 100 0.5 1 1.5 2 A 0 1.5 1 0.5 ) 0 γ ( ε 0 −0.5 −1 −1.5 0 FIG. 3: Quasi-energy spectrum of the driven modified SSH model as a function of the amplitude of the vector potential A0 at high frequency regime ω = 10 and in OBC with 50 dimers in the chain. Here only NN approximation is consid- ered γa = γb = 0. In this case we consider λ = 0.3. Left inset figures indicate eigen wave function in terms of site po- sitions. In the high frequency regime, right-down inset figure shows quasi-energy band structure of the driven modified SSH model for different values of the wave vectors in PBC. x = 2πA0b0 Φ0 , y = 2πA0(a0 − b0) Φ0 , z = 2πA0a0 Φ0 above the Bessel T R T equation we have used the In the lowing definition of 1 fol- function Jp(x) = 0 eix sin(ωt)−ipωtdt. Let us first look at the quasi- energy spectrum of the bulk system for n, m = 2 in the two regimes of high and low driving frequency. Fig.2 (a,b) represents Floquet side bands separated by the label n in the high frequency regime. Different replicas which are called Floquet side bands are related to n number of photons assisted in the spectrum. In the degenerated energies of spectrum, the driving induced gaps appear around the energy ω/2. It should be noted that these dynamical gaps are not originated from breaking of time reversal symmetry6. These dynamical gaps are propor- tional to the amplitude of the vector potential. In high frequency regime (ω ≫ γ0), Floquet bands are decoupled from each other and so FB Hamiltonian in Eq. 11 becomes block diagonal. High frequency regime corresponds to x, y, z → 0 and so all Bessel functions goes to zero for n 6= m. In this case, Hamiltonian 11 can be represented in terms of the Pauli matrices. Gn(k) = EI (k)I + ~d(k).~σ (13) 2γa cos(ka0)Jm−n(z) ρ(k) ρ∗(k) 2γb cos(ka0)Jm−n(z)  (11) ρ(k) = γ′Jn−m(x) + γJm−n(y)eika0 (12) where I and ~σ = (σx, σy, σz) are referred to the identity matrix and Pauli matrices, respectively. The components 1.5 1 0.5 0 ) 0 γ ( ε −0.5 −1 0.5 ) x ( u 0 −0.5 0 0.5 ) x ( u 0 −0.5 0 50 site position 100 50 100 −1.5 0 0.5 site position 1.5 1 4 OBC ) 0 γ ( ε 1.5 1 0.5 0 −0.5 −1 −1.5 0 1 2 A 0 3 PBC 4 4 OBC 1.5 1 0.5 1.5 1 0.5 0 ) 0 γ ( ε −0.5 −1 −1.5 0 1 2 A 0 3 0 0 ) γ ( ε PBC 4 4 −0.5 −1 −1.5 0 0.5 ) x ( u 0 −0.5 0 0.5 ) x ( u 0 −0.5 0 50 site position 100 50 site position 100 2 A 0 2.5 3 3.5 0.5 1 1.5 2 A 0 2.5 3 3.5 FIG. 4: Quasi-energy spectrum of the driven modified SSH model as a function of A0. Here γa = γb = 0.1 and λ = 0.3. Left inset figures indicate decaying of the two degenerate edge modes. Band structure of the bulk system (PBC) is presented in the right-down inset figure for different wave vectors. FIG. 5: Quasi-energy spectrum of the driven modified SSH model as a function of A0. Here γa = −γb = 0.2 and λ = 0.3. Left inset figures indicate decaying of the two degenerate edge modes. Band structure of the bulk system (PBC) is presented in the right-down inset figure for different wave vectors. of vector ~d and diagonal terms EI (k) are IV. BAND STRUCTURE IN PRESENCE OF DRIVEN ELECTRIC FIELD   . EI (k) = (γa + γb)J0(z) cos(ka0) dx(k) = γ′J0(x) + γJ0(y) cos(ka0) dy(k) = γJ0(y) sin(ka0) dz(k) = (γa − γb)J0(z) cos(ka0) Here if γa = γb, then dz = 0 and therefore the inversion symmetry σxG(k)σx = G(−k) is preserved. As a result, there is no gap opening. However, in this case, the en- ergy shift is not zero EI (K) 6= 0 which leads to breaking of the particle-hole symmetry and also chiral symmetry. However, the energy shift is zero when γa = −γb, while dz 6= 0. Therefore, in this case, the inversion symmetry is broken while the particle-hole symmetry is preserved. So, a band gap is opened in the spectrum. In the special amplitude of the electric field where J0(z) = 0 at z = z∗ as the zeros of the Bessel function, the chiral symmetry σzG(K)σz = −G(k) and also the inversion symmetry are both preserved leading to degenerate zero-mode edge state. In high frequency regime and for n = m, the quasi- energy of Hamiltonian shown in Eq.11 is read as ǫ±(k) = EI (k) ± ~d, where the condition for the gap closing point at k = π/a0 is derived as the following. [γ′J0(x) − γJ0(y)]2 + [γa − γb]2J0(z)2 = 0 In the case γa = γb, the gap closing point would occur at Λ = γ ′ J0(x) J0(y) = 1 and a topological phase is expected γ to appear for Λ < 1. In general case of γa 6= γb, the bulk system has a gap with no electron-hole symmetry in the band structure. At high frequency regime, it will be shown that based on the value of Λ, there is a trivial to non-trivial phase transition. In fact, by variation in the amplitude of the vector potential (A0), one can control topological proper- ties of the system. To show the edge modes, the spectra of Floquet quasi-energy in terms of A0 is investigated for both: the limited number of dimer chains (50 dimers) in the open boundary condition (OBC) and also the peri- odic boundary condition (PBC). In all following cases, driving frequency is considered to be ω = 10. Case 1: In the NN approximation, there are two zero edge state modes with double degeneracy which are seen in Fig.3. In this case, because γa = γb = 0, the edge modes are protected by both the electron-hole and in- version symmetries. There is no gap opening in PBC indicating in the right-down inset figure 3 and the band structure is symmetric around the zero mode. The left inset figures 3 show that paired edge states located on the opposite edges of dimer chain are pinned to the zero energy and they decay exponentially with the localiza- tion length ξ that depends on Λ as the following form, Λ ≈ exp[−a0/ξ]. So the localization length enhances when Λ → Λcr., where Λcr. = 1 − 1 M+1 . Here M is the number of dimers in a finite dimer chain. There is no edge states for the amplitudes of the electric field satis- fied Λ > Λcr.. In fact, by application of the driven field, hopping energies are renormalized by zero Bessel func- tions. So hopping energies are smaller than the hopping parameters of the undriven SSH model. Therefore, the band width of the driven SSH model is thinner than the undriven one. Case 2: Let us consider contribution of the NNN hop- pings in the Hamiltonian. At the first step, we investi- 1.5 1 0.5 ) 0 γ ( ε 0 −0.5 −1 −1.5 0 50 site position 100 OBC 1.5 1 0.5 0 PBC ) 0 γ ( ε 0.5 ) x ( u 0 −0.5 0 0.5 ) x ( u 0 −0.5 0 50 site position 0.5 1 1.5 −0.5 −1 −1.5 0 2.5 100 2 A 0 1 2 A 0 3.5 3 3 4 4 0.5 ) x ( u 0 −0.5 0 0.5 ) x ( u 0 −0.5 0 50 100 site position 50 site position 100 ) 0 γ ( ε 0.2 0.4 0.6 0.8 1 A 0 1.2 1.4 5 OBC 4 2 0 −2 0 PBC 0.5 1.6 1 A 0 1.5 1.8 2 2 3 2 1 0 ) 0 γ ( ε −1 −2 −3 0 FIG. 6: Quasi-energy spectrum of the driven modified SSH model as a function of A0. Here γa = 0.2, γb = 0.1 and λ = 0.3. Left inset figures indicate decaying of the two degen- erate edge modes. Band structure of the bulk system (PBC) is presented in the right-down inset figure for different wave vectors. FIG. 7: Quasi-energy spectrum of the driven modified SSH model as a function of A0. Here γa = 0.5, γb = 0.1 and λ = 1.5. Left inset figures indicate decaying of the two degen- erate edge modes. Band structure of the bulk system (PBC) is presented in the right-down inset figure for different wave vectors. gate the symmetric NNN hoppings in which γa = γb 6= 0. Quasi-energy spectrum of Fig.4 shows breaking of the electron-hole symmetry in the spectrum. In addition, the edge modes have no zero energy. The edge modes cross zero energy only at the special electric fields as z = z∗. In this case, the inversion symmetry is preserved and so there is no gap in the bulk system (PBC). As we will show later, adding the term EI (k) whenever dz(k) = 0 just shifts energy spectrum and does not change the topo- logical phase diagram in compared to the case 1. In this case, the chiral symmetry is broken. Case 3: It is interesting to have a look at the spectrum of the anti-symmetric NNN hoppings in Fig.5 where γa = −γb. In this case, the degeneracy of the two edge modes located at the opposite ends of the chain is broken. The spectrum is symmetric around zero energy. Here dz 6= 0, so a gap is opened in the bulk system (PBC) except for the points z = z∗. In addition, the Zak topological phase differs from the Zak phase related to the cases 1 and 2. Crossing points of the edge states (OBC) occur at the points z = z∗. Case 4: In general case, NNN hoppings are considered to be γa 6= γb. In this case, there is no symmetry at all. So as shown in Fig.6, band structure is asymmetric and also gapped. Furthermore, the edge modes are non- degenerate and have non-zero energy. Emergence of the edge modes crossing each other at the points z = z∗, proposes that these systems are topologically non-trivial. All the above cases are calculated for λ < 1, where in the undriven SSH model there exists a non-trivial topological phase. However, for λ > 1 where the undriven system is in the trivial phase, ac electric field can induce the non- trivial topological phases as seen in Fig.7. V. ZAK PHASE FOR DRIVEN MODIFIED SSH MODEL To elucidate the relation between the existence of the edge modes in a finite chain with the topological invari- ants in the bulk system, the Zak phase is calculated based on the Hamiltonian 11 for the different cases. The Zak phase is a parameter for measuring the solid angle that the pseudo-spinor of G0(k) rotates on the Bloch sphere when k spans the Brillouine zone. The pseudo-spinor of G0(k) can be written as: ψ+(k)i =  ψ−(k)i =  cos θ(k) 2 e−iφ(k) sin θ(k) 2 sin θ(k) 2 e−iφ(k) − cos θ(k) 2     (14) where φ(k) = arctan(dy/dx) and θ(k) = arccos(dz/d). It should be noticed that the constant term EI (k) does not modify the pseudo-spinor. For the case of dz = 0, the pseudo-spinors evolve with the wave vector in the equatorial plane of the Bloch sphere when θ(k) = π/2. In this case, the Zak phase is multiple of π. However, in the modified SSH model with NNN hoppings (dz 6= 0), the pseudo-spinors move away from the equatorial plane leading to Zak phase value which is not necessarily a multiple of π. For a Bloch function ψ(k), the Zak phase can be expressed as the following: ΦZ = iZ π/a0 −π/a0 hψ(k)∂kψ(k)idk ∆=0.1 λ=0.3 a) π / ) 0 ( Ζ Φ 1.1 1 0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0 Λ=0.75 Λ=1.25 0.5 0.4 0.3 0.2 0.1 Ζ Φ δ 0 -0.1 -0.2 -0.3 -0.4 -0.5 -2 -1 0 ∆ 1 0 0.5 1.5 2 1 Λ FIG. 8: The Zak phase difference δΦZ,− = ΦZ,−(∆) − ΦZ (0) in terms of a) ∆ (for two regimes Λ < 1 and > 1) and b) Λ (with ∆ = 0.1). If one consider the NNN hoppings, the Zak phase is changed from ΦZ(0) to ΦZ (dz) for the upper and lower bands. δΦZ,± = ΦZ,±(dz) − ΦZ(0) = ∓Z π/a0 −π/a0 dz 2d ∂kφ(k)dk (15) where the Zak phase for the case of dz = 0 are written as the following form. ΦZ (0) = 1 2 Z π/a0 −π/a0 ∂kφ(k)dk = 1 2Z π −π 1 + Λ cos(x) fΛ(x) dx (16) where in these integrals, the azimuthal angle of the pseudo-spinor and the function f are derived as tan φ(k) = sin(ka0) Λ+cos(ka0) and fΛ(x) = sin2 x + (Λ + cos x)2. The integration result for ΦZ(0) is ΦZ = π if Λ < 1, while ΦZ = 0 if Λ > 1. So there is a non-trivial topo- logical phase which corresponds to the regions with the existence of the edge modes in OBC. Even if the undriven system is in the trivial phase in a regime with λ > 1, it would be possible to tune the parameter Λ by using ap- plied ac electric field such that the driven SSH model (for dz = 0) shows non-trivial topological phase (Fig.7). The phase difference for the driven modified SSH model can be derived as the following. δΦZ,± = ∓ 1 2Z π −π (1 + Λ cos(x))∆ cos x fΛ(x)pfΛ(x) + ∆2 cos2 x dx (17) γJ0(y) where the dimensionless parameter ∆ is defined as ∆ = (ta−tb)J0(z) . The Zak phase in this case is a fraction of π. Considering the NNN hoppings ∆ breaks the inversion and electron-hole symmetries. Hence the edge modes are not protected and the non-trivial topological phase (for Λ < 1) will be changed. Let us look at the results shown in Fig.(8.a) for the variation of the Zak phase difference 6 λ=0.3 λ=1.5 b) π / Ζ Φ 1.4 1.2 1 0.8 0.6 0.4 0.2 0 -0.2 0 d) π / Ζ Φ 1.4 1.2 1 0.8 0.6 0.4 0.2 0 -0.2 0 1 2 Α0 3 4 0.5 1 Α0 1.5 2 -0.1 0 1 2 Α0 3 4 λ=0.3 c) π / Ζ Φ 1.4 1.2 1 0.8 0.6 0.4 0.2 0 -0.2 0 1 2 Α0 3 4 FIG. 9: The Zak phase for different cases as a function of the field amplitude. a) for the case with ta = tb b) for ta = −tb = 0.2 c) for ta = 0.5, tb = 0.1. All cases are in the regime λ < 1. d) for ta = 0.5, tb = 0.1 with λ = 1.5. Here the driven frequency is considered to be ω = 10. for the lower band between the modified (∆ 6= 0) and standard (∆ = 0) SSH model as a function of ∆. By increasing ∆, the pseudo-spin vector goes away from the equatorial plane leading to smaller solid angle variation whenever k spans the first Brillouine zone. In the case Λ > 1 where the phase of standard SSH model is trivial, by inducing NNN hoppings (∆), the Zak phase becomes non-zero, while for Λ < 1, the Zak phase decreases to zero. Furthermore, as it is seen in Fig.(8.b), the deviation of the Zak phase from what we know from the standard SSH model, depends on Λ. In fact, the evolution of the pseudo-spin vectors with k is large close to the transition point (Λ = 1) . In other words, the variation in the solid angle inducing by ∆ is large at the vicinity of the transition point. It should be noted that if we consider staggered on- site energy in the driven SSH model with the nearest neighbour approximation (γa = γb = 0), again the z- component of the vector ~d is non-zero but independent of k (dz = ∆0). The formula derived in Eq.15 can be still used for calculating the Zak phase of staggered driven SSH model as well. Finally, the Zak phase for the different cases was calcu- lated by using equations 16 and 17. As seen in Fig.(9.a), the Zak phase profile for the case of the driven SSH model (case 1) and also the case with the symmetric NNN hop- pings (case 2) are exactly similar. The Zak phase is zero or π depending on the value of A0. The regions with the π value for the Zak phase (which correspond to Λ < 1) are in agreement with the regions exhibiting the edge modes in figures 3 and 4. In the case with the symmet- ric NNN hoppings, the edge modes are degenerated but with non-zero energy. However, there exist non-trivial topological regions with the Zak phase value of π. For the cases 3 and 4, the Zak phase as a function of the field amplitude is presented in figures 9.b and 9.c. The results show some variations in the Zak values in compared to the Zak phase of the cases 1 and 2. Al- though the regions with non-trivial topological phases still remain unchanged, the Zak phase has a relatively large shift close to the transition points in compared to the Zak phase for the cases 1 and 2. The Zak phase at the points with z = z∗ is what one expects from the standard driven SSH model. At the end, the Zak phase is calculated for the case with λ > 1 where the undriven SSH system is in the trivial phase. However, as shown in Fig.(9.d), the Zak phase becomes nonzero in some regions of the field am- plitudes. As shown in Fig.7, the regions with the Zak phases around π represent non-zero edge modes leading to non-trivial topological states. 7 In this paper, it is shown how application of an ac elec- tric field can change topological properties of the system. Since the hopping parameters are renormalized by means of time-periodic field, the Zak phase is dependent on the amplitude and frequency of the ac electric field. At high frequency regime, the effect of the NNN hopping param- eters on the edge modes and the Zak phase are studied leading to a phase variation measurable directly in ultra- cold atoms trapped in optical lattices. Furthermore, we also develop a general Floquet-Bloch band theory for D dimensional lattice with a basis in its unit cell. This for- malism prepares a straightforward and simple form for the FB band structure calculation. VII. ACKNOWLEDGEMENT VI. CONCLUSION As a conclusion, we investigate topological invariants and the edge modes of the modified driven SSH model. H.C. thanks the International Center for Theoretical Physics (ICTP) for their hospitality and support during a visit in which part of this work was done. We would like to acknowledge the instructive comments of Fatemeh Adinehvand in the early stages of the work. ∗ Electronic address: cheraghchi@du.ac.ir 1 M. K onig, S. Wiedmann, C. Br une, A. Roth, H. Buh- mann, L. W. Molenkamp, X.-L. Qi, and S.-C. Zhang, Quantum spin Hall insulator state in HgTe quantum wells, Science 318, 766 (2007). 2 M. Z. Hasan and C. L. Kane, Colloquium: Topological insula- tors, Rev. Mod. Phys. 82, 3045 (2010). 3 T. Oka and H. Aoki, Photovoltaic Hall effect in graphene, Phys. Rev. B. 79, 081406 (2009). 4 T. Kitagawa, E. Berg, M. Rudner, and E. Demler, Topo- logical characterization of periodically driven quantum sys- tems, Phys. Rev. B. 82, 235114 (2010). 5 T. Kitagawa, T. Oka, A. Brataas, L. Fu, and E. Dem- ler, Transport properties of nonequilibrium systems under the application of light: Photoinduced quantum Hall in- sulators without landau levels, Phys. Rev. B. 84, 235108 (2011). 6 V. Dal Lago, M. Atala, and L. E. F. Foa Torres, Floquet topological transitions in a driven one-dimensional topo- logical insulator, Phys. Rev. A. 92, 023624 (2015). 7 Y. H. Wang, H. Steinberg, P. Jarillo-Herrero, and N. Gedik, Observation of floquet-bloch states on the surface of a topological insulator, Science, 342, 453 (2013). 8 M. C. Rechtsman, J. M. Zeuner, Y. Plotnik, Y. Lumer, D. Podolsky, F. Dreisow, S. Nolte, M. Segev, and A. Szameit, Photonic Floquet topological insulators, Nature (London), 496, 196 (2013). 9 Z. Gu, H. A. Fertig, D. P. Arovas, and A. Auerbach, Floquet Spectrum and Transport through an Irradiated Graphene Ribbon, Phys. Rev. Lett. 107, 216601 (2011). 10 M. Atala, M. Aidelsburger, J. T. Barreiro, D. Abanin, T. Kitagawa, E. Demler, and I. Bloch, Direct measurement of the Zak phase in topological Bloch bands, Nat. Phys. 9, 795 (2013). 11 G. Jotzu, M. Messer, R. Desbuquois, M. Lebrat, T. Uehlinger, D. Greif, and T. Esslinger, Experimental re- alization of the topological Haldane model with ultracold fermions, Nature (London) 515, 237 (2014). 12 L. Li, Z. Xu, and S. Chen, Topological phases of general- ized SuSchriefferHeeger model, Phys. Rev. B. 89, 085111 (2014). 13 L. Li, C. Yang, S. Chen, Winding numbers of phase tran- sition points for one-dimensional topological systems, Eu- roPhys. Lett. 112, 10004 (2015). 14 W. P. Su, J. R. Schrieffer, and A. J. Heeger, Solitons in polyacetylene, Phys. Rev. Lett. 42, 1698 (1979). 15 P. Delplace, D. Ullmo, and G. Montambaux, Zak phase and the existence of edge states in graphene, Phys. Rev. B. 84, 195452 (2011). 16 X. Li, E. Zhao, and W. V. Liu, Topological states in a lattice containing ultracold atoms in ladder-like optical higher orbital bands, Nat. Commun. 4, 1523 (2013). 17 S. Ganeshan, K. Sun, and S. Das Sarma, Topological Zero- Energy Modes in Gapless Commensurate Aubry-Andr- Harper Models, Phys. Rev. Lett. 110, 180403 (2013). 18 S. Ryu and Y. Hatsugai, Topological Origin of Zero-Energy Edge States in Particle-Hole Symmetric Systems, Phys. Rev. Lett. 89, 077002 (2002). 19 A. Gomez-Leon and G. Platero, Floquet-bloch theory and topology in periodically driven lattices, Phys. Rev. Lett. 110, 200403 (2013). 20 G. Grosso and G. P. Parravicini, Solid State Physics, Aca- demic Press, ISBN: 0-12-304460-X (2000) 10 (a) 5 0 −5 ) v e m E ( n=2 n=1 n=0 n=−1 n=−2 10 (b) 5 0 −5 −10 −4 −2 0 2 −10 −4 4 −2 0 2 4 (c) 5 0 ) v e m E ( (d) 6 4 2 0 −2 −4 −5 −4 −2 0 K(1/nm) 2 4 −4 −2 0 K(1/nm) 2 4 'zak.dat' e s a h P k a Z 1.2 1 0.8 0.6 0.4 0.2 0 -0.2 0 0.2 0.4 0.6 0.8 1 A0 1.2 1.4 1.6 1.8 2 'zak.dat' e s a h P k a Z 1.6 1.4 1.2 1 0.8 0.6 0.4 0.2 0 -0.2 0 0.5 1 1.5 2 A0 2.5 3 3.5 4 'zak.dat' e s a h P k a Z 2 1.5 1 0.5 0 -0.5 0 0.5 1 1.5 2 A0 2.5 3 3.5 4 'zak.dat' 1.2 1 0.8
0.4 0.2 e s a h P k a Z 0 0 0.5 1 1.5 2.5 3 3.5 4 2 A0 

1905.01100

1

1905

Current-induced atomic forces in gated graphene nanoconstrictions

Electronic current densities can reach extreme values in highly conducting nanostructures where constrictions limit current. For bias voltages on the 1 volt scale, the highly non-equilibrium situation can influence the electronic density between atoms, leading to significant inter-atomic forces. An easy interpretation of the non-equilibrium forces is currently not available. In this work, we present an ab-initio study based on density functional theory of bias-induced atomic forces in gated graphene nanoconstrictions consisting of junctions between graphene electrodes and graphene nano-ribbons in the presence of current. We find that current-induced bond-forces and bond-charges are correlated, while bond-forces are not simply correlated to bond-currents. We discuss, in particular, how the forces are related to induced charges and the electrostatic potential profile (voltage drop) across the junctions. For long current-carrying junctions we may separate the junction into a part with a voltage drop, and a part without voltage drop. The latter situation can be compared to a nano-ribbon in the presence of current using an ideal ballistic velocity-dependent occupation function. This shows how the combination of voltage drop and current give rise to the strongest current-induced forces in nanostructures.

cond-mat.mes-hall

cond-mat

Current-induced atomic forces in gated graphene nanoconstrictions S. Leitherer,1 N. Papior,2 and M. Brandbyge1 1Center for Nanostructured Graphene, Department of Physics, Technical University of Denmark, DK-2800 Kongens Lyngby, Denmark 2Department of Applied Mathematics and Computer Science, Technical University of Denmark, DK-2800 Kongens Lyngby, Denmark Electronic current densities can reach extreme values in highly conducting nanostructures where constrictions limit current. For bias voltages on the 1 volt scale, the highly non-equilibrium situation can influence the electronic density between atoms, leading to significant inter-atomic forces. An easy interpretation of the non-equilibrium forces is currently not available. In this work, we present an ab-initio study based on density functional theory of bias-induced atomic forces in gated graphene nanoconstrictions consisting of junctions between graphene electrodes and graphene nano-ribbons in the presence of current. We find that current-induced bond-forces and bond-charges are correlated, while bond-forces are not simply correlated to bond-currents. We discuss, in particular, how the forces are related to induced charges and the electrostatic potential profile (voltage drop) across the junctions. For long current-carrying junctions we may separate the junction into a part with a voltage drop, and a part without voltage drop. The latter situation can be compared to a nano- ribbon in the presence of current using an ideal ballistic velocity-dependent occupation function. This shows how the combination of voltage drop and current give rise to the strongest current- induced forces in nanostructures. I. INTRODUCTION The current densities in nano-scale, ballistic conduc- tors can reach extreme values compared to macroscopic Ohmic conductors. For example, the break-down volt- ages of atomic chains of Au are beyond 1 V corresponding to a current-density1 on the order of 1010 A/cm2, and the current-carrying capacity of narrow graphene conductors can reach almost 109 A/cm2 before breakdown.2 From a technological point of view the nano regime poses chal- lenges in terms of stability and reproducibility, since in this extreme scaling limit the position of a few atoms control the device operation. On the other hand, atomic control of the structure by external driving forces offers an enormous potential for further downscaling. F.ex. it has been demonstrated in experiments how the current/field may be used to toggle switch atomic-scale contacts between different conduc- tance states corresponding to different atomic configura- tions of metallic nano-contacts.3 In this paper we will concentrate on another impor- tant example, namely graphene nanostructures, which are now being created and changed using high applied voltages and consequently electrical current. Due to its excellent electrical and mechanical properties, graphene is a promising material for two-dimensional (2D) nano- electronic applications.4 So-called "electro-burning" has been employed in experiments to fabricate nano-gaps be- tween graphene electrodes.5 These electrodes of single or few-layer graphene has in some cases subsequently been bridged by single molecules.6 -- 9 Using similar techniques, the fabrication of electrically switchable graphene break junctions has been reported.10,11 Electron microscopy allows for structural, atomic-scale studies of graphene structures in the presence of high current and applied voltage.12 It has been seen how the structure of edges are changed by the current/voltage13 or how layers fuse.14 Current-induced motion/cleaning of adsorbed species on graphene has also been investigated.2 Under high bias and current density a number of differ- ent, possibly intertwined, effects play crucial roles for the atomic configuration, such as motion driven by locally in- duced fields, Joule heating and temperature gradients, as well as current-induced forces due to a steady momentum transfer from electronic current to ions.15 -- 17 Common for these structures and effects in graphene nanostructures is that the electrons may to a large degree be in the ballistic quantum transport regime, as seen e.g. by the appear- ance of interference phenomena.5,18,19 Experiments per- formed at high voltage bias on a bilayer constriction show an uniaxial lattice expansion of more than 5% at a cur- rent density on the order of 109 A/cm2 before breaking.20 The understanding of the role of voltage and currents in such systems and processes are still rudimentary. We consider here a simple, narrow graphene ribbon system using first principles calculations based on Density Func- tional Theory combined with Non-Equilibrium Greens functions (DFT-NEGF). We have previously studied the electron-phonon interaction in transport and the voltage drop dependence on gating in this system.21,22 In this paper we consider the current-induced forces in the pres- ence of steady-state electronic current, and analyze these in terms of the changes in electronic distributions. Our DFT-NEGF calculations, presented in the first part of this work, return forces for systems that are de- fected in the sense of having a scattering region. How- ever, for ballistic bulk systems one could imagine a cur- rent flowing which is far from any scattering potential, and approximately behaves as though states are occupied depending on their velocity. We present this approach and compare to the DFT-NEGF forces in the last part of this paper (Sec. III B 2). a) b) c) FIG. 1. Graphene nanoconstrictions with electrostatic bot- tom gate. a) represents the unit cell of a generic constriction, with the blue square indicating the position and shape of the electrostatic gate. Edge atoms of the graphene are saturated by hydrogen in order to avoid dangling bonds. b) Short con- striction and c) semi-infinite GNR constrictions considered in this work. II. SETUP AND METHOD The systems we investigate are constrictions consist- ing of graphene nanoribbons (GNRs) of varying lengths, placed between graphene electrodes (cf. Fig. 1). In addi- tion, the junctions are electrostatically gated. Their ge- ometries are relaxed at zero bias using the Siesta pack- age and their properties are studied at finite bias using the nonequilibrium electronic transport package Tran- Siesta. Computational details are described in Sec. V. We apply the field-effect gate model of Ref. 22. A charged plane is placed at 15 A underneath the graphene constriction. The plane carries a charge density of n = g · 1013e−/cm2, where g defines the gating levels, with g < 0/g > 0 referring to n/p-doping. Placing the elec- trostatic gate allows for a tuning of the conductance of the junction, while on the other hand, the position of the voltage drop in the constriction can be controlled.22 Thus we explicitly include the role of the gate-induced carriers on the screening properties and potential profile. We focus on two distinct geometries, shown in Fig. 1 (b,c): The first (b) is a graphene constriction with a very short GNR, and the second junction (c), consists of a large region of pristine graphene connected to a semi- infinite GNR. The results are presented in Sec. III A, Sec. III B, respectively. Since our aim here is to study generic features of the local current and potential drop, and the relation to inter- atomic forces, we neglect the role of spin-polarization at the zig-zag edges.23 2 III. RESULTS AND DISCUSSION A. Short constriction We first consider the left-right symmetric graphene nanojunction with a short GNR (cf. Fig. 1 (b)). We em- ploy periodic boundary conditions in the direction trans- verse to the constriction with a corresponding k-point sampling. 1. Conductance properties In Fig. 2 (a-c), we discuss the transport properties of the junction, in particular the transmission probability, total currents, and real-space "bond currents". In the following we implicitly assume k-dependence. Thereby, the transmission through the constriction is calculated from, T (E, V ) = Tr(cid:2)GΓLG†ΓR(cid:3) , (1) a) b) d) c) 1.0 V Imax=0.52 µA FIG. 2. a) Zero-bias transmission for different gate charges g (gray without gate), b) total current as a function of bias voltage for selected gate charges, c) real-space bond currents at 1 V, g = −2, and d) maximal absolute force over bias voltage for different gate charges. 3 µ R -0.5 where G = [(E + iη)S − H − Σ]−1 is the nonequilibrium retarded Green's function with device Hamiltonian H, Σα, and Γα = overlap S and selfenergies Σ = Pα=L,R i(Σα† − Σα). The total current is given by, a) µ L EF g<0 V>0 d) g>0 V>0 µ L EF I(V ) = 2e h Z T (E, V )[fL(E, V ) − fR(E, V )] dE, (2) with fL/R being the Fermi distributions in the electrodes, where the chemical potentials at finite bias are shifted according to µL/R = EF ± eV /2. Fig. 2a shows the zero-bias transmission for different values of the gating parameter g. Gating leads to a dop- ing of the junction, i.e. the charge-neutrality point in the DOS is shifted relative to its position at g = 0. Accord- ingly the transmission is shifted further into the conduc- tance window with g. This results in a higher conductiv- ity especially at small bias, cf. total current in Fig. 2b. In the high bias regime the current is not significantly en- hanced by the gating, because the transmission at high energies is nearly 1. A spatial distribution of the current flowing through the junction can be obtained by calculating bond currents.24 -- 26 The energy-dependent spectral bond cur- rents from atom n to m are defined as, µ R b) +0.5 -0.5 e) +0.5 c) f) ∂Jnm(E, V ) = 2e  Xµ,ν Im(cid:8)Aα µν (E, V )[H(V ) − ES]νµ(cid:9) (3) where α = L, R refers to the electrode, ν ∈ n and µ ∈ m are orbital indices, and the spectral function is given by Aα(E, V ) = GΓαG†. (4) The bond current is obtained by integrating Eq. (3) over the Fermi window, defined by fL − fR at the correspond- ing bias: FIG. 3. a) Energy scheme of the electrode DOS for negative g and positive bias voltage, b) electrostatic potential profile, and c) bias-induced charge redistribution at 1 V for g = −2. (d,e,f) DOS, potential and charges at 1 V for g = +2. Jnm(V ) = 1 2π Z ∂Jnm(E, V )[fL(E, V ) − fR(E, V )] dE. (5) In Fig. 2c the bond currents at a bias of 1 V are shown. The highest current density appears at the entrance to the constriction and along the edge atoms in the con- striction. In the pristine graphene, bond currents obtain smaller values and spread out across the lattice. They obey the law of particle conservation, i.e. through any section dividing the left and right part their total sum is conserved. The current pattern exhibits a somewhat left-right/top-bottom symmetry, which obviously stems from the junction symmetry. We want to study the interatomic forces in the graphene constriction, which are induced when a finite bias voltage is applied. These forces are calculated from the non-equilibrium electron density defined by the den- sity matrix D, which we obtain from the DFT-NEGF formalism.27 In particular, the force acting on atom n with coordinate ~Rn is given through the force operator ~Fn and the density operator D via ~Fn = Trh~FnDi = − Tr(cid:20) ∂H ∂ ~Rn D(cid:21) , (6) and the non-equilibrium density operator, D = Z (cid:2)AL(E, V )fL(E, V ) − AR(E, V )fR(E, V )(cid:3) dE. (7) In Fig. 2d, we plot the maximum absolute force in- duced by the non-equilibrium between all atoms in the short GNR constriction, depending on the gate parame- ter and the bias voltage (for a spatial distribution of the forces, see below). The maximum force is seen to increase with voltage roughly following the current, where as both are more weakly influenced by the gate parameter. We find forces of ∼ 0.2 nN at 1 V. Theoretical models were compared to tunnel-to- contact experiments of atomic point contacts in order to a) a) b) d) e) 4 c) f) FIG. 4. a) Induced bond forces, Fb, and change in overlap population ∆OP at 1.0 V, g = −2. The forces Fb are shown as vectors in light red and blue. Force vectors are pointing inwards (outwards) to indicate bond compressing (stretching), while the vector thickness corresponds to the force strength. The change in OP is depicted as in-/decreasing density along the bond, in red for positive and blue for negative ∆OP. Induced forces/charges below a cutoff of Fb = 0.004 nN/∆OP= 1 · 10−4 e are set to zero. b) Correlation between bond forces Fb and ∆OP, c) correlation between bond forces Fb and bond currents Jnm, (d,e,f) equivalent pictures for g = +2. explicitly relate the conductance to the atomic forces at low bias ∼ 1mV, as for example presented in Ref. 28. Be- low, we will present a detailed analysis of the forces and compare these to the local current and potential drop at the higher voltages. tion is highest at the interface of the potential drop. Our analysis shows that forces are highly correlated with such charge redistributions and in the following we will outline simple relations between the charge redistri- butions and forces. 2. Potential drop and finite bias charge redistribution 3. Bond forces and overlap population At finite bias, the chemical potential in the elec- trodes is symmetrically shifted and an electrical field be- tween the electrodes exists across the junction, result- ing in a rearrangement of charge. We present in Fig. 3a the schematic picture of the electrode density of states (DOS) and the energy levels of the junction at finite bias for g = −2, (b) the electrostatic potential landscape Φ(1 V, g = −2) − Φ(0 V, g = −2) and (c) the induced charge ρ(1 V, −2) − ρ(0 V, −2) in the short GNR con- striction. In Figs. d-f we present the same analysis for g = +2. The energy scheme, a) and d), illustrates how the non- symmetric coupling is induced via the electrode having the largest DOS in the bias window. This results in an electrostatic potential pinning of the contriction. In b) it pins to the right electrode presenting the larger DOS in the voltage window, and opposite for the case in e), see Ref. 22 for details. Such relative changes in the electro- static potential also results in a different charge redistri- bution. In c) and f) we show how the charge redistribu- To simplify the representation of the forces, Eq. (6), we project them onto the atomic bonds. These bond forces are defined as the difference of the forces on atom n and m, projected onto the bond vector ~rnm = ~rm − ~rn: ~Fb,nm = ( ~Fm − ~Fn) · ~rnm ~rnm (8) With this definition, positive (negative) bond forces can be interpreted as compressive (repulsive). Note that our structures are relaxed at zero bias, thus Fb refers to the bias induced forces. Fig. 3a depicts induced bond forces at 1 V for g = 2 (d for g = −2). Compressive (repulsive) bond forces are shown as arrows in light red (light blue). We draw the force arrows at both atoms of a bond to indicate if the force is stretching or compressing the bond. In a),d) we also show how we can relate the forces to the charge redistribution in the junction. In particular, we have calculated the amount of charge in the bonds, also termed overlap population (OP), similar to the anal- ysis in Ref. 29. This approach is based on interpreting the bond population as a measure of the bond strength.30 The OP is given by a sum over atomic orbitals (i, j) be- longing to the atoms n, m, OP = Xα=L,R Xi∈n j∈m Oα ij, with Oα ij = Sij Z dǫ Aα ij (ǫ)fα(ǫ − µα). (9) (10) a) 5 b) c) and the spectral function Aα, where α = L, R, since it has contributions from left and right-originating states. To obtain the bias-induced bond charge, we calculate the change in overlap population with bias, ∆OP = OP(V )− OP(0). In Fig. 3a and d, the nonzero ∆OP are depicted as density along the bond; in particular the line thickness corresponds to ∆OP and red (blue) indicates if it has positive (negative) sign. We find that the bond forces and the change in overlap population are clearly corre- lated. An increase (decrease) of charge in the bond corre- sponds to a positive (negative) bond force, corresponding to bond elongation (compression). This correlation be- tween bond force and population is also revealed by the scatter plots in Fig. 3b for g = 2 (and e for g = −2). Similarly to the bond forces, we plot the bond currents in Fig. 2c. While the bond-currents show a left-right sym- metry, this symmetry is fully absent in the bond forces. As shown in Fig. 3c and f, the bond current and force strength do not clearly correlate. This suggests that even though certain atoms experience a high current density it is not necessarily reflected in forces acting on it, or at least its effect is minor compared to other effects. We note that recent work calculating the current-induced forces in graphene nanoribbons based on single-orbital tight-binding model find a correlation between the local currents and bond-forces.31 But this clearly will depend on the level of description of the connection to electrodes and the associated potential drop. To get further insight into the bias-induced bond popu- lations, we analyze the crystal orbital overlap population (COOP) curve32, which is the energy-resolved overlap population. For the bond between atoms n and m, the COOP is defined as COOP(E) = 2 Xi∈n,j∈m Sij Aα ij(E), (11) with α = L, R referring to left- and right-coming states. The sign of the COOP curve determines whether the states contributing to the bond have bonding (positive) or anti-bonding (negative) character.32 Therefore filling of bonding/depletion of antibonding states will lead to a strengthening of the bond force, and vice versa. Note FIG. 5. a) Chemical bond of atom m and n with compressive bond force (red), and bond of atom n and k with repulsive bond force (blue) at 1 V and g = −2. b) DOS of left-/right- going states, AL and AR, on atoms m and n (left panel) and atoms n and k (right panel) at 1 V. States in AL (AR) below µL = 0.5 eV (µR = −0.5 eV) are occupied. c) COOP analysis for bond m-n (left panels) and bond n-k (right panels). Bond m-n is strengthened by current, as at 1 V bonding states get populated in AL (red shaded area in left COOP). The DOS that is depleted in AR is small. Bond n-k is weakened by current, as antibonding states get filled (blue shaded area). that integrating the COOP (weighted by the Fermi dis- tribution) gives the OP. In Fig. 5 we present this analysis for two bonds in the junction of Fig. 3a, which experience a high bond force: One bond that is compressed, m-n, and one that is stretched, n-k, under influence of the current (cf. Fig. 5a). The DOS of left- and right traveling states at 1 V, shown in b), is similar for atoms m-n (left panel) and atoms n-k (right panel). However, for atoms m-n, the states that are energetically located in the conductance window have bonding character, as indicated by a positive COOP (left panels in c), while on atoms n-k they are antibonding (right panels in c). Moslty relevant are states in AL, since those get filled by shifting µL up, while the occupation of right states (bottom panels) does not change significantly by the downshifting of µR. The positive ∆OP for the bond m-n can be traced back to an increased filling of bonding states, while on atoms n-k antibonding states become occupied, resulting in a negative ∆OP. B. Wide-narrow GNR constriction In order to study in more detail the influence of the potential profile on the forces, we consider a wide-narrow constriction, where the right electrode is a semi-infinite GNR (cf. Fig. 1c). We focus on a negative doping of g = −2 and positive bias voltages. 1. Potential drop, charge redistribution, bond currents and forces at non-equilibrium In longer GNR constrictions, a pinning of the poten- tial to one of the electrodes can be achieved, leading to a very localized potential drop at the transition between the GNR and the graphene. In Fig. 6a we show the po- tential profile at positive bias (here 0.75 V) in a graphene constriction with a horizontally extended GNR. For a detailed discussion of the bias-dependence of the volt- age drop in very similar GNR constrictions, we refer to Ref. 22. In Fig. 6b, we show the the bias-induced charge den- sity. We find that the largest amount of charge is accu- mulated near the potential drop. This is due to the fact that at finite bias, reflection of incoming channels takes place at the scatterer, i.e. the constriction entrance where the potential drop is located. These scattering processes induce Landauer dipoles in this region.33 Farther away a) b) c) FIG. 6. a) Electrostatic potential profile b) bias-induced charge redistribution and c) bond currents in the wide-narrow GNR constriction at 0.75 V, g = −2. 6 from the potential drop, we find a smaller amount of in- duced charge density in the GNR, which converges at longer distances. As the states closest to EF are edge states, this charge is mainly localized on the zigzag edges of the GNR. Figure 6c depicts bond currents through the extended GNR constriction. Due to particle conservation, the bond currents are of the same size all along the GNR. Note that there is little to no correspondence between the charge redistribution and bond currents. In a similar way as the charge density profile, the non-equilibrium forces are maximal in the region of the potential drop. This is illustrated in Fig. 7a, where we show the change in bond forces and overlap population. Again, we find compres- sive/repulsive forces for bonds where a large amount of bond charge is induced/depleted (cf. Fig. 7c). The max- imum/minimum forces along the transport direction in the junction are depicted in Fig. 7b. For a detailed analysis of the forces, two regions can be distinguished in the junction: One is the region of the wide-narrow transition (dotted square in Fig. 6b), where the potential drop is located. Here we find the largest forces with maximum strength of Fb ≤ 0.38 nN. In this region there are contributions to the forces from the reflected charge density as well as from the density of transmitting channels, beyond the usual electrostatic forces. Deeper in the GNR, the forces and bond populations become significantly smaller (Fb ≤ 0.05 nN) and reach a periodic pattern. In this region (bold square in Fig. 6b), the electrostatic potential profile is very flat and nearly equivalent to the right electrode chemical potential, µR. Thus, the forces in this region are not related to a po- tential drop, but are ideally solely originating from the flow of current. The correspondence between induced bond forces and ∆OP is still given, with the accumu- lated charge coming from the current in the occupied, transmitted conductance channels. 2. Forces without voltage drop In the extended GNR constriction, we have studied forces in region 2, where a flat potential profile has es- tablished. This allows for a comparison with a perfectly ballistic bulk system, where a current flows without the electrical field in the potential drop. This enables the use of periodic boundary conditions and a Bloch band description. Specifically, we may employ a bulk-like cal- culation scheme where states are occupied according to their band velocity, vnk = 1  ∂εn(k) ∂k (12) where n is the band-index. The idealized, ballistic occu- pation function corresponds to a situation where current is fed into the nano-ribbon from ideal electrodes without any scattering in the voltage window. This is of course 7 c) a) b) ] N n [ b F 0.2 0.1 0 -0.1 -0.2 -0.3 15 20 25 30 35 40 45 50 55 60 z [Ang] FIG. 7. transport direction z, c) correlation between Fb and ∆OP, at 0.75 V, g = −2. a) Distribution of induced bond forces Fb and ∆OP in the GNR constriction, b) maximal values of Fb along the idealized and will overestimate the current. The non- equilibrium distribution function relative to equilibrium is, δf (nk) = Θ(vnk · e) [fL(εn(k), V ) − fR(εn(k), V )] (13) where Θ is the Heaviside step function and e is the di- rection of the external bias driving the current. The chemical potentials for left- and right-movers are µL = EF +eV /2 and µR = EF −eV /2 with V being the applied voltage. The quasi-Fermi level, EF , is determined in the self-consistent DFT cycle such that the charge is neutral in the unit-cell. We will denote this type of calculations as ballistic-bulk calculations. We have performed a ballistic-bulk calculation for the GNR for g = −2. Fig. 8 illustrates the filling of bands in the GNR at a bulk bias of 0.75. In Fig. 9, we compare the DFT-NEGF forces in the constriction far away from the potential drop with the forces from the ballistic-bulk calculation. The cutout in Fig. 9a corresponds to the bold square in Fig. 6b, while 8b shows the unit cell of the ballistic-bulk calculation. We recover a very similar force pattern for both calcu- lations, with forces only perpendicular to the transport direction. In both setups, the inner atoms of the GNR are contracting, while the edge atoms and hydrogen are slightly pushed outwards. The correlation between bond forces and induced bond populations is also revealed in the ballistic-bulk calculation, cf. Fig. 9c. Due to the sym- metry of the single unit cell in the bulk calculation, it returns symmetric forces. The DFT forces show some deviations, as Fig. 9a is taken out from the large junc- tion. Also, they are lower in magnitude compared to the bulk forces. The maximum bulk forces in Fig. 9b are 0.2 nN for 0.75 V, comparable to the forces in the region of the voltage drop in the constriction. On the other hand the DFT-NEGF forces in the "bulk" part of the constriction are ∼ 0.05 nN, and thus a factor of 4 smaller than ballistic-bulk, while the ratio of the currents at this FIG. 8. Bulk-bias applied to GNR. Shown is the bandstruc- ture, where a positive bias of 0.75 V along the GNR direction changes the occupation of right-moving states with positive velocity such that states are filled up to µL = eV /2. Simi- larly are the left moving states with negative velocity emptied above µR = −eV /2. Filled bands are indicated in red. voltage is, however, roughly a factor of 20. So it is clear that the quasi-gap in the transmission seen in Fig. 2a due to the connection to the graphene electrode is important. IV. SUMMARY Summing up, we have analyzed non-equilibrium forces due to the presence of current in graphene nanoconstric- tions by employing first principles transport calculations. We have shown that the induced forces are related to a re- arrangement of bond charges due to left/right incoming scattering states. The forces and charges are maximal  a) b) c) "current-induced". 8 Our theoretical work can be help to understand current-induced strains, bond-breaking processes34,35, and mechanisms that lead to the destruction of devices at the atomic scale. Funding by Villum Fonden (Grant No. 00013340) and the Danish Research Foundation (Project DNRF103) for the Center for Nanostructured Graphene (CNG) is ac- knowledged. V. COMPUTATIONAL DETAILS A. DFT parameters FIG. 9. a) Force pattern far away from the potential drop at 0.75 V from TranSiesta calculation, b) forces and c) ∆OP from bulk calculation at 0.75 V. in the region where the potential drop takes place, be- cause scattering happens there and dipoles are induced. We have further demonstrated forces which exist with- out potential drop and can thus be considered as purely 1 C. Sabater, C. Untiedt, and J. M. van Ruitenbeek, Beilstein Journal of Nanotechnology 6, 2338 (2015). 2 J. Moser, a. Barreiro, and a. Bachtold, Applied Physics Letters 91, 163513 (2007). 3 C. Schirm, M. Matt, F. Pauly, J. C. Cuevas, P. Nielaba, and E. Scheer, Nature Nanotechnology 8, 645 (2013). 4 A. K. Geim and K. S. Novoselov, Nat. Mat. 6, 183 (2007). 5 H. Sadeghi, J. A. Mol, C. S. Lau, G. A. D. Lambert, Briggs, Proceedings of the National Academy of Sciences 112, 2658 (2015), http://www.pnas.org/content/112/9/2658.full.pdf. J. Warner, and C. J. The were done using calculations the Siesta/TranSiesta code with the PBE-GGA func- tional for exchange-correlation and a SZP basis-set.36 Spin polarization is not considered. The mesh cutoff was 300 Ry. In Siesta we used an optimized k-point sampling according to the bias window. The transport calculations were averaged over 25 to 50 transverse k-points. In the bulk calculations 1000 k-points along the ribbon are used. Physical quantities like transmission, current, overlap population and COOP were extracted using TBtrans and SISL.37 VI. REFERENCES 18729415, https://doi.org/10.1021/nl801774a. 12 P. J. Harris, Carbon 122, 504 (2017). 13 X. Jia, M. Hofmann, V. Meunier, B. G. Sumpter, J. Campos-Delgado, J. M. J. M. Romo-Herrera, H. Son, Y.-P. Hsieh, A. Reina, J. Kong, M. Terrones, and M. S. Dresselhaus, Science 323, 1701 (2009). 14 A. Barreiro, F. Borrnert, M. H. Rummeli, B. Buchner, and L. M. K. Vandersypen, Nano Letters 12, 1873 (2012). 15 D. Dundas, E. J. McEniry, and T. N. Todorov, Nature nanotechnology 4 2, 99 (2009). 16 M. Di Ventra, S. T. Pantelides, Phys. Rev. Lett. 88, 046801 (2002). and N. D. Lang, 6 F. Prins, A. Barreiro, J. W. Ruitenberg, J. S. Seldenthuis, N. Aliaga-Alcalde, L. M. K. Vandersypen, and H. S. J. van der Zant, Nano Letters 11, 4607 (2011). 7 K. Ullmann, P. B. Coto, S. Leitherer, A. Molina- and H. B. Weber, Ontoria, N. Martin, M. Thoss, Nano Letters 15, 3512 (2015). 8 S. Leitherer, P. B. Coto, K. Ullmann, H. B. Weber, and M. Thoss, Nanoscale 9, 7217 (2017). 9 H. Sun, Z. Jiang, N. Xin, X. Guo, S. Hou, and J. Liao, ChemPhysChem 19, 2258 (2018). 10 H. Zhang, W. Bao, Z. Zhao, J.-W. Huang, B. Stand- ley, G. Liu, F. Wang, P. Kratz, L. Jing, M. Bockrath, and C. N. Lau, Nano Letters 12, 1772 (2012), pMID: 22429115, https://doi.org/10.1021/nl203160x. 11 B. Standley, W. Bao, H. Zhang, J. Bruck, C. N. Lau, and M. Bockrath, Nano Letters 8, 3345 (2008), pMID: 17 J.-T. Lu, R. B. Christensen, J.-S. Wang, P. Hedegard, and M. Brandbyge, Phys. Rev. Lett. 114, 096801 (2015). 18 V. M. Garc´ıa-Su´arez, A. Garc´ıa-Fuente, D. J. Carrascal, E. Burzur´ı, M. Koole, H. S. J. van der Zant, M. El Abbassi, M. Calame, and J. Ferrer, Nanoscale 10, 18169 (2018). 19 P. Gehring, H. Sadeghi, S. Sangtarash, C. S. Lau, J. Liu, A. Ardavan, J. H. Warner, C. J. Lambert, G. A. D. Briggs, and J. A. Mol, Nano Letters 16, 4210 (2016). 20 F. Borrnert, A. Barreiro, D. Wolf, M. I. Katsnelson, B. Buchner, L. M. K. Vandersypen, and M. H. Rummeli, Nano Letters 12, 4455 (2012). 21 T. Gunst, T. Markussen, K. Stokbro, and M. Brandbyge, Phys. Rev. B 93, 245415 (2016). 22 N. Papior, T. Gunst, D. Stradi, and M. Brandbyge, Phys. Chem. Chem. Phys. 18, 1025 (2016). 23 G. Z. Magda, X. Jin, I. Hagym´asi, P. Vancs´o, Z. Osv´ath, P. Nemes-Incze, C. Hwang, L. P. Bir´o, and L. Tapaszt´o, Nature 514, 608 (2014). 30 R. S. Mulliken, The Journal of Chemical Physics 23, 1833 (1955), https://doi.org/10.1063/1.1740588. 31 N. Asoudegi, M. Soleimani, and M. Pourfath, 24 S. Nakanishi and M. Tsukada, J. Appl. Phys 125, 144503 (2019). 9 Physical Review Letters 87, 126801 (2001). 25 T. N. Todorov, Journal of Physics: Condensed Matter 14, 3049 (2002). 26 G. Solomon, C. Herrmann, T. Hansen, V. Mujica, and 32 R. Hoffmann, Rev. Mod. Phys. 60, 601 (1988). 33 R. Landauer, IBM Journal of Research and Development 1, 223 (1957). 34 T. N. Todorov, J. Hoekstra, Phys. Rev. Lett. 86, 3606 (2001). and A. P. Sutton, M. Ratner, Nature Chemistry 2, 223 (2010). 27 M. Brandbyge, J.-L. Mozos, P. Ordej´on, J. Taylor, and 35 A. Erpenbeck, C. Schinabeck, U. Peskin, and M. Thoss, K. Stokbro, Phys. Rev. B 65, 165401 (2002). 28 M. Ternes, C. Gonz´alez, C. P. Lutz, P. Hapala, and A. J. Heinrich, F. J. Giessibl, P. Jel´ınek, Phys. Rev. Lett. 106, 016802 (2011). 29 M. Brandbyge, K. Stokbro, J. Taylor, J.-L. Mozos, and P. Ordej´on, Phys. Rev. B 67, 193104 (2003). Phys. Rev. B 97, 235452 (2018). 36 N. Papior, Lorente, T. N. Garc´ıa, A. sen, Computer Physics Communications 212, 8 (2017). and M. Frederik- Brandbyge, 37 N. R. Papior, "sisl," (2018). 

1102.4512

4

1102

Weak antilocalization in HgTe quantum wells and topological surface states: Massive versus massless Dirac fermions

HgTe quantum wells and surfaces of three-dimensional topological insulators support Dirac fermions with a single-valley band dispersion. In the presence of disorder they experience weak antilocalization, which has been observed in recent transport experiments. In this work we conduct a comparative theoretical study of the weak antilocalization in HgTe quantum wells and topological surface states. The difference between these two single-valley systems comes from a finite band gap (effective Dirac mass) in HgTe quantum wells in contrast to gapless (massless) surface states in topological insulators. The finite effective Dirac mass implies a broken internal symmetry, leading to suppression of the weak antilocalization in HgTe quantum wells at times larger than certain t_M, inversely proportional to the Dirac mass. This corresponds to the opening of a relaxation gap 1/t_M in the Cooperon diffusion mode which we obtain from the Bethe-Salpeter equation including relevant spin degrees of freedom. We demonstrate that the relaxation gap exhibits an interesting nonmonotonic dependence on both carrier density and band gap, vanishing at a certain combination of these parameters. The weak-antilocalization conductivity reflects this nonmonotonic behavior which is unique to HgTe QWs and absent for topological surface states. On the other hand, the topological surface states exhibit specific weak-antilocalization magnetoconductivity in a parallel magnetic field due to their exponential decay in the bulk.

cond-mat.mes-hall

cond-mat

Weak antilocalization in HgTe quantum wells and topological surface states: Massive versus massless Dirac fermions G. Tkachov and E. M. Hankiewicz Institute for Theoretical Physics and Astrophysics, Wurzburg University, Am Hubland 97074 Wurzburg, Germany (Dated: November 19, 2018) HgTe quantum wells and surfaces of three-dimensional topological insulators support Dirac fermions with a single-valley band dispersion. In the presence of disorder they experience weak antilocalization, which has been observed in recent transport experiments. In this work we conduct a comparative theoretical study of the weak antilocalization in HgTe quantum wells and topological surface states. The difference between these two single-valley systems comes from a finite band gap (effective Dirac mass) in HgTe quantum wells in contrast to gapless (massless) surface states in topological insulators. The finite effective Dirac mass implies a broken internal symmetry, leading to suppression of the weak antilocalization in HgTe quantum wells at times larger than certain τM, inversely proportional to the Dirac mass. This corresponds to the opening of a relaxation gap τ −1 M in the Cooperon diffusion mode which we obtain from the Bethe-Salpeter equation including relevant spin degrees of freedom. We demonstrate that the relaxation gap exhibits an interesting nonmono- tonic dependence on both carrier density and band gap, vanishing at a certain combination of these parameters. The weak-antilocalization conductivity reflects this nonmonotonic behavior which is unique to HgTe QWs and absent for topological surface states. On the other hand, the topological surface states exhibit specific weak-antilocalization magnetoconductivity in a parallel magnetic field due to their exponential decay in the bulk. PACS numbers: I. INTRODUCTION Recently discovered materials -- graphene,1,2 two- dimensional (2D)3 -- 6 and three-dimensional (3D)7 -- 12 topological insulators (TIs)13,14 -- exhibit a Dirac-like band dispersion which is responsible for their unusual electronic and optical properties. In graphene the low-energy electron spectrum can be approximated by two spin-degenerate Dirac cones at the corners of the Brillouin zone.15 The 2D TIs have been realized in HgTe quantum wells (QWs)4,5 which have a single double-degenerate Dirac valley, as predicted by band- structure calculations and inferred from transport mea- surements.16 The double degeneracy of the HgTe QW bands allows for an energy gap at the Dirac point, with- out breaking time-reversal invariance, which paves the way to study Dirac fermions with a finite (positive and negative) effective mass and related mass disorder.17 -- 19 In comparison with HgTe QWs, the ideal 3D TI exhibits a single non-degenerate gapless Dirac cone on the surface of the material, whereas its bulk is insulating.20 In this case, the opening of the gap in the surface Dirac spectrum requires time-reversal symmetry breaking and has been predicted to cause the surface quantum Hall effect21 -- 24 and rich magneto-electric phenomena21 -- 27 related to ax- ion electrodynamics.28 The number of the Dirac valleys is an essential fac- tor in understanding quantum electron transport in dis- ordered samples. The transport studies of graphene have reported weak localization29 and more complex quantum-interference patterns30 instead of the antilo- calization effect expected for the symplectic universal- ity class (e.g. Refs. 31 -- 47). Such a situation can oc- cur if the two graphene's valleys are coupled by scatter- ing off atomically sharp defects.38,41,43,46,47 In contrast, in single-valley Dirac systems such scattering processes are forbidden, and recent experiments on HgTe QWs48,49 and 3D TIs50 -- 52 have found a positive (antilocalization) quantum-interference conductivity. In this work we conduct a comparative study of the weak antilocalization (WAL) in HgTe QWs and on sur- faces of 3D TIs. The goal is to elucidate the difference between these two systems which comes from the finite ef- fective Dirac mass in HgTe QWs in contrast to the mass- less surface states in 3D TIs. Like in the conventional 2D electron systems (2DESs) with Bychkov-Rashba or Dres- selhaus spin-orbit interactions,32,34 the WAL in HgTe QWs and on surfaces of 3D TIs reflects the broken rota- tion symmetry in relevant spin space. However, in addi- tion to the lack of this continuous symmetry, the effec- tive Dirac mass in HgTe QWs implies a broken discrete symmetry which for a single-cone system would play the role of time reversal. In this sense, there is a formal anal- ogy between the effective Dirac mass and an out-of-plane Zeeman field in a 2DES.44,53 Therefore, by analogy with weak ferromagnets44,53 we find that the WAL in HgTe QWs is suppressed at times larger than certain τM, in- versely proportional to the effective Dirac mass. Such suppression is however absent for the massless surface states in 3D TIs, which can be used to experimentally differentiate between the two systems. Before going to the calculation details given in Sec. III and IV, in the next section we briefly announce some of our results for HgTe QWs and TIs. a -1 ΤM 0.3 0.2 0.1 M= +5 meV M= 0 meV M=-7.5 meV M=-15 meV 0 3 6 9 n
´1011cm-2
b -1 ΤM 0.3 0.2 0.1 n=1´1011cm-2 n=1.5´1011cm-2 n=2´1011cm-2 n=2.5´1011cm-2 -20 -10 M
meV
10 FIG. 1: (Color online) Relaxation gap τ −1 M [in units of inverse life-time τ −1, see Eq. (2)] versus carrier density n (a) and band gap M (b); A = 380 meV·nm and B = 850 meV·nm2 (from Ref. 16). II. OVERVIEW OF THE RESULTS To calculate the quantum-interference (Cooperon) con- ductivity correction, δσxx, we adopt the effective Dirac model of HgTe QWs4 and obtain the following expression for δσxx: δσxx(n,M) = −α 2e2 πh ln τ −1 τ −1 M + τ −1 ϕ , (1) τ −1 M = 2 τ (M + Bk2 F )2 F + (M + Bk2 F )2 , A2k2 kF = √2πn. (2) FIG. 2: (Color online) Quantum-interference conductivity correction δσxx [in units of e2/πh, see Eq. (1)] versus car- rier density n and band gap M; A = 380 meV·nm, B = 850 meV·nm2 (from Ref. 16) and τ0/τϕ = 0.01 [see also Eq. (22) for τ and τ0 and Eq. (53) for α]. 2 ∆Σxx 8 6 4 2 0 ∆Σxx 8 6 4 2 a 0.05 b n=1.9´1011cm-2 n=2.5´1011cm-2 n=3´1011cm-2 n=3.5´1011cm-2 n=4´1011cm-2 Τ0
Τj 0.1 M= -10 meV M= -5 meV M=0 meV M=+5 meV M=+10 meV 0.05 Τ0
Τj 0.1 FIG. 3: (Color online) Quantum-interference conductivity correction δσxx [in units of e2/πh, see Eq. (1)] versus nor- malized dephasing rate τ0/τϕ for different carrier densities and M = −10 meV (a) and for different band gaps and n = 1.9 × 1011 cm−2 (b); A = 380 meV·nm and B = 850 meV·nm2 (from Ref. 16). See also Eq. (22) for τ and τ0 and Eq. (53) for α. Here the symmetry-breaking-induced relaxation gap τ −1 M is proportional to the total mass term, M + Bk2 F , in the effective Dirac model of HgTe QWs.4 M is the band gap at the Dirac point, Bk2 F is the positive quadratic correc- tion accounting for the curvature of the filled conduction band (kF is the Fermi momentum determined by the 2D carrier concentration n), whereas constant A determines the linear (Dirac) part of the spectrum (τ the elastic life- time). In Eq. (1) the factor of 2 accounts for the double degeneracy of the Dirac valley in HgTe QWs, constant α approaches −1/2 for M + Bk2 F → 0 (as discussed in detail in Sec. III) and τϕ is the dephasing time. The unique feature of the HgTe QWs is that the band gap M can take both positive and negative values de- pending on the QW thickness.5,16 Therefore, the relax- ation gap τ −1 M (2) exhibits an interesting nonmonotonic behavior as a function of both M and carrier density n [see Fig. 1], vanishing when these parameters satisfy the condition, M + 2πnB = 0. (3) It represents a line in parameter space (M, n) on which conductivity (1) reaches the maximum δσxx = (e2/πh) ln(τϕ/τ ) [see also Fig. 2]. Such a nonmonotonic behavior of δσxx(n,M) can be used to experimentally identify the symmetry breaking and the resulting re- laxation gap τ −1 In particular, the predicted carrier- M . density dependence should hold for the QWs where the potential impurity scattering is much stronger than scat- tering off random gap fluctuations. This regime can be identified from the carrier-density dependence of the QW mobility.17 As to the dependence on the gap M , it can be extracted from sample-to-sample measurements. The band structure of MBE grown HgTe QWs is controllable to a great extent.5,6,16 For the experiment we suggest here one should select several QWs with distinctly different gaps and comparable dephasing times (inferred from the temperature dependence of the conductivity) and other band structure parameters (inferred from the Hall and Shubnikov-de Haas measurements). Alternatively, the presence of the relaxation gap τ −1 M can be established from the dependence of δσxx on the dephasing rate 1/τϕ, which is directly related to the temperature dependence (e.g. the dephasing rate due to electron-electron interac- tions is linearly proportional to the temperature).54 The dependence of δσxx on 1/τϕ is shown in Figs. 3(a) and (b). In these figures the upper curves correspond to the logarithmically divergent δσxx with τ −1 In con- trast, for finite τ −1 M (rest of the curves) the conductivity correction shows only weak dependence on the dephasing rate. M = 0. As to the 3D TIs, we focus on compounds Bi2Se3 and Bi2Te3 where the surface states can be described by the effective two-band Dirac Hamiltonian, accounting for the hexagonal warping of the bands [see, e.g., Ref. 55,56]. The warping term is cubic in momentum k and enters for- mally as the Dirac mass term. However, since it preserves the time-reversal symmetry, we find that for weak warp- ing the quantum-interference conductivity correction has the same form as for the conventional 2DES with spin- orbit interaction:31,32,34 δσxx = −α e2 πh ln τϕ τ , α = − 1 2 . (4) The specific of the surface state shows up mainly in the dependence of the conductivity ∆σxx(B) = δσxx(B) − δσxx(0) on magnetic field B applied parallel to the TI surface: ∆σxx(B) = − e2 2πh ln 1 + k!, Bk = B2 B2 . (5)  eλpℓtrℓϕ Such dependence reflects the finite penetration length, λ, of the surface state into the bulk, i.e. the magnetic flux through the effective width of the surface state [see Eq. (5) for Bk , where ℓtr and ℓϕ are the transport mean free path and dephasing length, respectively]. Quantum transport in the in-plane magnetic fields has been studied theoretically for disordered metallic films57 and electron quantum wells.58 The present case differs from the previ- ous studies in that the topological surface states have a different microscopic profile of the transverse wave func- tions. We discuss the dependence of ∆σxx on the mag- netic field orientation in Sec. IV in connection with recent experiments on Bi2Te3 (Ref. 52). 3 FIG. 4: (Color online) Energy bands of a HgTe quantum well [see Eq. (7) and text] in meV versus in-plane wave-numbers kx and ky in nm−1. The Fermi level lies in the conduction band at E = 0. We choose A = 380 meV·nm, B = 850 meV·nm2, D = 700 meV·nm2 (from Ref. 16), and EF = 100 meV. III. HGTE QUANTUM WELLS A. Effective Hamiltonian We use the effective 4-band Hamiltonian of HgTe wells4 which can be written in the following form: 0 H + Hi(cid:19) , HHgT e =(cid:18) H + Hi H = σ(Ak + Mkz) + (Dk2 − EF )σ0, H = −σ(Ak + Mkz) + (Dk2 − EF )σ0, 0 Hi = V (r)σ0. (6) (7) (8) (9) The two diagonal blocks in HHgT e describe pairs of states related to each other by time reversal symmetry (Kramers partners). In the upper block the Hamilto- nian H has a matrix 2 × 2 structure with Pauli matrices σx,y,z and unit matrix σ0 acting in space of two lowest- energy subbands of the HgTe quantum well:4 an s-like electron one E1, 1 2i and a p-like heavy hole one H1, 3 2i. For the lower block the basis states have the opposite spin projections: E1,− 1 2i and H1,− 3 2i. The linear term in H (proportional to constant A and momentum operator k = −i∇r) originates from the hybridization of the s- and p-like subbands. Mk is the effective Dirac mass: Mk = M + Bk2, (10) where constant M determines the band gap at the gamma (k = 0) point of the Brillouin zone (see Fig. 4). The positive quadratic terms Bk2 and Dk2 take into ac- count the details of the band curvature in HgTe quantum wells.4,5 Finally, Hi in Eq. (6) accounts for interaction with randomly distributed short-range impurities. The impuritity potential V (r) is characterized by the correla- tion function, b ζ(r − r′) = hhV (r)V (r′)ii = ζk =Z ζ(r)e−ikrdr = πN τ0   πN τ0 , δ(r − r′), (11) (12) parametrized in terms of the characteristic scattering time τ0 and the density of states (DOS) at the Fermi level, N , for one Dirac cone [brackets hh...ii denote averaging over impurity positions and ζk is the Fourier transform of the disorder correlation function]. We emphasize that the mass term Mk (10) is sym- metric with respect to momentum inversion k → −k. Hence, Hamiltonian H does not possess the symmetry under transformation (13) k, σ → −k,−σ. Within a given block of Eq. (6), i.e. in subband pseu- dospin space, such a transformation plays the role of "time reversal". At the same time, the real time-reversal symmetry is ensured by the matrix form of HHgT e (6). Physically, this means that the Kramers partners reside on two Dirac cones superimposed at k = 0 point.16 Note that the zero off-diagonal elements in HHgT e (6) imply conservation of the spin projections of E1i and H1i subbands, which is a good approximation for symmetric HgTe wells.5,16,59,60 In this case, each of the Dirac cones contributes independently to transport processes, which is accounted for by the factor of 2 in the expressions for the conductivity corrections discussed in subsection III D. B. Disorder-averaged single-particle Green's functions and elastic life-time We begin by calculating the disorder-averaged re- for an n-type tarded/advanced Green's functions G HgTe well under weak-scattering condition R/A kF vF τ ≫ 1, (14) where τ is the elastic scattering time and vF and kF = √2πn are the Fermi velocity and wave-vector, respec- tively. In the standard self-consistent Born approxima- tion (see, e.g. Refs. 61,62) the equation for G is In k representa- shown diagrammatically in Fig. 5(a). tion it reads R/A 4 R C A R C A a c α β d = + R C A + R α α α C = + β A β β R A = R A α β + + R α C A β R A FIG. 5: (Color online) Diagrammatic representations for (a) equation for disorder-averaged Green's function (thick line) in self-consistent Born approximation (thin line is the unper- turbed Green's function of the disorder-free system, dashed line is the disorder correlator), (b) bare and dressed Hikami boxes for the Cooperon correction to Drude conductivity, (c) Bethe-Salpeter equation for the Cooperon, and (d) equation for the renormalized current vertex in the ladder approxima- tion. R/A Here the Green's function G band carrier with dispersion ξk =pA2k2 + M2 0k describes a conduction- k +Dk2− EF in the absence of disorder [index s = ±1 labels the Kramers partners residing on the different Dirac cones]. The valence band contribution is neglected under as- sumption that the energy separation between the valence and conduction bands is much bigger than the disorder- induced band smearing, 2qA2k2 F + M2 kF ·n = 2(EF − Dk2 F ) ≫ /τ. (17) Because of the large band-structure constant A ≈ 380 meV·nm [see, e.g. Ref. 16] the requirement (17) can be satisfied simultaneously with the weak-scattering condi- tion (14). We also note that the matrix structure of G (16) accounts for the carrier chirality and is of primary importance throughout the paper. R/A 0k R/A The solution to Eq. (15) can be sought in the form k = 1 ) 2 (σ0 + σek)g , where g k = 1/(ǫ−ξk−Σ R/A R/A R/A R/A k k G and Σ k satisfies the equation R/A R/A G k = G 0k + G R/A 0k Z dk′ (2π)2 ζk−k′G R/A R/A k′ G k , (15) Σ R/A k =Z ζk−k′ 1 + ekek′ ǫ − ξk′ − ΣR/A k′ 2 dk′ (2π)2 . (18) R/A G 0k = 1 2 σ0 + σek ǫ − ξk , ek = s Ak + Mkz pA2k2 + M2 k , ek2 = 1,(16) Approximate solution61,62 of the latter equation near the Fermi surface, k ≈ kF , yields the disorder-averaged Green's function as R/A G k = 1 2 σ0 + σekF ·n ǫR/A − ξk , ǫR/A = ǫ ± i 2τ , (19) where τ is the elastic life-time given by N  × 1 τ 2  = ∓ Im Σ R/A kF ·n = ×Z dφn′ ζkF ·n−n′ 1 + e⊥n · e⊥n′ + ekn · ekn′ 2 in the orthonormal eigenfunctions of the pseudospin of a two-electron system: 5 Cαβα′β′ =Xij Cij Ψi αβΨj∗ α′β′, Xαβ Ψj αβΨi∗ αβ = δji. (24) The indices i, j = 0, x, y, z label the pseudospin-singlet (0) and pseudospin-triplet (x, y, z) states. The conduc- tivity corrections will be eventually expressed in terms of the coefficients Cij for which we derive the following algebraic equations (see, also, Appendix A): (20) , ekn = snq1 − e2 ⊥n, e⊥n = s MkF ·n z F + M2 qA2k2 kF ·n . (21) τ 4τ0Xs In Eq. (20) the unit vectors n and n′ specify the direc- tions of the incident and scattered momentum states, re- spectively, and ekn and e⊥n are the in- and out-of-plane components of the unit vector ekF ·n = ekn + e⊥n. For isotropic ζkF ·n−n′ (12) and MkF ·n (10) one finds the elastic time τ = τ0 1 + e2 ⊥ , (22) C0j = δ0j + τ 2τ0h1 − e+ · e−iC0j Cij (q) = τ 2 τ0 δij + (25) Trh(σ0 − σe−kF ·n)σi(σ0 + σekF ·n)σsiCsj (q), where the square brackets stand for integral over the di- rections of the momentum k = kF n on the Fermi surface: h...i =Z 2π 0 dφn 2π ... 1 − iτ ω + iτ vF n · q . (26) Evaluating the traces of the products of the Pauli matri- ces in Eq. (25) we find τ 2 τ0 τ 2 τ0 + τ 2τ0 Xb=x,y,z Caj = δaj + h(e+ − e− + ie+ × e−) · biCbj , (27) τ 2τ0h(e+ − e− − ie+ × e−) · aiC0j h(1 + e+ · e−)(a · b) − i(e+ + e−) · a × b τ + 2τ0 Xb=x,y,z −(e+ · a)(e− · b) − (e− · a)(e+ · b)iCbj . (28) We separated the singlet C0j and triplet C(a,b)j Cooper- ons with respect to the first index so that a, b run over x, y, z only. Respectively, vectors a and b run over the unit vector basis of the Cartesian system. We also intro- duced a convenient shorthand notation e± = e±kF ·n. (29) As discussed in subsection III A, the specifics of the ef- fective Hamiltonian for HgTe quantum wells consists in the symmetry of the mass term (10). Being an even func- tion of k, it breaks the symmetry of Hamiltonian H in Eq. (7) under transformation k, σ → −k,−σ. The sym- metry breaking is encoded in the unit vectors e± (29) which have antiparallel in-plane and parallel out-of-plane components, e± = ±ek + e⊥, (30) where ek and e⊥ are defined by Eq. (21). In view of the identities e+ + e− = 2e⊥, e+ · e− = 1 − 2e2 k , e+ − e− = 2ek, e+ × e− = 2ek × e⊥, (31) which is shorter than the disorder-related time scale τ0 [see, Eq. (11)]. This is due to the allowed backscattering into an opposite momentum state (nn′ ≈ −1) which is absent in the gapless case.63 The backscattering is the consequence of the symmetry breaking due to the mass term. The strength of the symmetry breaking is con- trolled by the out-of-plane component e⊥n of the unit vector ekF ·n [see, Eq. (21)]. C. Cooperon The quantum-interference corrections to the Drude conductivity can be expressed diagrammatically by the Hikami boxes shown in Fig 5b. Apart from the (19) they in- single-particle Green's functions G k volve the disorder-averaged two-particle Green's func- tion, Cαβα′β′(q), known as the Cooperon. In this subsec- tion we will set up and solve the equation for Cαβα′β′(q). For the potential disorder defined by Eq. (11) the dia- gram for the Cooperon equation (see, Fig 5c) is read off as follows R/A τ 2 τ0 δαα′ δββ′ +  πN τ0Z dk (2π)2 R G αγ′ (k, ǫ + ω)G A βδ′(q − k, ǫ)Cγ′δ′α′β′ (q),(23) Cαβα′β′(q) = ×Xγ′δ′ where the Greek indices label the states in pseudospin (σ) space. The prefactor τ 2/τ0 in the first term is due to the chosen normalization of Cαβα′β′(q). To solve Eq. (23) it is convenient to first expand the Cooperon Eqs. (27) and (28) reduce to D = Dτ (2 + 5e2 ⊥ + e4 ⊥)/e2 k 6 , Dτ = v2 F τ /2, (40) τ τ0 he2 kiC0j h(ek + iek × e⊥) · biCbj, C0j = δ0j + + τ τ0 Xb=x,y,z τ 2 τ0 τ 2 τ0 Caj = δaj + τ τ0 h(ek − iek × e⊥) · aiC0j (32) (33) (34) (35) Cxx(q, ω) = 2τ e2 k + 2τ (τ −1 M − iω) e4 k Cyy(q, ω) = 2τ e2 k + 2τ (τ −1 M − iω) e4 k 2 − e4 2 + 5e2 (1 + 3e2 k cos2 φq ⊥ + e4 ⊥ ⊥)2 − e6 ⊥ + e4 ⊥ 2 + 5e2 k sin2 φq 2 − e4 2 + 5e2 k (1 + 3e2 sin2 φq ⊥ + e4 ⊥ ⊥)2 − e6 ⊥ + e4 ⊥ k 2 + 5e2 cos2 φq (41) C00(q, ω), (42) C00(q, ω), where angle φq in Eqs. the Cooperon momentum direction: (cos φq, sin φq, 0). (41) and (42) indicates · q = q Let us analyze Eqs. (39) -- (42). The symmetry break- ing has a three-fold effect on the Cooperons. First, it results in a relaxation gap τ −1 M in the singlet Cooperon C00 (39), which implies suppression of the quantum in- terference for times larger than τM (even in the absence of the phase breaking, i.e. for ω → 0). Second, the symmetry breaking affects the diffusion constant D in Eq. (40). The diffusion constant renormalization comes from the off-diagonal Cooperons.41 In the absence of ⊥ = 0) one finds41,63 the symmetry breaking (i.e. D = 2Dτ = v2 F τtr/2 with τtr = 2τ . Finally, the expres- sions for the triplet Cooperons Cxx (41) and Cyy (42) contain additional terms ∝ τ /τM = 2e2 ⊥, remaining fi- nite in the limit ω → 0. Despite the smallness of the parameter e2 ⊥, these terms give a noticeable contribution to the net conductivity correction. We will return to this point when discussing Eq. (53) in the next subsection. for e2 D. Hikami boxes and net conductivity correction We now turn to the evaluation of the Hikami boxes for the conductivity corrections, shown in Fig. 5b. With the Cooperon defined by Eq. (23) the first and second diagrams in Fig. 5b correspond to the following analytical expressions: he2 ⊥(a · b) − ie⊥ · (a × b) + (ek · a)(ek · b) or, explicitly, τ + τ0 Xb=x,y,z −(e⊥ · a)(e⊥ · b)iCbj , h τ0 τ − he2 − hex − ieye⊥iC0j + h τ0 kii C0j − hex + ieye⊥iCxj − hey − iexe⊥iCyj = τ δ0j , xii Cxj − hexey − ie⊥iCyj = τ δxj, τ − he2 ⊥ + e2 − hey + iexe⊥iC0j − heyex + ie⊥iCxj ⊥ + e2 τ − he2 + h τ0 (36) yii Cyj = τ δyj , Czj = τ 2 τ0 δzj. (37) For the quantum-interference conductivity corrections we will only need the q- and ω-dependent diagonal Cooper- ons C00, Cxx and Cyy. Each of them is obtained from coupled Eqs. (34) -- (36) where index j should be set to 0, x or y, respectively. The coefficients in these equa- tions are evaluated by expanding65 the denominator in Eq. (26) in the small Cooperon momentum q and fre- quency ω, τ vF n · q ≪ 1, τ ω ≪ 1. (38) In doing so, we keep the lowest order terms that yield the nonzero angle average h...i and compete with the small symmetry-breaking parameter e⊥ [see, Eq. (21)]. Under these approximations we obtain the following expressions for the diagonal Cooperons: C00(q, ω) = Dq2 + τ −1 1 M − iω , τ −1 M = 2e2 ⊥ τ , (39) δσ(1) xx = e2 πN τ 2Z dǫ 2πω dq (2π)2 Cββ′γγ′(q, ω) [f (ǫ) − f (ǫ + ω)] Z ×Z [f (ǫ) − f (ǫ + ω)]Z (2π)2Z k,ǫ V x dk′ (2π)2 (G dk A dǫ 2πω Z ×Z dk (2π)2 (G A k,ǫ V x R k G k,ǫ+ω)γ′β(G R q−k,ǫ+ω V x A q−k G q−k,ǫ)γβ′, dq (2π)2 Cββ′γγ′(q, ω) R R R k G k,ǫ+ωG q−k′,ǫ+ω)γ′β(G q−k,ǫ+ω G R k′,ǫ+ω V x A k′ G k′,ǫ)γβ′, (44) 7 (43) δσ(2) xx = e22 π2N 2τ0τ 2 where V x k is the matrix current vertex renormalized by disorder (11) in the usual ladder approximation (see, e.g. Ref. 62 and diagram in Fig. 5d) and f (ǫ) is the Fermi distribution function. We will skip the details regarding the third diagram in Fig. 5a since it finally gives the same result as Eq. (44). Evaluation of the k integrals in Eqs. (43) and (44) yields (see, also, Appendix B): δσ(1) xx = e2Dτ 2hv2 F [(σ0 + σekF ·n)V x kF ·n(σ0 + σekF ·n)]γ′β[(σ0 + σe−kF ·n)V x −kF ·n(σ0 + σe−kF ·n)] γβ′Z dq (2π)2 Cββ′γγ′(q),(45) δσ(2) xx = e2Dτ τ 8hv2 F τ0 × [(σ0 + σekF ·n)V x kF ·n(σ0 + σekF ·n)]γ′β1(σ0 + σe−kF ·n) γγ1 × (σ0 + σe−kF ·n′)β1β[(σ0 + σekF ·n′)V x kF ·n′(σ0 + σekF ·n′)] γ1β′Z dq (2π)2 Cββ′γγ′(q), (46) 0 rections n on the Fermi surface: (...) = R 2π where the bar denotes averaging over the momentum di- ...dφn/2π. We note that the correction δσ(2) xx is entirely due to the carrier chirality. If we omit the chirality matrix σekF ·n in Eq. (46), the independent averaging of the current ver- tices V x kF acquires the standard prefactor τtr/τ , where τtr is the transport scattering time: The renormalized vertex V x kF ·n gives δσ(2) xx = 0. 1 τtr = × = N  Z dφn′ (1 − n · n′)ζkF n−n′ 1 + e⊥n · e⊥n′ + ekn · ekn′ 1 + 3e2 ⊥ 2 . 2τ0 and satisfies the identity (σ0 + σekF ·n)V x kF ·n(σ0 + σekF ·n) = 2 (49) × (σ0 + σekF ·n), which helps to perform the averaging in Eqs. (45) and (46). The conductivity corrections δσ(1) xx can xx and δσ(2) vF nx τtr τ then be expressed in terms of the singlet and triplet Cooperons as follows δσ(1) xx = 2e2Dτ τ (cid:17)2Z π (cid:16) τtr − 2n2 x(1 − e2 δσ(2) xx = − 2e2Dτ π × (cid:2)(1 + e2 dq x(1 − e2 x(1 − e2 (2π)2(cid:2)2n2 y)Cxx(q) − 2n2 x)2Z τ 2 tr (2π)2 τ τ0 ⊥)C00(q) − Cxx(q) − e2 (n2 dq e2 k ⊥)C00(q) x)Cyy(q)(cid:3),(50) ⊥Cyy(q)(cid:3),(51) (47) (48) where n2 x = 1/2. We have also summed up the con- tributions of both Dirac cones of the HgTe QW spec- trum, which yields the factor of 2 in front of the integrals. Noticing further that on average over the directions of q the triplet Cooperons (41) and (42) coincide, we express the net conductivity correction in the form: δσxx = δσ(1) xx + 2δσ(2) ×(cid:20)(cid:18)1 − e2 ⊥ − xx = 2e2Dτ π (cid:16) τtr 2(cid:19)C00(q) −(cid:18)1 + e2 k e2 τ (cid:17)2Z qdq 2π e2 k 2(cid:19)Cxx(q)(cid:21), ⊥ − (52) 0 where (...) = R 2π ter Eq. (42). Note that the terms ∝ 1 ∓ e2 Eq. (50) for δσ(1) xx , whereas the terms ∝ e2 Eq. (51) for δσ(2) xx , multiplied by 2. ...dφq/2π with φq defined in text af- ⊥ come from k/2 come from In Eq. (52) the singlet Cooperon C00(q) results in a positive conductivity correction (antilocalization) com- ing from the pairs of states with antiparallel projections of σ, which prevents the constructive interference.32,34 This can also be viewed as the manifestation of π Berry phase.38 In contrast, the conductivity correction due to the triplet Cooperons is negative (localization) because the interference of the states with the parallel projections of σ is constructive. When integrating the singlet and triplet contributions in Eq. (52) we insert Eq. (39) and the terms with the diffusion pole structure (∝ C00(q)) in Eqs. (41) and (42).64 With the upper integration cut-off (Dτ )−1/2 and replacement −iω → τ −1 in Eq. (52), we obtain Eq. (1) for the logarithmic correction to the Drude conductivity, where the relaxation gap τ −1 M is defined in Eq. (39) and the prefactor α is given by ϕ α = − ×(e2 e2 2 + 5e2 ⊥ + e4 k k − τ(cid:18) 1 τM + ⊥ 1 + 3e2 ⊥(cid:19)2 ⊥ ×(cid:18) 1 + e2 τϕ(cid:19) 1 + 3e2 e4 1 ⊥ k (53) 2(1 + 3e2 2 + 5e2 ⊥)2 − e6 ⊥ + e4 ⊥ ) . k This expression is factorized into the three parts that have different origins: the first comes from the renormal- ization of the diffusion constant D in Eq. (40), the second is due to the vertex renomalization [see, Eqs. (52), (22) Α a 6 3 -0.5 b Α n
´1011cm-2
9 M= +5 meV M= 0 meV M=-5 meV M=-10 meV M=-15 meV -20 -10 M
meV
10 n=2´1011cm-2 n=2.5´1011cm-2 n=3´1011cm-2 n=3.5´1011cm-2 n=4´1011cm-2 -0.5 8 and (48)], and the third includes the contributions of the three Hikami boxes in Fig. 5b with the interplay of the singlet and triplet Cooperons [see, Eq. (52)]. It should be noted that in the presence of the k, σ → −k,−σ symme- try there is partial cancellation of these three factors41, yielding α = −1/2 for the symplectic disorder class31 (see, also Eqs. (75) and (76) for topological insulators in Sec. IV). Equation (53), as well as the equation for the conduc- tivity correction (1), is valid under conditions τ /τM = 2e2 ⊥ ≪ 1, τ /τϕ ≪ 1, (54) longer than the elastic life-time τ . when the carrier diffusion is limited by the time-scale, In min{τϕ, τM}, particular, for e2 ⊥ = 0 and τ /τϕ → 0 the parameter α → −1/2, and we recover the result δσxx = 2× e2 2πh ln τϕ τ for the symplectic class [the factor of 2 accounts for the two Dirac cones, see Eq. (6)]. For a finite e2 ⊥ the bro- ken k, σ → −k,−σ symmetry leads to the deviation of α from −1/2 [see, Fig. 6]. The deviation is quite signifi- cant because the expansion in powers of e2 ⊥ involves large numerical coefficients: α ≈ − 1 2(cid:26)1 − 17e2 ⊥ 2 − τ 2(cid:18) 1 τM + 1 τϕ(cid:19)(cid:18)1 + 35e2 ⊥ 2 (cid:19)(cid:27) .(55) This behavior can be seen as the crossover between the symplectic and unitary classes. IV. SURFACE STATES IN TOPOLOGICAL INSULATORS A. Effective Hamiltonian Unlike HgTe wells the spectrum of surface states in topological insulators (TIs), such as Bi2Se3 or Bi2Te3, consists of a single Dirac cone. Consequently, the effec- tive Hamiltonian for the surface state in TIs55,56 has the form of the single-block Hamiltonian of Eq. (6): HT I = H + Hi. (56) where H and Hi are given by Eqs. (7) and (9), respec- tively. There are two further distinctions between Hamil- tonians for Bi2Se(Te)3 and HgTe wells. First, here the basis functions correspond to 1 2 spin projections, i.e. Pauli (σx,y,z) and unit (σ0) matrices act on real spin indices. Second, the Mk-term in Hamiltonian H in Eq. (7) is cubic (odd) in momentum k:55,56 2 and − 1 Mk = W 2 (k3 + + k3 −), k± = kx ± iky, (57) FIG. 6: (Color online) Parameter α (53) versus carrier density n (a) and band gap M (b); A = 380 meV·nm, B = 850 meV·nm2 (from Ref. 16) and τ0/τϕ = 0.01. causing no gap at k = 0. This term does not break the k, σ → −k,−σ invariance, which is now the real time- reversal symmetry. Instead, it causes hexagonal warping 9 W=0 W=0 1 -1 1 -1 FIG. 8: (Color online) Polar plots of surface-state spectrum E(k, φn) (in units of energy Ak) as a function of momentum direction specified by angle φn ∈ (0, 2π) [see, also, Eq. (58)]. For W 6= 0 the spectrum shows hexagonal warping (we chose W 2k4/2A2 = 0.4 and Dk/A = 1). On average over all angles the warping results in larger E(k, φn) (dashed circle) com- pared to the W = 0 case (solid circle). B. Disorder-averaged Green's functions and scattering times As in the case of HgTe QWs, we assume that the Fermi level lies in the conduction band of the topological sur- face state [Fig. 7] and the condition of weak scatter- ing, Eq. (14), is fulfilled. With perturbative treatment of the warping described above, the retarded/advanced Green's functions and the elastic life-time for the surface state in a TI are given by Eqs. (19) -- (21). The time- reversal symmetry of the surface state is encoded in the odd momentum dependence of the out-of-plane vector e⊥n. There is no backscattering in this case63, and the elastic life-time (20) for isotropic ζkF ·n−n′ is τ = τ0. (61) In the similar way, Eq. (47) gives the transport scattering time as τtr = τ0 1 − e2 k /2 ≈ 2τ0 kF ·n/A2k2 F 1 + M2 . (62) Its dependence on ek (and, hence, on the variance M2 kF ·n) originates from the anisotropy of the Dirac- fermion scattering probability in momentum space. C. Cooperon To write the Cooperon equations (27) and (28) for the surface state in a TI we note that the time-reversal sym- metry implies the identities: e− = −e+, e+ · e− = −1, e+ − e− = 2e, e+ × e− = 0. (63) FIG. 7: (Color online) Energy bands of a topological surface state in meV [see, Eqs. (56), (57) and text] versus in-plane wave-numbers kx and ky in nm−1. The Fermi level lies in the conduction band at E = 0. We choose A = 300 meV·nm, W = 50 meV·nm3 (see, e.g., Ref. 56), D = 0 meV·nm2, and EF = 600 meV. of the surface-state spectrum55,56 (see, also Figs. 7 and 8): W 2k6 E(k, φn) =rA2k2 + (1 + cos 6φn) + Dk2. (58) We will treat the warping as weak perturbation onto the isotropic spectrum, assuming the smallness of the param- eter: 2 W 2k4 2A2 ≪ 1. (59) Then the main effect of the warping consists in the in- crease of E(k, φn) on average over all angles φn (see, dashed circle in Fig. 8). This amounts to replacing M2 by its angle average, k M2 k =⇒ M2 k = W 2k6 2 , (60) in Eq. (58). In fact, the same replacement can be done in all even functions of Mk, e.g. Fermi momentum, Fermi velocity, DOS etc. Then, the specific of the surface states is captured by the odd carrier chirality, σe−k = −σek, in Eq. (16). Therefore, the results of the integration over the single-particle momenta k (given in Appendices A and B) apply also in the present case. This allows us to use Eqs. (20) and (47) for the scattering times, Cooperon equations (27) and (28) as well as the Hikami boxes (45) and (46) to obtain the weak antilocalization conductivity corrections for the surface state in TIs. Equations (27) and (28) therefore reduce to he · biCbj , (64) C0j = τ0δ0j + h1iC0j + Xb=x,y,z Caj = τ0δaj + he · aiC0j + Xb=x,y,z h(e · a)(e · b)iCbj , or, explicitly, (65) (66) (67) (68) (69) [1 − h1i] C0j − hexiCxj − heyiCyj = τ0δ0j, − hexiC0j +(cid:2)1 − he2 xi(cid:3) Cxj = τ0δxj, yi(cid:3) Cyj = τ0δyj, − heyiC0j +(cid:2)1 − he2 (cid:2)1 − he2 ⊥i(cid:3) Czj = τ0δzj. In the above equations we use the short-hand notation e ≡ e+ for the unit vector ekF ·n whose in- and out- of-plane components are given by Eq. (21). Solving Eqs. (66) -- (69) for the diagonal Cooperon coefficients, we have C00(q, ω) = , D = 1 Dq2 − iω Cxx = Cyy = τtr(cid:18)1 − e2 k 4(cid:19) − Czz = τ0 1 − he2 ⊥i v2 F τtr 2 , (70) e2 k 4 iτtrωC00(q, ω),(71) . (72) Note that the singlet Cooperon C00 (70) remains gap- less also in the presence of the warping because it does not break the time-reversal symmetry. The warping only modifies the diffusion constant D through the transport scattering time (62). The triplet Cooperons Cxx and Cyy are already averaged, for presentation purposes, over the directions of q. In the absence of the symmetry breaking [cf. Eqs. (41) and (42)] and for τtr/τϕ ≪ 1, the triplet Cooperons Cxx and Cyy as well as Czz can be neglected compared to C00 in the conductivity corrections. D. Hikami boxes and net conductivity correction Repeating the calculations described in subsec- tion III D, we express the conductivity corrections δσ(1) xx (45) and δσ(2) xx (46) in terms of the diagonal Cooperons: δσ(1) xx = e2Dτ dq τ (cid:17)2Z π (cid:16) τtr xiCxx − 2hn2 (2π)2(cid:2)2hn2 xiC00 yiCyy − 2hn2 xe2 xe2 − 2hn2 xe2 ⊥iCzz(cid:3),(73) a B b z B surface state surface state 10 λ FIG. 9: (Color online) Schematic view of a 3D topological insulator with a surface state subject to (a) perpendicular and (b) parallel magnetic field B. In the latter case the B-field dependence of the surface weak-antilocalization conductivity is due to the magnetic flux through the effective thickness of the surface state, determined by the decay length, λ, into the bulk. δσ(2) xx = − e2Dτ π (cid:16) τtr τ (cid:17)2(cid:16) ek 2(cid:17)2Z dq (2π)2(cid:2)C00 − Cxx(cid:3),(74) Keeping only the singlet Cooperon C00, we obtain the following expression for the net correction: δσxx = δσ(1) e2Dτ xx + 2δσ(2) xx = = k 2(cid:17)Z e2 π (cid:16) τtr π Z τ (cid:17)2(cid:16)1 − dq (2π)2 C00(q). e2D dq (2π)2 C00(q) (75) (76) k After the partial cancellation in Eq. (75): Dτ (τtr/τ )2(1− e2 /2) = D (τtr/τ ) (τ /τtr) = D, the prefactor in the final equation (76) depends only on the transport scattering time through the diffusion constant D [see, Eq. (70) and Ref. 41]. Inserting Eq. (70) into Eq. (76), integrating over q with the upper cut-off (Dτ )−1/2, and replacing −iω → τ −1 ϕ , we obtain Eq. (4), already discussed in Sec. I. Below we will focus on the dependence of δσxx on the strength and oriention of an external magnetic field B, which can be used in experiments to identify the surface states in three-dimensional topological insulators. E. Magnetoconductivity in perpendicular field For B applied perpendicularly to the surface (see Fig. 9a) we write Eq. (70) for the singlet Cooperon in the two-dimensional position representation,65 "D(cid:18)i∇ + 2e  A(r)(cid:19)2 − iω# C00(r, r′) = δ(r − r′), (77) including the vector potential A(r) = (−By, 0, 0) of the magnetic field (r = (x, y, 0)). The solution is given by the Hilbert-Schmidt expansion in the Landau wave-function basis, which yields the well-known expression65 for the magnetoconductivity ∆σxx(B) = δσxx(B) − δσxx(0): e2 2πh"ln B⊥ B − ψ 1 2 + B⊥ B !#, ∆σxx(B) = B⊥ =  2e ℓtrℓϕ , (78) (79) 11 DΣxx -20 -10 10 20 B
B¦ -1 B ¦ surface B
surface where the magnetic field B⊥ corresponds to the Aharonov-Bohm flux of order of h/e through a typi- cal area encircled by the interfering trajectories,65 ℓtr = vF τtr and ℓϕ = vF τϕ, and ψ(x) is the digamma function. FIG. 10: (Color online) Quantum-interference magnetocon- ductivity ∆σxx(B) [in units of e2/2πh, see Eqs. (78) and (84)] for in- and out-of-plane field orientations. The mag- netic field scales, on which ∆σxx(B) decreases, are fixed such that Bk /B⊥ = 2pℓtrℓϕ/λ = 20. F. Magnetoconductivity in parallel field In the case of the parallel magnetic field B (see Fig. 9b) the vector potential, A(z) = (B× z)z, depends explicitly on the coordinate z perpendicular to the surface (z is the unit vector). Therefore, the penetration of the surface state into the bulk needs to be taken into account. To do so we first transform the diffusion operator in equa- tion [Dq2 − iω]C00(q) = 1 into the three-dimensional position representation and then make the Peierls sub- stitution i∇r → i∇r + 2e a−1R drdze−iqrχ∗(z)[D(i∇r + 2e  A(z))2 − iω]eiqrχ(z) (80)  A(z) as follows ×C00(q) = 1, where the in-plane (r) integration goes over the surface area a, and the out-of-plane (z) integral involves the nor- malized wave function of the surface state, χ(z), which decays exponentially for z → −∞ on the length λ in- versely proportional to the bulk band gap:20 χ(z) = = . (81) ez/λ pλ/2 e−z/λ pλ/2 In the latter form (i.e. written with z) this equation can formally be used in the entire space −∞ < z < ∞. This helps to simplify further calculations because the system is symmetrically extended to the other half-space, 0 < z < ∞, and the z integral in Eq. (80) can be calculated as −∞ dz.... Having done this integration, we return to q representation of the Cooperon: R dz... = (1/2)R ∞ C00(q, ω) = Dq2 + τ −1 1 B − iω , τ −1 B = 2De2B2λ2/2 = Dλ2/2ℓ4 B, (82) (83) where τB is the time-scale for the suppression of the quan- tum interference by the magnetic flux through the thick- ness of the surface state, λ. The latter is assumed much smaller than the magnetic length ℓB =p/2eB. Note that Eq. (83) has a six-time larger numerical coefficent than the result for the quasi-2D quantum wells,57 which reflects the difference in the electron confinement. Using Eqs. (76) and (82) we calculate the magneto- conductivity ∆σxx(B) = δσxx(B) − δσxx(0): ∆σxx(B) = − = − e2 2πh e2 2πh Bk = 2pℓtrℓϕ λ ln 1 + ln 1 + B⊥. τϕ τB! k!, B2 B2 (84) (85) 2pℓtrℓϕ the in- and out-of-plane geometries have dis- Clearly, for a sufficiently small penetration length λ ≪ tinctly different magnetic-field scales, Bk ≫ B⊥, on which ∆σxx(B) decreases with B [see Fig. 10]. The in- plane magnetoconductivity ∆σxx(B) (84) can be veri- fied against recent magnetotransport measurements on Bi2Te3 (Ref. 52). Acknowledgments We thank L. W. Molenkamp, H. Buhmann, P. Brouwer, B. Trauzettel, P. Recher, C. Brune, C. Gould, C.-X. Liu and B. Buttner for helpful discussions. Appendix A: Derivation of Cooperon equations (25) equation αβΨj coefficients Cij α′β′ Cαβα′β′ follows from Eq. (23): the for The Pαβα′β′ Ψi∗ Cij (q) = τ 2 τ0 δij +  πN τ0Xs Z dk (2π)2h Xαβγ′δ′ R αβG αγ′(k, ǫ + ω)G ×Ψi∗ A βδ′(q − k, ǫ)Ψs γ′δ′iCsj (q). = (A1) It can be rewritten in a simpler form with the help of the identity Here the tilde denotes the operation 12 Xαβγ′δ′ Xδ′βαγ′hG = TrhG R Ψi∗ αβG αγ′ (k, ǫ + ω)G A βδ′(q − k, ǫ)Ψs γ′δ′ = A T (q − k, ǫ)iδ′β(cid:2)Ψi†(cid:3)βα G AT (q − k, ǫ)Ψi†G R (k, ǫ + ω)Ψsi , R αγ′ (k, ǫ + ω)Ψs γ′δ′ (A2) where Tr and T denote the trace and transposition oper- ations, respectively, in σ space. We therefore have Cij(q) = τ 2 τ0 δij + AT ×G (q − k, ǫ)Ψi†G  dk TrhZ πN τ0Xs (k, ǫ + ω)ΨsiCsj(q).(A3) (2π)2 R For the orthonormal basis functions given by Ψj = σjσy √2 , j = 0, x, y, z, (A4) Eq. (A3) assumes the following form: Cij(q) = τ 2 τ0 δij + R fGA(q − k, ǫ)σiG  dk TrhZ 2πN τ0Xs (k, ǫ + ω)σsiCsj(q). (2π)2 AiT fGA = σyhG σy, (A6) which flips the pseudospin: eσ = σy σT σy = −σ. Thus, for q → 0 Eq. (A5) describes interference between the state with k, σ and its "time-reversed" partner with −k,−σ. Next, we evaluate the k integral in Eq. (A5) using the sharpness of the Green's functions at the Fermi level un- der condition (14) and the smallness of the Cooperon momentum q and frequency ω [see, Eq. (38)]. In the de- q−k it is sufficient to keep only the linear term in q, i.e. ξq−k ≈ ξk − vq + ... (which should be compared with energy /τ ): nominator of fGA q−k ≈ 1 2 σ0 − σe−k ǫA − ξk + vq . fGA (A7) the same At proximate eq−k ≈ e−k, neglecting small in the numerator we time, ap- terms F + M2 kF ∼ Aq/EF ≪ 1. Aq/qA2k2 With Eqs. (19), (A7) and under condition (14) the k (A5) integral can be evaluated as follows Z ≈ = = R 1 2π (k, ǫ + ω)σs = dk (2π)2 GA(q − k, ǫ)σiG 4 Z dφn N 4 Z dφn 22Z dφn (σ0 − σe−kF ·n)σi(σ0 + σekF ·n)σsZ dξk (σ0 − σe−kF ·n)σi(σ0 + σekF ·n)σs 4Z dφn 1 (σ0 − σe−kF ·n)σi(σ0 + σekF ·n)σs 1 − iτ ω + iτ vF n · q  × 2πN τ 2π 2π N , 2π Z N (ξk, n)dξk 1 (σ0 − σe−k)σi(σ0 + σek)σs (ǫA − ξk + vq)(ǫR − ξk) (ξk − ǫA − vF n · q)(ξk − ǫR) 2πi i/τ + ω − vF n · q (A8) (A9) (A10) (A11) where the DOS N (ξk, n) is replaced by its Fermi surface value N and, then, the ξk integral is calculated in the complex plane. Inserting Eq. (A11) into Eq. (A5) yields Eq. (25). Appendix B: Evaluation of k integrals in Hikami boxes in Eqs. (43) and (44) To calculate the k integral in Eq. (43) we first expand the Green's functions GR q−k,ǫ in small Cooperon momentum q as we did in Eq. (A7) of Ap- pendix A: q−k,ǫ+ω and GA R G q−k,ǫ+ω ≈ 1 2 σ0 + σe−k ǫR − ξk + ω + vq , (B1) A q−k,ǫ ≈ G 1 2 σ0 + σe−k ǫA − ξk + vq (B2) Then, the integral is converted to R dφn the ξk integration is again done in the complex plane following the same steps as in Eqs. (A8) -- (A11). The final result is 2π R N dξk... and Z dk (2π)2 (G A k,ǫ V x R k G k,ǫ+ω)γ′β (G R q−k,ǫ+ω V x q−k G A ×[(σ0 + σekF ·n)V x 4πN τ 3 1 24 q−k,ǫ)γβ′ ≈ kF ·n(σ0 + σekF ·n)]γ′β[(σ0 + σe−kF ·n)V x 3 −kF ·n(σ0 + σe−kF ·n)] 13 (B3) γβ′, where (...) =R 2π ...dφn/2π is the averaging over the mo- mentum direction n. As the expression above is indepen- dent of ǫ, the energy integral in Eq. (43) is /2π, which along with Eq. (B3) yields Eq. (45). 0 To evaluate the integrals over k and k′ in Eq. (44) we set q, ω → 0 in the single-particle Green's functions and re-group them as follows Z dk (2π)2Z dk′ (2π)2 R A (G k G k,ǫ V x ≈Z dk (2π)2 [G R R k,ǫ+ωG q−k′,ǫ+ω)γ′β(G q−k,ǫ+ω G A R (k)V x(k)G (k)]γ′β1G R γγ1(−k)Z R k′,ǫ+ω V x k′ G A k′,ǫ)γβ′ ≈ β1β(−k′)[G R R (B4) (k′)V x(k′)G A (k′)]γ1β′. dk′ (2π)2 G Each of the integrals can now be done in the similar way as in Eqs. (A8) -- (A11) of Appendix A: Z A dk (2π)2 [G R (k)V x(k)G (k)]γ′β1 G R γγ1(−k) ≈ − 2πiN τ 2 2 1 23 Z dk′ (2π)2 G R β1β(−k′)[G R (k′)V x(k′)G A (k′)]γ1β′ ≈ − 2πiN τ 2 2 1 23 × [(σ0 + σekF ·n)V x kF ·n(σ0 + σekF ·n)]γ′β1(σ0 + σe−kF ·n) , (B5) γγ1 × (σ0 + σe−kF ·n′)β1β[(σ0 + σekF ·n′)V x kF ·n′(σ0 + σekF ·n′)] γ1β′.(B6) Inserting these into Eq. (44) we obtain Eq. (46). 1 K. S. Novoselov, A. K. Geim, S. V. Morozov, D. Jiang, Y. Zhang, S. V. Dubonos, I. V. Grigorieva and A. A. Firsov, Science 306, 666 (2004). 2 for a review, see A.H. Castro Neto, F. Guinea, N.M. Peres, K.S. Novoselov, and A.K. Geim, Rev. Mod. Phys. 81, 109 (2009) and references therein. 3 C. L. Kane and E. J. Mele, Phys. Rev. Lett. 95, 226801 (2005). 4 B. A. Bernevig and T. L. Hughes and S. C. Zhang, Science 314, 1757 (2006). 5 M. Konig, S. Wiedmann, C. Brune, A. Roth, H. Buhmann, L. W. Molenkamp, X.-L. Qi and S.-C. Zhang, Science 318, 766 (2007). 6 A. Roth, C. Brune, H. Buhmann, L. W. Molenkamp, J. Maciejko, X.-L. Qi, and S.-C. Zhang, Science 325, 294 (2009). 7 L. Fu and C. L. Kane, Phys. Rev. B 76, 045302 (2007). 8 J. E. Moore and L. Balents, Phys. Rev. B 75, 121306(R) (2007). 9 D. Hsieh, D. Qian, L. Wray, Y. Xia, Y. S. Hor, R. J. Cava, and M. Z. Hasan, Nature 452, 970 (2008). 10 D. Hsieh, Y. Xia, L. Wray, D. Qian, A. Pal, J. H. Dil, F. Meier, J. Osterwalder, C.L. Kane, G. Bihlmayer, Y. S. Hor, R. J. Cava and M.Z. Hasan. Science 323, 919 (2009). 11 Y. Xia, D. Qian, D. Hsieh, L. Wray, A. Pal, H. Lin, A. Bansil, D. Grauer, Y. S. Hor, R. J. Cava, and M. Z. Hasan, Nature Phys. 5, 398 (2009). 12 Y. L. Chen, J. G. Analytis, J.-H. Chu, Z. K. Liu, S.-K. Mo, X. L. Qi, H. J. Zhang, D. H. Lu, X. Dai, Z. Fang, S. C. Zhang, I. R. Fisher, Z. Hussain, and Z.-X. Shen, Science 325, 178 (2009). 13 for a recent review on topological insulators, see M. Z. Hasan and C. L. Kane, Rev. Mod. Phys. 82, 3045 (2010) and references therein. 14 for a recent review on topological insulators and supercon- ductors, see X.-L. Qi and S.-C. Zhang, arXiv:1008.2026 and references therein. 15 see, e.g., G. W. Semenoff, Phys. Rev. Lett. 53, 2449 (1984). 16 B. Buttner, C. X. Liu, G. Tkachov, E. G. Novik, C. Brune, H. Buhmann, E. M. Hankiewicz, P. Recher, B. Trauzettel, S. C. Zhang and L. W. Molenkamp, Nature Phys. 7, 418 (2011). 14 17 G. Tkachov, C. Thienel, V. Pinneker, B. Buttner, C. Brune, H. Buhmann, L. W. Molenkamp, and E. M. Han- kiewicz, Phys. Rev. Lett. 106, 076802 (2011). 18 G. Tkachov and E. M. Hankiewicz, Phys. Rev. Lett. 104, 166803 (2010). 148, 39 (2007). 44 D. Neumaier, K. Wagner, S. Geiler, U. Wurstbauer, J. Sad- owski, W. Wegscheider, and D. Weiss, Phys. Rev. Lett. 99, 116803 (2007). 45 V. A. Guzenko, T. Schapers, and H. Hardtdegen, Phys. 19 G. Tkachov and E. M. Hankiewicz, Phys. Rev. B 83, Rev. B 76, 165301 (2007). 155412 (2011). 20 B. A. Volkov and O. A. Pankratov, Pis'ma Zh. Eksp. Teor. Fiz. 42, 145 (1985) [JETP Lett. 42, 178 (1985)]. 21 X.-L. Qi, T. L. Hughes, and S.-C. Zhang, Phys. Rev. B 78, 195424 (2008). 22 A. M. Essin, J. E. Moore, and D. Vanderbilt, Phys. Rev. Lett. 102, 146805 (2009). 23 W.-K. Tse and A. H. MacDonald, Phys. Rev. Lett. 105, 057401 (2010). 46 K.-I. Imura, Y. Kuramoto, and K. Nomura, Phys. Rev. B 80, 085119 (2009); Europhys. Lett. 89, 17009 (2010). 47 J. Wurm, A. Rycerz, I. Adagideli, M. Wimmer, K. Richter, and H. U. Baranger, Phys. Rev. Lett. 102, 056806 (2009). 48 E. B. Olshanetsky, Z. D. Kvon, G. M. Gusev, N. N. Mikhailov, S. A. Dvoretsky and J. C. Portal, JETP Lett. 91, 347 (2010). 49 B. Buttner, C. Brune, H. Buhmann, and L. W. Molenkamp (private communication). 24 W.-K. Tse and A. H. MacDonald, Phys. Rev. B 82, 50 J. G. Checkelsky, Y. S. Hor, R. J. Cava, and N. P. Ong, 161104(R) (2010). 25 J. Maciejko, X.-L. Qi, H. D. Drew, and S.-C. Zhang, Phys. Rev. Lett. 105, 166803 (2010). 26 I. Garate and M. Franz, Phys. Rev. Lett. 104, 146802 (2010). 27 G. Tkachov and E. M. Hankiewicz, Phys. Rev. B 84, 035405 (2011). 28 F. Wilczek, Phys. Rev. Lett. 58, 1799 (1987). 29 S. V. Morozov, K. S. Novoselov, M. I. Katsnelson, F. Schedin, L. A. Ponomarenko, D. Jiang, and A. K. Geim, Phys. Rev. Lett. 97, 016801 (2006). 30 F. V. Tikhonenko, A. A. Kozikov, A. K. Savchenko, and R. V. Gorbachev, Phys. Rev. Lett. 103, 226801 (2009). 31 S. Hikami, A.I. Larkin, and N. Nagaosa, Prog. Theor. Phys. 63, 707 (1980). 32 S. V. Iordanskii, Yu. B. Lyanda-Geller, and G. E. Pikus, Pis'ma Zh. Eksp. Teor. Fiz. 60, 199 (1994) [JETP Lett. 60, 206 (1994)]. 33 Y. B. Lyanda-Geller and A. D. Mirlin, Phys. Rev. Lett. 72, 1894 (1994). 34 W. Knap, C. Skierbiszewski, A. Zduniak, E. Litwin- Staszewska, D. Bertho, F. Kobbi, J. L. Robert, G. E. Pikus, F. G. Pikus, S. V. Iordanskii, V. Mosser, K. Zekentes, and Yu. B. Lyanda-Geller, Phys. Rev. B 53, 3912 (1996). 35 A. Altland and M. R. Zirnbauer, Phys. Rev. B 55, 1142 (1997). 36 I. L. Aleiner and V. I. Falko, Phys. Rev. Lett. 87, 256801 (2001). 37 P. W. Brouwer, J. N. H. J. Cremers, and B. I. Halperin, Phys. Rev. B 65, 081302(R) (2002). 38 H. Suzuura and T. Ando, Phys. Rev. Lett. 89, 266603 (2002). 39 J. B. Miller, D. M. Zumbuhl, C. M. Marcus, Y. B. Lyanda- Geller, D. Goldhaber-Gordon, K. Campman, and A. C. Gossard, Phys. Rev. Lett. 90, 076807 (2003). 40 O. Zaitsev, D. Frustaglia, and K. Richter, Phys. Rev. Lett. 94, 026809 (2005). 41 E. McCann, K. Kechedzhi, V.I. Fal'ko, H. Suzuura, T. Ando, and B.L. Altshuler, Phys. Rev. Lett. 97, 146805 (2006). 42 I. L. Aleiner and K. B. Efetov, Phys. Rev. Lett. 97, 236801 (2006). 43 K. Kechedzhi, E. McCann, V. I. Fal'ko, H. Suzuura, T. Ando, and B. L. Altshuler, Eur. Phys. J. Special Topics arXiv:1003.3883. 51 J. Chen, H. J. Qin, F. Yang, J. Liu, T. Guan, F. M. Qu, G. H. Zhang, J. R. Shi, X. C. Xie, C. L. Yang, K. H. Wu, Y. Q. Li, and L. Lu, Phys. Rev. Lett. 105, 176602 (2010). 52 H.-T. He, G. Wang, T. Zhang, I.-K. Sou, G. K. L. Wong, J.-N. Wang, H.-Z. Lu, S.-Q. Shen, and F.-C. Zhang, Phys. Rev. Lett. 106, 166805 (2011). 53 V. K. Dugaev, P. Bruno, and J. Barnas, Phys. Rev. B 64, 144423 (2001). 54 I. L. Aleiner, B. L. Altshuler, and M. E. Gershenson, Waves Random Media 9, 201 (1999). 55 L. Fu, Phys. Rev. Lett. 103, 266801 (2009). 56 C.-X. Liu, X.-L. Qi, H. J. Zhang, X. Dai, Z. Fang, S.-C. Zhang, Phys. Rev. B 82, 045122 (2010). 57 B. L. Altshuler and A. G. Aronov, Pis'ma Zh. Eksp. Teor. Fiz. 33, 515 (1981) [JETP Lett. 33, 499 (1981)]. 58 J. S. Meyer, A. Altland, and B. L. Altshuler, Phys. Rev. Lett. 89, 206601 (2002). 59 D. G. Rothe, R. W. Reinthaler, C.-X. Liu, L. W. Molenkamp, S.-C. Zhang, and E. M. Hankiewicz, New J. Phys. 12, 065012 (2010). 60 Inversion symmetry breaking leads to off-diagonal terms in Eq. (6), which is another difference compared, e.g., to the topological surface states. However, such terms are sig- nificantly smaller than the linear (Dirac) term in Hamil- tonian (7), e.g. the bulk inversion asymmetry (BIA) term ∆BIA ∼ 1 meV, whereas AkF ≈ 50 − 60 meV for carrier densities n = (2.5 − 3.5) × 1011 cm−2 considered in this paper. Therefore, the block-diagonal model (6) is a rea- sonable approximation. For more details on BIA see M. Konig, H. Buhmann, L. W. Molenkamp, T. Hughes, C.- X. Liu, X.-L. Qi, and S.-C. Zhang, J. Phys. Soc. Jpn. 77, 031007 (2008). 61 A. A. Abrikosov, L. P. Gorkov and I. E. Dzyaloshinski, Methods of Quantum Field Theory in Statistical Physics (Prentice-Hall, Englewood Cliffs, NJ, 1963). 62 J. Rammer, Quantum Transport Theory (Westview Press, 2004). 63 T. Ando, J. Phys. Soc. Jpn. 74, 777 (2005). 64 Under conditions (54) the contribution of the first terms in Eqs. (41) and (42) is smaller than the logarithmic cor- rection (1). 65 B. L. Altshuler, D. Khmel'nitzkii, A. I. Larkin and P. A. Lee, Phys. Rev. B 22, 5142 (1980). 

1612.08602

2

1612

Observation of Acoustic Valley Vortex States and Valley-Chirality Locked Beam Splitting

The Letter reports an experimental observation of the classical version of valley polarized states in a two-dimensional hexagonal sonic crystal, where the inversion-symmetry breaking of scatterers induces an omnidirectional frequency gap. The acoustic valley states, which carry specific linear momenta and orbital angular momenta, were selectively excited by external Gaussian beams and conveniently confirmed by the pressure distribution outside the crystal, according to the criterion of momentum conservation. The vortex nature of such intriguing crystal states was directly characterized by scanning the phase profile inside the crystal. In addition, we observed a peculiar beam splitting phenomenon, in which the separated beams are constructed by different valleys and locked to the opposite vortex chirality. The exceptional sound transport, encoded with valley-chirality locked information, may serve as the basis of designing conceptually novel acoustic devices with unconventional functions.

cond-mat.mes-hall

cond-mat

Observation of Acoustic Valley Vortex States and Valley-Chirality Locked Beam Splitting Liping Ye,1 Chunyin Qiu,1* Jiuyang Lu,4 Xinhua Wen,1 Yuanyuan Shen,1 Manzhu Ke,1 Fan Zhang,3 and Zhengyou Liu1,2 1Key Laboratory of Artificial Micro- and Nano-structures of Ministry of Education and School of Physics and Technology, Wuhan University, Wuhan 430072, China 2Institute for Advanced Studies, Wuhan University, Wuhan 430072, China 3Department of Physics, University of Texas at Dallas, Richardson, Texas 75080, USA 4Department of Physics, South China University of Technology, Guangzhou 510641, China Abstract: We report an experimental observation of the classical version of valley polarized states in a two-dimensional hexagonal sonic crystal, where the inversion-symmetry breaking of scatterers induces an omnidirectional frequency gap. The acoustic valley states, which carry specific linear momenta and orbital angular momenta, were selectively excited by external Gaussian beams and conveniently confirmed by the pressure distribution outside the crystal, according to the criterion of momentum conservation. The vortex nature of such intriguing crystal states was directly characterized by scanning the phase profile inside the crystal. In addition, we observed a peculiar beam splitting phenomenon, in which the separated beams are constructed by different valleys and locked to the opposite vortex chirality. The exceptional sound transport, encoded with valley-chirality locked information, may serve as the basis of designing conceptually novel acoustic devices with unconventional functions. PACS numbers: 42.70.Qs, 43.20.+g, 63.20.−e *Corresponding author. cyqiu@whu.edu.cn 1 / 12 Sound is one of the most common waves in daily life. However, sound wave is not easy to control by external fields, e.g., electric and magnetic fields, since it lacks intrinsic degrees of freedom (DOFs) like the charge and spin in electrons. As a fundamental property of waves, sound can be scattered or diffracted when suffering inhomogeneity in the propagation path. This offers an efficient route of steering sound by artificial structures, e.g., sonic crystals (SCs) [1-4], acoustic metamaterials [5-8] and metasurfaces [9-12]. By utilizing these unnatural media, the propagation of sound can be tailored in unprecedented ways, such as negative refraction [1-3,13-15], super-resolution imaging [3,16,17], cloaking [18-20], and abnormal wavefront shaping [9-12]. Recently, the acoustic structures have also been demonstrated to be good platforms in exploring topological physics predicted originally in electronic systems [21-28]. In condensed matter systems, the property of valley electrons has sparked extensive interest in recent years [29-43]. The discrete valley index, labeling the degenerate energy extrema in momentum space, can be treated as a new quantum DOF other than charge and spin when the intervalley scattering is negligibly weak. Numerous fascinating phenomena associated with valley-contrasting properties have been studied, such as valley filters and valley Hall effects, which are paving the road for the applications of valleytronics (e.g., in quantum computing and information processing). Inspired by the concept of valley DOF in electronic states, recently, Lu et al. have theoretically studied the valley states of acoustic waves in two-dimensional (2D) SCs, and revealed that such states carry notable vortex signatures [44]. Particularly, the acoustic valley (AV) states locked to specific vortex chirality can be independently accessed by external sound according to their linear or angular momenta. Therefore, analog to the electronic case, the additional valley DOF in the artificially designed SC structure, associated with the valley-chirality locking property, could also construct a good carrier of information and thus enables a brand new manner to control sound. In this Letter, we present the first experimental observation of the AV vortex states in a 2D hexagonal SC. Valley-selective excitation has been realized by an 2 / 12 external Gaussian beam at a particular incident angle, and confirmed by the Fourier spectrum of the pressure field outside the SC sample, together with an elaborate characterization of the vortex chirality through detecting the phase profile around the vortex core. Furthermore, we demonstrate a unique beam splitting behavior, where two spatially separated beams are locked to different vortex chirality. The experimental data agree fairly well with the full-wave simulations performed by COMSOL Multiphysics. As such, by exciting the AV states we have experimentally demonstrated a novel manipulation on sound: making vortex arrays and controlling their chirality according to the moving directions. In addition, as a natural property of the chiral phased waves, the sound vortices carry orbital angular momentum (OAM) inherently [45], and thus enable many tantalizing applications, e.g., to generate mechanical torques on the trapped objects by delivering the OAM of sound to matter [46-52]. FIG. 1. (a) A local photograph of the SC sample made of a hexagonal array of regular triangle plexiglass rods, with p and q labeling two inequivalent lattice centers. (b) The lowest two bands plotted around the point. (c) Eigenfields for the valley states and . The arrow indicates the orientation of vortex-like energy flow. (d) Schematic of the experimental setup. All measurements are performed inside a 2D air-filled waveguide formed by two parallel plexiglass plates. (e) Phase scanning 3 / 12 pqa2a15.77cmqppq-0.4-0.20.00.20.42.83.03.23.43.6-0.4-0.20.00.20.4Frequency (kHz)kx (/a)ky (/a)K2K1Y1n1nqpmax00probeK1-stateK2-statePhase/Energy FlowAmplitude(b)(c)(d)(e)090180270(a)XK1K2Kalong a circular loop by counterclockwise rotating the probe, with labeling the azimuthal angle. As shown in Fig. 1(a), the SC is built by a 2D hexagonal array of regular triangle plexiglass rods. The length of lattice vectors and the side length of each rod . These dimensions are carefully designed to allow a field scanning inside the unit cell, meanwhile considering a trade-off with the total sample size limited by our measuring platform. (Principally, a sample containing more unit cells is preferred to reducing the finite size effect.) The orientation DOF of the anisotropic rod, depicted by the rotation angle , offers a flexible control of the band gap [53]. Specifically, here is utilized to open a sizable omnidirectional band gap between the lowest two bands [Fig. 1(b)]. Below we focus on the AV states (3.06 kHz) and (3.49 kHz) that locate at the corner point of the first Brillouin zone. Similar to the orbital motion of valley electrons, each state is featured by a hexagonal array of sound vortices [Fig. 1(c)], centered at the equivalent lattice centers q (or p) and rotated counterclockwise (or clockwise). Intriguingly, the AV vortex states carry quantized phase winding numbers ( ), owing to the threefold rotation symmetry of the system [28,44]. The counterparts at the inequivalent valley , possess invariant vortex cores (with zero amplitudes) but opposite chirality, as required by time-reversal symmetry. It has been pointed out that [44], different from the electronic case where the excitation of the valley polarized states resorts to additional fields [54-60], the AV states can be directly accessed by sound stimuli, either by Gaussian beams based on the principle of momentum conservation, or by point-like chiral sources according to the azimuthal selection rule. Here we employ the first approach since a Gaussian beam can be prepared easily in experiment. Figure 1(d) shows our experimental setup. The sample contains 276 rods in total and has a shape of regular triangle with side lengths 130 cm. Its surface normal is selected along the direction to avoid intervalley scattering at the SC boundary [44]. Thanks to the macroscopic characteristic of the 4 / 12 125.77 cmaaa3.0 cmlo301K2KK1nKMsystem, the triangular plexiglass rods, fabricated by laser-cutting, can be orientated and arranged into the hexagonal lattice precisely, which enables the valley transport nearly free of short range scattering inside the SC. To guarantee the experimental system of 2D nature, the whole structure is positioned in an air-filled planar sound waveguide (of height 1.2 cm) formed by two plexiglass plates, which allows only the propagation of the fundamental waveguide mode for the concerned wavelength (>9.3 cm). The point-like sound signal, launched from a narrow tube with inner diameter ~0.8 cm, is transferred into a Gaussian beam when it is reflected by a carefully designed parabolic concave mirror [61]. The width of the Gaussian beam is controlled by the geometry of the concave mirror. The pressure field inside or outside the SC sample can be scanned by a movable microphone of diameter ~0.7 cm (B&K Type 4187). Specifically, to confirm the vortex chirality of the AV states, a small hole (of diameter ~1.1 cm) is drilled through the top plate, and the probe microphone is obliquely inserted into the waveguide. As illustrated in Fig. 1(e), by rotating the probe (at an angular step of ) one can detect the pressure response along a circular contour centered at a given lattice center (e.g., p point for the state here). Finally, the sound signal is analyzed by a multi-analyzer system (B&K Type 3560B), from which the phase and amplitude of the pressure field can be extracted simultaneously. FIG. 2. (a) Pressure patterns measured and simulated outside the sample, stimulated by an obliquely incident Gaussian beam at the frequency. The green arrows 5 / 12 o302K0.00.51.01.52.00.00.51.01.52.02.53.0-maxkx (/a)Phase ()max0102030 Y (cm)(a)(c)Exp.0204060801000102030 Y (cm)X (cm)Sim.-1.0-0.50.00.51.00102030 K0 Y (cm)Sim.K0102030 maxY (cm)(b)Exp.2Kindicate the propagation of sound wavefronts. (b) The corresponding spatial Fourier spectra performed along the x direction. (c) Measured (circles) and simulated (lines) phase distributions for three different circular loops encircling the equivalent p points. The error bars represent the standard deviation of the measurement. To experimentally stimulate the state, we launch a Gaussian beam of width ~23cm onto the SC sample. Specifically, the incident angle satisfies , where is the wavenumber in air space and is a projection of the point to the sample boundary. By this incidence, the state is anticipated to be well excited owing to the momentum conservation parallel to the SC boundary. Instead of directly scanning the whole field inside the sample, an easier task is carried out to validate the valley selection, i.e., detecting the sound signal leaked out. For example, we consider the spatial region earmarked by the rectangle in Fig. 1(d). As displayed visually by Fig. 2(a), the experimentally measured wavefront pattern agrees well with the COMSOL simulation. A nearly pure excitation of the state can be further checked from the corresponding Fourier spectra performed parallel to the horizontal sample boundary [Fig. 2(b)]. As predicted by the theory, the experimental data shows a bright stripe at , where the momentum broadening stems mostly from the finite size effect. This is in striking contrast to the position , a projection of the point. Figure 2(c) exemplifies the phase profiles for three circular loops located in different unit cells. In all cases, the experimental measurements, in good agreement with the simulations, exhibit a phase reduction of over the loop, another critical signature for the excitation of the state. Note that the quantized phase winding number ( ) is easy to capture experimentally, since the vortex singularity is generic and structurally stable in real space [45]. Such immunity to noise perturbation is appealing in practical applications. Similarly, the state (with ) can also be excited and detected independently (see Supplementary Information). 6 / 12 2Ko340sinkK0k23KaK2K2KxkKxkKK22K1n1K1nThe above study states that, by selectively exciting a single valley state, one can realize a hexagonal lattice of sound vortices. Their chirality can be controlled by the incident frequency and direction of the Gaussian beam. As such, the distinguishable valley signature, i.e., valley-chirality locking, enables a new DOF to manipulate sound. This is greatly meaningful for scalar acoustics that lacks intrinsic DOFs like charge and spin. In addition, as a fundamental morphology of sound profile, the vortex matter of sound (imprinted with nonzero OAM) is attracting fast growing attention due to its promising applications [46-52]. Usually, a sound vortex is realized by an array of active sound sources with elaborate phase lags [46-51,62-64]. Recently, artificial structures, e.g., spiral gratings [65-67] and metamaterials [68,69], have been proposed to make sound vortices. Strikingly different from these passive schemes, which produce only a single vortex that travels spirally along its axis, the current method generates a compact array of vortices simultaneously, and the energy transports in the 2D lattice plane. FIG. 3. (a) Trigonal warping effect of the EFC (purple solid lines), illustrated by a frequency slightly above . The red dashed line indicates the circular EFC of air at the same frequency. (b) Amplitude distribution of the pressure field simulated for a narrow Gaussian beam incident normally from the bottom. (c) Experimentally measured pressure amplitudes along the labeled horizontal lines. The red arrows guide the propagation of the split sub-wave-packets. (d) Phase profile (circles) scanned along two circular loops located on the left- and right-moving beams [see green points in (b)], together with the numerical data (lines) for comparison. Below we demonstrate another fascinating manipulation of the AV vortex states, 7 / 12 0.00.40.81.21.62.00.00.51.01.52.0(a)gvgvgv3.66kHzgvgvgvKKKKKK0max-80-60-40-20020406080 X (cm) (b)(c)KKKPhase ()(d)K2Ki.e., stimulating the valleys and simultaneously and separating them in different spatial regions. This valley-dependent beam splitting stems essentially from a trigonal warping effect of the band structure [44,70]. As exemplified by the EFCs of 3.66 kHz [Fig. 3(a)], if an incident beam covers a broad range of momentum distribution, the forward-moving states around the and points will be excited at the same time. Considering the finite lengths of the trigonal EFCs in momentum space, the beams constructed by these states must be finite widths, and separate as they propagate because of the different group velocities [ depicted by the arrows in Fig. 3(a)]. This is displayed clearly in Fig. 3(b). Interestingly, each beam carries a chiral feature of the corresponding valley, since the vortex-like field profile remains even for the state off the band edge. To experimentally confirm the valley-chirality locked beam splitting effect, we have fabricated a rectangular SC sample of size 164x125 cm2 (made of 684 rods, as used in the simulation). The size of the reflective concave mirror is reduced to generate a narrower Gaussian beam (of width ~10.0 cm). By inserting the detecting probe inside the SC, we have experimentally measured the pressure distributions (at 3.66 kHz) along several equidistant horizontal lines. As observed in Fig. 3(c), the amplitude distribution shows a clear separation of the wave-packet as the sound travels forward, in which the traces of the sub-wave-packets agree well with the theoretical prediction. The main contribution of Bloch states for each beam has been checked further by performing Fourier transform (along the x direction) for the corresponding sub-wave-packet: as specified in Fig. 3(c), the and valleys are responsible for the right- and left-moving beams, respectively. To identify the vortex chirality carried by these two beams, we have scanned the phase profiles for two circular loops that locate separately on the two beams. The experimental data in Fig. 3(d), consistent with the COMSOL simulations, display obvious anticlockwise ( ) and clockwise ( ) phase growths for the left- and right-moving beams, respectively. In some sense, this peculiar beam splitting behavior resembles the valley Hall effect of electronic systems: the deflection of each sound beam is linked with particular vortex chirality. Interestingly, both split beams travel in 8 / 12 KKKKgvKK1n1na 'wrong' way when they are refracted out from the sample: the left-moving beam deflects rightward and the right-moving beam deflects leftward. This anomalous propagation of sound, so-called negative refraction, has sparked intense interest in past decades [1-3,13-15], which can be explained straightforwardly from the shape of EFC. FIG. 4. OAM-reversed valley transports in the samples constructed by -inverted SCs (here ), simulated for (a) the selective excitation of the state and (b) the beam splitting behavior. It is worth mentioning that, the operating frequency can be continuously tailored by mechanically rotating the triangular scatterers. Particularly, if the rotation angle is inverted, the vortex chirality can even be switched without any change of the incident wave, considering a mirror operation of the system. Several fancy sound transports can be imagined further if such -inverted SCs are combined together. For example, Fig. 4(a) shows an alternate excitation of the valley vortices with the opposite OAMs in the neighboring SC blocks with . The whole sample is well ignited because the momentum conservation is fulfilled perfectly at each SC interface. Similar idea can be extended to the valley-chirality locked beam splitting. As demonstrated in Fig. 4(b), the OAMs of the sound vortices are reversed separately once the beams traverse the SC interface (where the reflection can be remarkably reduced by adding a few SC layers with gradient distribution). Note that so far we have been focusing on the AV states of the lowest two bands. The AV states located on the higher bands, which accommodate more fine features, also deserve special 9 / 12 3030(b)0max(a)303030303030o302Ko30attention. For example, the valley state on the third band is characterized by a pair of oppositely-rotated vortices in a single unit cell, centered at the p and q points respectively. Numerical and experimental data are provided in the Supplemental Information. In conclusion, the universal valley physics has been successfully validated in the acoustic system (associated with reconfigurable operation frequencies and vortex chirality by mechanically rotating the triangular rods). This may stimulate extensive interest in the exploration of the valley-dependent phenomena in various artificial crystals and lead to novel manipulations on the corresponding classical waves. In addition to the implications in fundamental researches, such compact sound vortex array with controllable chiriality, which is unattainable through the currently existing approaches [62-69], could be potentially useful for patterning and rotating objects without contact, given the interaction of sound with matter [44,46-52,64,67,69]. Acknowledgements The authors thank C. T. Chan for fruitful discussions. This work is supported by the National Basic Research Program of China (Grant No. 2015CB755500); National Natural Science Foundation of China (Grant Nos. 11674250, 11374233, 11534013, and 11547310); Postdoctoral innovation talent support program (BX201600054). F. Zhang was supported by the UT-Dallas research enhancement funds. References: [1] S. Yang, J. H. Page, Z. Liu, M. L. Cowan, C. T. Chan, and P. Sheng, Phys. Rev. Lett. 93, 024301 (2004). [2] M. Lu, C. Zhang, L. Feng, J. Zhao, Y. Chen, Y. Mao, J. Zi, Y. Zhu, S. Zhu, and N. Ming, Nat. Mater. 6, 744–748 (2007). [3] A. Sukhovich, B. Merheb, K. Muralidharan, J. O. Vasseur, Y. Pennec, P. A. Deymier, and J. H. Page, Phys Rev Lett. 102, 154301 (2009). [4] X. Zhang and Z. Liu, Phys. Rev. Lett. 101, 264303 (2008). [5] Z. Liu, X. Zhang, Y. Mao, Y. Y. Zhu, Z. Yang, C. T. Chan, and P. Sheng, Science 289, 1734 (2000). [6] N. Fang, D. Xi, J. Xu, M. Ambati, W. Srituravanich, C. Sun, and X. Zhang, Nat. Mater. 5, 452 (2006). [7] M. Yang, G. Ma, Z. Yang, and P. Sheng, Phys. Rev. Lett. 110, 134301 (2013). 10 / 12 [8] S. H. Lee, C. M. Park, Y. M. Seo, Z. G. Wang, and C. K. Kim, Phys. Rev. Lett. 104, 054301 (2010). [9] Y. Li, B. Liang, Z. Gu, X. Zou, and J. Cheng, Sci. Rep. 3, 2546 (2013). [10] J. Zhao, B. Li, Z. Chen, and C. W. Qiu, Sci. Rep. 3, 2537 (2013). [11] K. Tang, C. Qiu, M. Ke, J. Lu, Y. Ye, and Z. Liu, Sci. Rep. 4, 6517 (2014). [12] Y. Xie, W. Wang, H. Chen, A. Konneker, B.-I. Popa, and S. A. Cummer, Nat. Commun. 5, 5553 (2014). [13] Z. Liang, and J. Li, Phys. Rev. Lett. 108, 114301 (2012). [14] J. Christensen, and F. J. Garcia de Abajo, Phys. Rev. Lett. 108, 124301 (2012). [15] V. M. Garcia-Chocano, J. Christensen, and J. Sanchez-Dehesa, Phys. Rev. Lett. 112, 144301 (2014). [16] J. Li, L. Fok, X. Yin, G. Bartal, and X. Zhang, Nat. Mater. 8, 931–934 (2009). [17] J. Zhu, J. Christensen, J. Jung, L. Martin-Moreno, X. Yin, L. Fok, X. Zhang, and F. J. Garcia- Vidal, Nat. Phys. 7, 52–55 (2011). [18] S. Zhang, C. Xia, and N. Fang, Phys. Rev. Lett. 106, 024301 (2011). [19] B.-I. Popa, L. Zigoneanu, and S. A. Cummer, Phys. Rev. Lett. 106, 253901 (2011). [20] L. Zigoneanu, B.-I. Popa, and S. A. Cummer, Nat. Mater. 13, 352 (2014). [21] Z. Yang, F. Gao, X. Shi, X. Lin, Z. Gao, Y. Chong, and B. Zhang, Phys. Rev. Lett. 114, 114301 (2015). [22] P. Wang, L. Lu, and K. Bertoldi, Phys. Rev. Lett. 115, 104302 (2015). [23] M. Xiao, G. Ma, Z. Yang, Z. Q. Zhang, and C. T. Chan, Nat. Phys. 11, 240 (2015). [24] V. Peano, C. Brendel, M. Schmidt, and F. Marquardt, Phys. Rev. X 5, 031011 (2015). [25] C. He, X. Ni, H. Ge, X. Sun, Y. Chen, M. Lu, X. Liu, and Y. Chen, Nat. Phys. 12, 1124 (2016). [26] Y. Peng, C. Qin, D. Zhao, Y. Shen, X. Xu, M. Bao, H. Jia, and X. Zhu, Nat. Commun. 7, 13368 (2016). [27] J. Lu, C. Qiu, L. Ye, X. Fan, M. Ke, F. Zhang, and Z. Liu, Nat. Phys. 2016, DOI: 10.1038/NPHYS3999. [28] C. Brendel, V. Peano, O. Painter, and F. Marquardt, arXiv:1607.04321 (2016). [29] A. Rycerz, J. Tworzydlo, and C. W. J. Beenakker, Nat. Phys. 3, 172 (2007). [30] D. Xiao, W. Yao, and Q. Niu, Phys. Rev. Lett. 99, 236809, (2007). [31] R. V. Gorbachev, J. C. W. Song, G. L. Yu, A. V. Kretinin, F. Withers, Y. Cao, A. Mishchenko, I. V. Grigorieva, K. S. Novoselov, L. S. Levitov, and A. K. Geim, Science 346, 448 (2014). [32] D. Xiao, G. Liu, W. Feng, X. Xu, and W. Yao, Phys. Rev. Lett. 108, 196802 (2012). [33] K. F. Mak, K. L. McGill, J. Park, and P. L. McEuen, Science 344, 1489 (2014). [34] X. Xu, W. Yao, D. Xiao, and T. F. Heinz, Nat. Phys. 10, 343-350 (2014). [35] H. Pan, X. Li, F. Zhang, and S. A. Yang, Phys. Rev. B 92, 041404(R) (2015). [36] G. W. Semenoff, V. Semenoff, and F. Zhou, Phys. Rev. Lett. 101, 087204 (2008). [37] I. Martin, Y. M. Blanter, and A. F. Morpurgo, Phys. Rev. Lett. 100, 036804 (2008). [38] F. Zhang, J. Jung, G. A. Fiete, Q. Niu, and A. H. MacDonald, Phys. Rev. Lett. 106, 156801 (2011). [39] Z. Qiao, J. Jung, Q. Niu, and A. H. Macdonald, Nano Lett. 11, 3453-3459 (2011). [40] F. Zhang, A. H. MacDonald, and E. J. Mele, Proc. Natl Acad. Sci. USA 110, 10546-10551 (2013). 11 / 12 [41] A. Vaezi, Y. Liang, D. H. Ngai, L. Yang, and E.-A. Kim, Phys. Rev. X 3, 021018 (2013). [42] L. Ju, Z. Shi, N. Nair, Y. Lv, C. Jin, J. V. Jr, C. Ojeda-Aristizabal, H. A. Bechtel, M. C. Martin, A. Zettl, J. Analytis, and F. Wang, Nature 520, 650-655 (2015). [43] J. Li, K. Wang, K. McFaul, Z. Zern, Y. Ren, K. Watanabe, T. Taniguchi, Z. Qiao, and J. Zhu, Nat. Nanotech. 11, 1060 (2016). [44] J. Lu, C. Qiu, M. Ke, and Z. Liu, Phys. Rev. Lett. 116, 093901 (2016). [45] J. F. Nye and M. V. Berry, Proc. R. Soc. A 336, 165 (1974). [46] K. Volke-Sepulveda, A. O. Santillan, and R. R. Boullosa, Phys. Rev. Lett. 100, 024302 (2008). [47] K. D. Skeldon, C. Wilson, M. Edgar, and M. J. Padgett, New J. Phys. 10, 013018 (2008). [48] A. O. Santillan, and K. A. Volke-Sepulveda, Am. J. Phys. 77, 209 (2009). [49] C. E. M. Demore, Z. Yang, A. Volovick, S. Cochran, M. P. MacDonald, and G. C. Spalding, Phys. Rev. Lett. 108, 194301 (2012). [50] A. Anhauser, R. Wunenburger, and E. Brasselet, Phys. Rev. Lett. 109, 034301 (2012). [51] Z. Y. Hong, J. Zhang, and B. W. Drinkwater, Phys. Rev. Lett. 114, 214301 (2015). [52] L. K. Zhang and P. L. Marston, Phys. Rev. E 84, 065601 (2011). [53] J. Lu, C. Qiu, S. Xu, Y. Ye, M. Ke, and Z. Liu, Phys. Rev. B 89, 134302 (2014). [54] K. Takashina, Y. Ono, A. Fujiwara, Y. Takahashi, and Y. Hirayama, Phys. Rev. Lett. 96, 236801 (2006). [55] O. Gunawan, Y. P. Shkolnikov, K. Vakili, T. Gokmen, E. P. De Poortere, and M. Shayegan, Phys. Rev. Lett. 97, 186404 (2006). [56] Z. Zhu, A. Collaudin, B. Fauque, W. Kang, and K. Behnia, Nat. Phys. 8, 89-94 (2011). [57] D. MacNeill, C. Heikes, K. F. Mak, Z. Anderson, A. Kormányos, V. Zólyomi, J. Park, and D. C. Ralph, Phys. Rev. Lett. 114, 037401 (2015). [58] K. F. Mak, K. He, J. Shan, and T. F. Heinz, Nat. Nanotechnol. 7, 494 (2012). [59] H. Zeng, J. Dai, W. Yao, D. Xiao, and X. Cui, Nat. Nanotechnol. 7, 490 (2012). [60] T. Cao, G. Wang, W. Han, H. Ye, C. Zhu, J. Shi, Q. Niu, P. Tan, E. Wang, B. Liu, and J. Feng, Nat. Commun. 3, 887 (2012). [61] J. Lu, C. Qiu, S. Xu, Y. Ye, M. Ke, and Z. Liu, Phys. Rev. B 89, 134302 (2014). [62] B. T. Hefner and P. L. Marston, J. Acoust. Soc. Am. 106, 3313 (1999). [63] R. Marchiano, and J-L. Thomas, Phys. Rev. E 71, 066616 (2005). [64] D. Baresch, J.-L. Thomas, and R. Marchiano, Phys. Rev. Lett. 116, 024301 (2016). [65] N. Jimenez, V. J. Sanchez-Morcillo, R. Pico, L. M. Garcia-Raffi, V. Romero-Garcia, K. Staliunas, Physics Procedia. 70, 245-248 (2015). [66] X. Jiang, J. J. Zhao, S. L. Liu, B. Liang, X. Zou, J. Yang, C. Qiu, and J. C. Cheng, Appl. Phys. Lett. 108, 203501 (2016). [67] T. Wang, M. Ke, W. Li, Q. Yang, C. Qiu, and Z. Liu, Appl. Phys. Lett. 109, 123506 (2016). [68] X. Jiang, Y. Li, B. Liang, J. Cheng, and L. Zhang, Phys. Rev. Lett. 117, 034301 (2016). [69] L. Ye, C. Qiu, J. Lu, K. Tang, H. Jia, M. Ke, S. Peng, and Z. Liu, AIP Advances 6, 085007 (2016). [70] J. L. Garcia-Pomar, A. Cortija, and M. Nieto-Vesperinas, Phys. Rev. Lett. 100, 236801 (2008). 12 / 12 

1010.5169

1

1010

Plasmon mass and Drude weight in strongly spin-orbit-coupled 2D electron gases

Spin-orbit-coupled two-dimensional electron gases (2DEGs) are a textbook example of helical Fermi liquids, i.e. quantum liquids in which spin (or pseudospin) and momentum degrees-of-freedom at the Fermi surface have a well-defined correlation. Here we study the long-wavelength plasmon dispersion and the Drude weight of archetypical spin-orbit-coupled 2DEGs. We first show that these measurable quantities are sensitive to electron-electron interactions due to broken Galileian invariance and then discuss in detail why the popular random phase approximation is not capable of describing the collective dynamics of these systems even at very long wavelengths. This work is focussed on presenting approximate microscopic calculations of these quantities based on the minimal theoretical scheme that captures the basic physics correctly, i.e. the time-dependent Hartree-Fock approximation. We find that interactions enhance the "plasmon mass" and suppress the Drude weight. Our findings can be tested by inelastic light scattering, electron energy loss, and far-infrared optical-absorption measurements.

cond-mat.mes-hall

cond-mat

Plasmon mass and Drude weight in strongly spin-orbit-coupled 2D electron gases Amit Agarwal,1, ∗ Stefano Chesi,2 T. Jungwirth,3, 4 Jairo Sinova,5, 3, 6 G. Vignale,7, 6 and Marco Polini1, 6, † 1NEST, Istituto Nanoscienze-CNR and Scuola Normale Superiore, I-56126 Pisa, Italy 2Department of Physics, University of Basel, Klingelbergstrasse 82, 4056 Basel, Switzerland 3Institute of Physics ASCR, v.v.i., Cukrovarnick 10, 162 53 Praha 6, Czech Republic 4School of Physics and Astronomy, University of Nottingham, Nottingham NG7 2RD, United Kingdom 5Department of Physics, Texas A&M University, College Station, Texas 77843-4242, USA 6Kavli Institute for Theoretical Physics China, CAS, Beijing 100190, China 7Department of Physics and Astronomy, University of Missouri, Columbia, Missouri 65211, USA Spin-orbit-coupled two-dimensional electron gases (2DEGs) are a textbook example of heli- cal Fermi liquids, i.e. quantum liquids in which spin (or pseudospin) and momentum degrees-of- freedom at the Fermi surface have a well-defined correlation. Here we study the long-wavelength plas- mon dispersion and the Drude weight of archetypical spin-orbit-coupled 2DEGs. We first show that these measurable quantities are sensitive to electron-electron interactions due to broken Galileian invariance and then discuss in detail why the popular random phase approximation is not capa- ble of describing the collective dynamics of these systems even at very long wavelengths. This work is focussed on presenting approximate microscopic calculations of these quantities based on the minimal theoretical scheme that captures the basic physics correctly, i.e. the time-dependent Hartree-Fock approximation. We find that interactions enhance the "plasmon mass" and suppress the Drude weight. Our findings can be tested by inelastic light scattering, electron energy loss, and far-infrared optical-absorption measurements. PACS numbers: 71.45.Gm, 71.10.-w, 71.70.Ej I. INTRODUCTION In recent years we have witnessed a tremendous explo- sion of interest in a large variety of novel two-dimensional (2D) quantum many-body systems. Prime examples are: (i) strongly spin-orbit-coupled 2D electron and hole gases, which are promising candidates for semiconductor spintronics1; (ii) graphene2 (a monolayer of carbon atoms arranged in a 2D honeycomb lattice), which has attracted a great deal of interest because of the massless-Dirac- fermion character of its carriers and because it may pave the way for carbon-based electronics3; (iii) 2D electron gases in HgTe/Hg(Cd)Te quantum wells where massless Dirac fermions are predicted to arise at a critical quan- tum well thickness4 -- 7; and, more recently, (iv) metallic surface states of 3D topological insulators8 -- 11. These systems share a unique common factor: their orbital degrees-of-freedom are intimately coupled to the electron spin (or sublattice pseudospin, in the case of graphene) degree-of-freedom. This coupling, being of rel- ativistic origin12, naturally breaks Galileian invariance and is thus the basic reason for a quite sensitive depen- dence of several observables to electron-electron inter- actions, even at very long wavelengths (see for example Refs. 13 -- 18). Furthermore, these systems exhibit coupled spin-charge collective dynamics, which is just beginning to be investigated in the contemporary literature19. In this article we focus our attention on an archetypical 2D electron gas model Hamiltonian with spin-orbit cou- pling (SOC). For the sake of simplicity we choose an ele- mentary form of SOC which is linear in momentum and has the canonical Rashba or Dresselhaus functional form. Since collective dynamics in quantum many-body sys- tems is controlled by isolated poles in dynamical linear- response functions20, we carry out a microscopic study of the density-density response function in the dynami- cal limit taking into account exactly SOC and treating electron-electron interactions beyond the random phase approximation (RPA). The RPA, which is commonly used to describe electron liquids, is indeed not capable to capture the subtle renormalization of the plasmon mode that occurs in non-Galileian-invariant quantum liquids. The study of many-body effects when both Rashba and Dresselhaus SOC terms are present in the Hamiltonian is beyond the scope of the present article: rotational in- variance of the Fermi contours is indeed spoiled by the si- multaneous presence of both effects and this complicates (and partly obscures) the basic interplay between SOC and many-body effects we want to highlight. Electron- electron interactions in 2D electron and hole gases in the presence of SOC have attracted a certain deal of atten- tion13,14,21 -- 40. Below we will make contact with the pre- existing literature whenever possible. Our manuscript is organized as follows. In Sect. II we present the model we have studied, we introduce the basic definitions, and outline the equation-of-motion approach we have used to relate the density-density response func- tion with the longitudinal current-current response func- tion. The latter is then evaluated microscopically within the time-dependent Hartree-Fock approximation in the long-wavelength limit in Sect. III. While the main focus of this paper is on the plasmon dispersion at long wave- lengths and on the Drude weight, in the same Section we briefly discuss interaction corrections to the spin Hall conductivity and the renormalization of the spin-orbit splitting of the bands due to electron-electron interac- 0 1 0 2 t c O 5 2 ] l l a h - s e m . t a m - d n o c [ 1 v 9 6 1 5 . 0 1 0 1 : v i X r a 2 where S is the area of the system and γk = γk(θk) is given by tan γk = α cos (θk) + β sin (θk) β cos (θk) + α sin (θk) . (9) The map k → neq(k) = (cos (γk),− sin (γk)) (10) between momentum and the unit vector neq, which parametrizes the noninteracting orientation of the spin texture in momentum space, establishes the helical na- ture of the model. Electron-electron interactions in Eq. (1) are described by the usual two-body spin-independent Hamiltonian Hint = 1 2S(cid:88)q(cid:54)=0(cid:88)k,k(cid:48)(cid:88)i,j vq ψ†k−q,i ψ†k(cid:48)+q,j ψk(cid:48),j ψk,i , (11) where vq = 2πe2/(
q) is the 2D Fourier transform of the Coulomb interaction (
being a high-frequency dielectric constant which depends on the specific semiconductor heterojunction in which the 2DEG is created). This spe- cific form of interaction potential applies to a strictly 2D system. The finite width of the quantum well hosting the 2DEG can be easily taken into account by introduc- ing a form factor F (q), which renormalizes the Fourier transform vq → Vq = vqF (q) (see, for example, Ref. 43). As is common in electron-gas theory20, the electron density n will be expressed below in terms of the more convenient dimensionless Wigner-Seitz parameter rs: rs = , (12) 1 (cid:112)πna2 B where aB =
¯h2/(mbe2) is the material Bohr radius. (5) B. Equations of motion and plasmons tions. In Sect. IV we present our main numerical results, while in Sect. V we summarize our findings and draw our main conclusions. II. GENERAL THEORY A. Model Hamiltonian We consider the following model Hamiltonian for a 2D electron gas (2DEG), H = H0 + HSOC + Hint , (1) incorporating the usual parabolic-band kinetic-energy term, SOC, and electron-electron interactions. More pre- cisely, the first term in Eq. (1) is given by H0 =(cid:88)k,i ε(k) ψ†k,i ψk,i , (2) with i =↑,↓ a real-spin label and ε(k) = ¯h2k2/(2mb), mb being the bare electron band mass. For the SOC term we choose a simple linear-in-momentum Rashba-Dresselhaus model41: HSOC = (cid:88)k,i,j ψ†k,i[α(σx ijkx) ijky − σy ijky)] ψk,j . + β(σx ijkx − σy (3) ij and σy Here σx ij are Pauli matrices, while α and β are the Rashba and Dresselhaus SOC constants, respectively. Di- agonalization of the sum of the first two terms in Eq. (1) ( H0 + HSOC) yields two bands (see, for example, Ref. 42), (4) with λ = ±1 the so-called "chirality" index, θk the angle between k and the x axis, and ελ(k) = ε(k) + λkΓ(θk) , Γ(θ) =(cid:112)α2 + β2 + 4αβ sin(θ) cos(θ) . The Fermi wave vectors for the two bands can be ex- pressed in terms of θk and of the Fermi energy εF: k(0) F,λ = −λ mbΓ(θk) ¯h2 +(cid:115)(cid:20) mbΓ(θk) ¯h2 (cid:21)2 + 2mbεF ¯h2 . (6) Note that for zero Fermi energy the Fermi contour of the minority λ = + band shrinks into a single point (i.e. k(0) F,+ = 0). For any εF ≥ 0 the electron density n can be expressed in terms of the Fermi energy as n = mbεF π¯h2 +(cid:18) mb ¯h2(cid:19)2 α2 + β2 π . (7) The eigenstates of H0 + HSOC corresponding to the eigenvalues (4) are given by the product of a plane wave and a spinor, Ψk,λ(r) = eik·r √S × 1 √2(cid:18) 1 λe−iγk (cid:19) , (8) Collective modes are isolated poles in appropriate dy- namical susceptibilities. Plasmons, in particular, are iso- lated poles in the dynamical density-density response function χρρ(q, ω). Note that we are deliberately not denoting the density-density response function by the symbol χρρ(q, ω), i.e. we are assuming that the sys- tem we are interested in is rotationally invariant and thus its density-density response function depends only on q = q. This happens when β or α is equal to zero. The case α = ±β deserves special attention and will be discussed at a greater length below (see Sect. III F). In full generality, this response function can be written as χρρ(q, ω) = , (13) response function20,44, which physically describes the where (cid:101)χρρ(q, ω) is the so-called "proper" density-density (cid:101)χρρ(q, ω) 1 − vq(cid:101)χρρ(q, ω) density response to the screened potential. The plasmon mode can be found by solving the equation, and it obeys the standard Heisenberg equation of motion (¯h = 1 from now on) 3 (14) 1 − vq(cid:101)χρρ(q, ω) = 0 . In this article we are not interested in the dispersion of the plasmon at finite q but only in its limit for q → 0. In this limit we can neglect44 the distinction between the proper and the full causal response function χρρ(q, ω). In Sect. II C we prove that lim q→0(cid:60)e χρρ(q, ω) = D πe2 lim ω→0 q2 ω2 , (15) where the quantity D depends on density and on SOC. Note the order of limits in Eq. (15) and everywhere below: the limit ω → 0 is taken in the "dynamical" sense20, i.e. ω (cid:29) vF,λq, where vF,λ is the Fermi velocity for each chiral band. Before proceeding with the formal proof of Eq. (15) we highlight its main physical consequences. Using Eq. (15) in Eq. (14) and solving for ω we find that, to leading order in q, ωpl(q → 0) =(cid:114) 2D
q1/2 ≡(cid:115) 2πne2 mpl
q1/2 , (16) where we have introduced the plasmon mass mpl = . (17) πne2 D The plasmon mass is thus completely controlled by the quantity D. In the same limit the imaginary-part of the low-frequency a.c. conductivity σ(ω) = ie2ωχρρ(ω)/q2 has the form (cid:61)m σ(ω) → D πω . (18) The a.c. conductivity is a causal response function, which implies that its poles can only lie infinitesimally below the real-frequency axis, i.e. σ(ω → 0) ∝ (ω + iη)−1. It then follows that the real-part of the conductivity has a δ-function Drude peak at ω = 0: (cid:60)e σ(ω) = Dδ(ω) . (19) The quantity D introduced in Eq. (15) is thus precisely the Drude weight. In the presence of disorder the δ- function peak in Eq. (19) is broadened into a Drude peak, but the Drude weight is preserved. i∂t ρq = [ρq, H] ≡ q · j(p) q , (21) which is simply the quantum mechanical version of the continuity equation. Here the so-called paramagnetic current-density operator20 has the following transparent form: ψ†k−q,i kx + qx/2 mb ψk,i + (β σx q − ασy q) (22) along the x direction and ψ†k−q,i ky + qy/2 mb ψk,i + (ασx q − β σy q) , (23) j(p) q,x =(cid:88)k,i q,y =(cid:88)k,i j(p) along the y direction. In Eqs. (22)-(23) we have intro- duced the spin-density operators ψ†k−q,iσµ ψk,j . (24) σµ ij q =(cid:88)k,i,j We now introduce the causal linear-response functions χAB(ω), which are defined by the Kubo "product"20, χAB(ω) = 1 S (cid:104)(cid:104) A; B(cid:105)(cid:105)ω S(cid:90) ∞ i 0 ≡ − dt(cid:104)[ A(t), B(0)](cid:105)eiωte−ηt , (25) where the symbol (cid:104) O(cid:105) denotes the expectation value of the operator O over the exact interacting ground state and η → 0+ is a positive infinitesimal. The dynamical response function (cid:104)(cid:104) A; B(cid:105)(cid:105)ω obeys the following identity (26) 1 ω(cid:104)[ A, B](cid:105) + i ω(cid:104)(cid:104)∂t A; B(cid:105)(cid:105)ω , (cid:104)(cid:104) A; B(cid:105)(cid:105)ω = or, (cid:104)(cid:104) A; B(cid:105)(cid:105)ω = 1 ω(cid:104)[ A, B](cid:105) − i ω(cid:104)(cid:104) A; ∂t B(cid:105)(cid:105)ω . (27) Using the continuity equation (21) and Eqs. (26)- (27), the density-density response function χρρ(q, ω) can be expressed in terms of the longitudinal paramagnetic current-current response function as, χρρ(q, ω) ≡ = 1 S (cid:104)(cid:104)ρq; ρ−q(cid:105)(cid:105)ω 1 1 ω(cid:104)(cid:104)q · j(p) S q · (cid:104)[j(p) q , ρ−q](cid:105) ω2 1 S q ; ρ−q(cid:105)(cid:105)ω + 1 S (cid:104)(cid:104)q · j(p) q ; q · j(p) −q(cid:105)(cid:105)ω ω2 . C. Rigorous definition of the Drude weight = We now proceed to demonstrate Eq. (15) using the "equations-of-motion" approach. The density operator corresponding to the Hamiltonian (1) is given by the usual expression ρq =(cid:88)k,i ψ†k−q,i ψk,i , We remind the reader that in the presence of a vector potential Ak the physical current-density operator jq is related to the paramagnetic one by (28) (20) jq = j(p) q + e mbcS(cid:88)k Aq−k ρk . (29) The paramagnetic current-current response function χj(p) (cid:96) j(p) m (q, ω) = 1 S (cid:104)(cid:104)j(p) q,(cid:96) ; j(p) −q,m(cid:105)(cid:105)ω (30) (here (cid:96), m label the coordinate indices) is thus related to the physical one by the simple equation20 χj(cid:96)jm(q, ω) = n mb δ(cid:96)m + χj(p) (cid:96) j(p) m (q, ω) . (31) For generic values (α, β) of the SOC constants, the dynamical response functions of the model described by Eq. (1) are anisotropic, i.e. they depend on the direction of q. However, in the cases of pure Rashba (β = 0) or pure Dresselhaus (α = 0) SOC the ground state of the Hamiltonian H0 + HSOC is rotationally invariant: for the sake of simplicity, in what follows we will restrict our at- tention to these two "extreme" cases. From now on we assume α (cid:54)= 0 and β = 0. In Sect. III F we will com- ment on how our results change in the pure Dresselhaus (α = 0 and β (cid:54)= 0) case and in the special case α = ±β. Last but not least, we also assume to be in the regime in which both chiral bands are occupied (εF > 0). In this situation the Fermi surface consists of two concen- tric circles. At low enough densities the topology of the Fermi surface changes dramatically, the occupied states becoming an annulus in momentum space. We will not tackle interaction effects in this interesting but hard to achieve experimentally regime. In a homogeneous and isotropic liquid we can de- compose the tensor χj(cid:96)jm (q, ω) into its longitudinal and transverse components with respect to the direction of q: χj(cid:96)jm (q, ω) = χL(q, ω) q(cid:96)qm q2 + χT(q, ω)(cid:18)δ(cid:96)m − q(cid:96)qm q2 (cid:19) . (32) Using this definition we immediately end up with the following result χρρ(q, ω) = 1 S q · (cid:104)[j(p) q , ρ−q](cid:105) ω2 + q2 ω2(cid:20)χL(q, ω) − n mb(cid:21) . (33) We stress that Eq. (33) is exact (provided that the ground state is homogenous and isotropic). The commutator on the r.h.s. of Eq. (33) can be cal- culated easily: indeed, the portion of the paramagnetic current operator due to spin-orbit coupling [second terms on the r.h.s. of Eqs. (22)-(23)] is proportional to the spin operator σq only, which commutes with the density oper- ator ρ−q. Thus the commutator is found to be equivalent to that of the 2DEG without any spin orbit coupling. It is related to the so-called f-sum rule20 and is given by 1 S [j(p) q , ρ−q] = q n mb . (34) Using Eq. (34) into Eq. (33) we are left with the following crucially important relation: q2 ω2 χL(q, ω) . χρρ(q, ω) = (35) Note that the f-sum rule is crucial for the cancellation of the diamagnetic n/mb term in the square brackets on the r.h.s of Eq. (33). 4 Eq. (35) is identical in form with Eq. (15) provided that we identify D with the following dynamical limit: (36) D ≡ πe2 lim ω→0 lim q→0(cid:60)e χL(q, ω) . This equation is extremely important because it gives us an operational definition of the Drude weight. In order to calculate it we need to compute the dynamical limit of the real part of the longitudinal current-current response function χL(q, ω). Such a microscopic calculation will be carried out below in Sect. III within the so-called time- dependent Hartree-Fock approximation. D. Broken Galileian invariance In a standard 2DEG without spin-orbit coupling (α = β = 0) the longitudinal current-current response function obeys the exact relation lim ω→0 lim q→0 χL(q, ω) = n mb , (37) a nonperturbative result (i.e. valid for any strength of electron-electron interactions as long as the 2DEG re- mains in a translationally-invariant and homogeneous ground state), which is completely independent of com- plicated exchange and correlation effects. In this case the Drude weight becomes D = πne2/mb and the plasmon mass reduces to the bare electron mass, mpl = mb. The physical reason behind the exact result (37) is the following. In the limit q → 0 χL(q, ω) measures the re- sponse of the system to a homogeneous time-dependent vector potential A(t), i.e. to a homogeneous electric field E(t) = −c−1dA(t)/dt. In a system with a single parabolic band the usual replacement p → p + eA(t)/c implies that a uniform vector potential couples identi- cally to all the electrons and thus only to the center-of- mass motion. This is immediately seen in first quantiza- tion: H0(A) = (cid:88)i = H0 + 1 2mb(cid:104)pi + e c A(t)(cid:105)2 e mbc PCM · A(t) + O(A2) , (38) where PCM =(cid:80)i pi is the centre-of-mass momentum. In the last equality terms of order A2 have been neglected since we are interested in the linear-response regime. Electron-electron interactions are thus completely trans- parent to A(t), since the latter does not probe the relative motion of electrons. Eq. (37) can be derived by a classical Newton's equa- tion for the centre-of-mass coordinate RCM: mbN d2RCM dt2 = −eN E(t) = e c N dA(t) dt , (39) where N is the total number of electrons. this equation we find VCM(t) = [e/(mbc)] A(t) or Integrating j(p) q=0(t) = nVCM(t) = n mb e c A(t) , (40) i.e. Eq. (37). To see more formally why Eq. (37) comes about, we can use the exact-eigenstate (Lehmann) representation20 for the current-current response function: χj(cid:96)jm (q, ω) = n mb δ(cid:96)m + −q,m0(cid:105) 1 S(cid:88)n (cid:32)(cid:104)0j(p) q,(cid:96)n(cid:105)(cid:104)nj(p) ω − ωn0 + iη (cid:33) , q,(cid:96)0(cid:105) (41) − (cid:104)0j(p) −q,mn(cid:105)(cid:104)nj(p) ω + ωn0 + iη where the limit η → 0+ is understood. In a translation- ally invariant system, with or without SOC, the exact eigenstates n(cid:105) are eigenstates of the total momentum. In the absence of SOC, moreover, j(p) (cid:96),q=0 coincides with the total momentum [see Eqs. (22)-(23)] and thus for α = β = 0 and q → 0 the second term in Eq. (41) van- In the presence of ishes and one is left with Eq. (37). SOC, however, j(p) (cid:96),q=0 does not coincide with the total momentum and thus Eq. (37) ceases to be true. When α (or β) is non-zero we have lim ω→0 lim q→0 χL(q, ω) (cid:54)= n mb . (42) Deviations from the trivial n/mb result are due to both single- and many-particle effects14. The single-particle contribution to the long-wavelength low-energy limit of χL(q, ω) can be found quite easily. In Sect. III C we will show that if electron-electron interac- tions are neglected lim ω→0 lim q→0 χ(0) L (q, ω) = n mb − α2 ν0 2 , (43) where ν0 = mb/π is the usual 2D parabolic-band density- of-states in the absence of SOC. The rest of the paper is mainly devoted to quantifying interaction-corrections to Eq. (43). E. Failure of the random phase approximation Before concluding this Section, we would like to em- phasize that the popular random phase approximation (RPA) is not capable of capturing the subtle renormal- izations of the Drude weight due to many-body effects. By definition, within RPA the proper density-density interacting value20: response function(cid:101)χρρ(q, ω) is approximated with its non- ρρ (q, ω) = L (q, ω) . (44) q2 ω2 χ(0) (cid:101)χρρ(q, ω) RPA→ χ(0) When Eq. (44) is substituted in Eq. (36) one finds im- mediately that the RPA Drude weight is identical to its noninteracting value: 5 DRPA = lim ω→0 ≡ D0 . lim q→0 χ(0) L (q, ω) = πe2(cid:20) n 2(cid:21) mb − α2 ν0 (45) More physically, the reason why RPA does not capture the subtle interaction renormalizations of D is the follow- ing. During a plasmon oscillation the Fermi circle oscil- lates back and forth in momentum space. Due to SOC this oscillatory motion of charge excites spin oscillations. Exchange interactions are of course very sensitive to the spin degrees-of-freedom. The RPA, however, is simply a time-dependent Hartree theory20, which treats exactly only the self-consistent electrical potential, VH(r, t) =(cid:90) d2r(cid:48) e2
r − r(cid:48) δn(r(cid:48), t) , (46) created by the electrons displaced away from the equi- librium position in the presence of the neutralizing back- ground, while completely neglecting the self-consistent exchange field associated with the spin degrees-of- freedom. From this argument it clearly emerges that the minimal theory which can capture interaction-corrections to Eq. (43) is the time-dependent Hartree-Fock theory. III. MICROSCOPIC TIME-DEPENDENT HARTREE-FOCK THEORY In this Section we present a microscopic theory of D that takes into account electron-electron interactions in an approximate manner. As discussed in Sect. II E, the minimal approximation that captures the renormaliza- tion of D due to many-body effects is the so-called time- dependent Hartree-Fock approximation (TDHFA). One of the pleasant properties of the TDHFA is that it is exact to first order in Coulomb interactions. Other ad- vantages, such as its relative simplicity, will be evident below. As we have amply discussed in the previous Sections, we want to study the response of the system described by the Hamiltonian (1) to a weak homogeneous external time-dependent electric field directed along, say, x. In the gauge in which the scalar potential is zero the elec- tric field is simply described by a time-dependent vector potential: E(t) = −[c−1dA(t)/dt] x. The vector po- tential enters the Hamiltonian (1) via the usual minimal coupling p → p + eA(t) x/c. The parabolic-band part becomes (cid:104)kx + H0(t) = (cid:88)k,i e c A(t)(cid:105)2 2mb + k2 y ψ†k,i ψk,i , (47) while the SOC part reads HSOC(t) = α(cid:88)k,i,j ψ†k,i(cid:110)σx ijky − σy ij(cid:104)kx + e c A(t)(cid:105)(cid:111) ψk,j . (48) Neglecting terms O(A2), which are beyond linear- response theory, we can write the sum of the two terms in Eqs. (47)-(48) as H0(t) + HSOC(t) = H0 + HSOC + σy totA(t) . − α e c e mbc P x CMA(t) (49) Thus, due to SOC, a magneto-electric effect appears45: a uniform electric field applied along the x direction acts as a uniform magnetic field in the y direction [last term in the r.h.s. of Eq. (49)]. Here σy tot = σy q=0. Electron-electron interactions are treated within the Hartree-Fock (HF) mean-field theory in which the two- body term in Eq. (11) is approximated as46 ψ†k−q,i ψ†k(cid:48)+q,j ψk(cid:48),j ψk,i ≈ − : ψ†k−q,i − : ψ†k(cid:48)+q,j ψk(cid:48),j : (cid:104) ψ†k(cid:48)+q,j ψk,i : (cid:104) ψ†k−q,i ψk,i(cid:105) ψk(cid:48),j(cid:105) , (50) where (cid:104). . .(cid:105) (: . . . :) denote the expectation value over (normal ordering with respect to) the HF ground state20. At this point we introduce the spin-density matrix, (cid:104) ψ†k,i ψk(cid:48),j(cid:105) = δk,k(cid:48)ρij(k) , (51) which just assumes that the mean-field ground state is translationally invariant. The interaction contribution to the total Hamiltonian reads and 6 Using Eq. (53) in Eq. (52), the total mean-field HF Hamiltonian can be written as HHF =(cid:88)k,i,j : ψ†k,i [δijB0(k) + σij · B(k)] ψk,j : (54) where the HF fields are defined by B0(k) = ε(k) + e mbc and P x CMA(t) −(cid:90) d2k(cid:48) (2π)2 vk−k(cid:48)f+(k(cid:48)) (55) B(k) = h(k) −(cid:90) d2k(cid:48) (2π)2 vk−k(cid:48)f−(k(cid:48)) n(k(cid:48)) . (56) In Eq. (56), h(k) = −α e c A(t) y + αk neq(k) (57) is an effective magnetic field, which has the external magneto-electric component45 of modulus Bext = α e c A (58) arising from the external vector potential and an internal component ∝ neq(k), while the last term is the exchange field due to the electron-electron interactions with f±(k) ≡ nk,+ ± nk,− 2 n(k) ≡ B(k) B(k) . (59) (60) Hint = − 1 S(cid:88)k,k(cid:48)(cid:88)i,j vk−k(cid:48)ρji(k(cid:48)) : ψ†k,i ψk,j : . (52) The noninteracting band-eigenstate occupation factors are given by We parametrize the spin-density matrix ρij(k) in a com- pact form24 in terms of the occupation factors, nk,±, of the noninteracting Hamiltonian H0 + HSOC in the eigen- state representation and in the absence of A(t): nk,+ + nk,− nk,+ − nk,− 2 2 δij + ρij(k) = n(k)· σji . (53) Here n(k) is a unit vector on the 2D plane which de- notes the orientation of the spins in the total "effective" magnetic field. The idea behind this parametrization is that a homogeneous external field (a field with q = 0) cannot change anything but the orientation of the spin, which is encoded in the unit vector n(k). Note that in the absence of the external field n(k) = neq(k). Eq. (53) is nevertheless approximate since it assumes the absence of interaction effects in the ground state of the system. More explicitly, the Fermi wave vectors are renormalized by electron-electron interactions32, k(0) F,± → kF,±. We will come back to this point below in Sect. III A. nk,± = Θ(εF − ε±(k)) , (61) where Θ(x) is the standard Heaviside step function. As we have already emphasized above, for pure Rasha SOC (β = 0) the momentum occupation factors nk,± are rotationally-invariant and depend only on k = k (the same is true also for pure Dresselhaus SOC, α = 0). It is thus extremely convenient to decompose the spherically symmetric inter-electron interaction vk−k(cid:48) in angular mo- mentum components, with vk−k(cid:48) = +∞(cid:88)m=−∞ Vm(k, k(cid:48)) =(cid:90) 2π 0 Vm(k, k(cid:48))eim(θk−θk(cid:48) ) , (62) dθ 2π e−imθ vqq=k−k(cid:48) , (63) θ being the angle between k and k(cid:48). A. Equilibrium HF theory In the absence of the external electric field, i.e. A(t) = 0, the unit vector n(k) coincides with the equilibrium one: n(k) → neq(k) = (sin (θk),− cos (θk)) . (64) Substituting Eq. (62) in Eq. (56) and performing the an- gular integration over θk(cid:48), we find that the equilibrium solution of Eq. (56) reads Beq(k) =(cid:20)αk −(cid:90) ∞ 0 dk(cid:48) 2π k(cid:48)f−(k(cid:48))V1(k, k(cid:48))(cid:21) neq(k) . (65) As expected, in the absence of the electric field, B(k) is oriented along neq(k) and it is isotropic. The modulus of Beq(k) is simply 7 Rigorously speaking, the HF bands and the corre- sponding Fermi wave vectors (kF,±) should be calculated in a fully self-consistent manner. We have done this and we find that the repopulation of the energy bands in the ground state due to interactions is a very small effect. In fact, the difference between k(0) F,± and kF,± is less than 0.5% over the entire range of parameters we have consid- ered. We have thus ignored this small effect throughout this article and used k(0) F,± in all calculations. B. The non-equilibrium problem: linearization of the HF equation We now proceed to solve Eq. (56) in the presence of the external electric field by linearizing it around the equi- librium solution Beq(k). To this end we write Beq(k) = αk + Σ1(k) , (66) and depends only on k with the self-energy Σ1(k) defined by and Σ1(k) = −(cid:90) ∞ 0 dk(cid:48) 2π k(cid:48) f−(k(cid:48))V1(k, k(cid:48)) . (67) For εF > 0 the factor f− in the integrand, being the difference in the occupation of the two bands, picks up contributions only from wave vectors k(cid:48) in the interval [k(0) F,λ is given by Eq. (6) with β = 0. The self energy can thus be written as F,−], where k(0) F,+, k(0) Σ1(k) = F,− 1 4π(cid:90) k(0) k(0) F,+ dk(cid:48) k(cid:48)V1(k, k(cid:48)) . (68) In a completely analogous manner, it is possible to find the equilibrium solution of Eq. (55) which reads B0,eq(k) = ε(k) −(cid:90) ∞ 0 dk(cid:48) 2π k(cid:48)f+(k(cid:48))V0(k, k(cid:48)) , (69) or, B0,eq(k) = ε(k) + Σ0(k) with Σ0(k) = − 1 − F,+ 0 1 2π(cid:90) k(0) 4π(cid:90) k(0) k(0) F,+ F,− dk(cid:48) k(cid:48)V0(k, k(cid:48)) dk(cid:48) k(cid:48)V0(k, k(cid:48)) . Finally, the complete HF bands are given by EHF,λ(k) = ε(k) + Σ0(k) + λ[αk + Σ1(k)] , (71) and the quasiparticle effective mass m(cid:63) can be defined as λ for the λ-th band k(0) F,λ λ ≡ m(cid:63) ∂EHF,λ(k) ∂k (cid:12)(cid:12)(cid:12)(cid:12)k=k(0) F,λ . (72) B(k) = Beq(k) + δB(k) n(k) = neq(k) + δB⊥(k) Beq(k) + O((δB)2) , (73) (74) where δB⊥(k) = δB(k) − neq(k)[ neq(k) · δB(k)] is the component of δB(k) perpendicular to neq(k). We now make the following Ansatz for δB(k): δB(k) = [δBL,1(k) cos (θk)] neq(k) − [δBT,1(k) sin(θk)] z × neq(k) . (75) We note that the Ansatz has to be consistent with the un- derlying model Hamiltonian. Indeed, for the pure Rashba model, neq(k) = k × z and z × neq(k) = k. Since the magneto-electric field is in the y-direction [see Eq. (57)] its components along z× neq and neq are proportional to sin(θk) and cos(θk), respectively. This justifies the par- ticular form of Eq. (75). Using Eq. (75) in Eq. (74) we find that δ n(k) ≡ n(k) − neq(k) = − δBT,1(k) sin(θk) Beq(k) k . (76) Substituting Eqs. (75) and (76) in Eqs. (56), integrat- ing over θk(cid:48), and keeping only terms that are linear in δB(k), we find (70) 1 2 [δBL,1(k) + δBT,1(k)] = Bext − (cid:90) ∞ 0 dk(cid:48) 4π k(cid:48)f−(k(cid:48))V0(k, k(cid:48)) δBT,1(k(cid:48)) Beq(k(cid:48)) , (77) and 1 2 [δBL,1(k) − δBT,1(k)] =(cid:90) ∞ 0 dk(cid:48) 4π k(cid:48)f−(k(cid:48))V2(k, k(cid:48)) . (78) × δBT,1(k(cid:48)) Beq(k(cid:48)) Summing and subtracting these two equations we fi- nally find the following integral equation for the trans- verse δBT,1(k) component: δBT,1(k) = Bext −(cid:90) ∞ where the kernel KT(k, k(cid:48)) is defined by 0 dk(cid:48) KT(k, k(cid:48))δBT,1(k(cid:48)) (79) KT(k, k(cid:48)) = = 1 4π 1 4π k(cid:48)f−(k(cid:48)) k(cid:48)f−(k(cid:48)) V0(k, k(cid:48)) + V2(k, k(cid:48)) Beq(k(cid:48)) V0(k, k(cid:48)) + V2(k, k(cid:48)) αk(cid:48) + Σ1(k(cid:48)) . (80) Once Eq. (79) has been solved self-consistently for δBT,1(k), the longitudinal component δBL,1(k) can be calculated from δBL,1(k) = Bext −(cid:90) ∞ 0 dk(cid:48) KL(k, k(cid:48)) δBT,1(k(cid:48)) (81) with . 1 4π V0(k, k(cid:48)) − V2(k, k(cid:48)) (82) k(cid:48)f−(k(cid:48)) αk(cid:48) + Σ1(k(cid:48)) KL(k, k(cid:48)) = For future reference, it is very convenient to rewrite Eq. (79) in a dimensionless form. To this end, we scale all the wave vectors with the 2DEG Fermi wave vec- tor kF = √2πn in the absence of SOC, all energies with εF,0 = k2 F/(2mb), the pseudopotentials Vm with 2πe2/(
kF), and, finally, we introduce the dimension- less SOC constant47 ¯α = mbα/kF and the dimensionless quantity u = δBT,1/Bext. From now on, symbols with a bar over them denote dimensionless quantities. In these units Eq. (79) reads u(x) = 1 + rs 2√2(cid:90) Λ− Λ+ dx(cid:48) x(cid:48) ¯V0(x, x(cid:48)) + ¯V2(x, x(cid:48)) 2¯αx(cid:48) + ¯Σ1(x(cid:48)) Here x = k/kF, x(cid:48) = k(cid:48)/kF, u(x(cid:48)) . (83) (84) Λ± = k(0) F,± kF = ∓¯α +(cid:112)1 − ¯α2 , and ¯Σ1(x) is the dimensionless version of the self-energy introduced in Eq. (68): 8 We need to evaluate the longitudinal response χL to a uniform vector potential A in the ω → 0 limit. We thus have to evaluate the change in longitudinal physical current due to a uniform electric field applied, say, along the x direction: δjx = lim A→0(cid:104)(cid:104)jq=0,x(cid:105)A − (cid:104)jq=0,x(cid:105)A=0(cid:105) , (87) where the physical current operator jq has been intro- duced in Eq. (29). Recalling the definition of the spin- density matrix ρij(k) we find that 1 S(cid:88)k,i δjx = n mb eA c + kx mb δρii(k) − α σy ijδρij(k) . (88) Since the diagonal components of the spin-density ma- trix do not change under the application of a uniform magnetic field we get the following important relation 1 S (cid:88)k,i,j δjx = n mb eA c − α δσy , (89) where δσy = 1 σy S (cid:88)k,i,j = −2(cid:90) d2k ij δρij(k) = 2(cid:90) d2k (2π)2 f−(k) δny (2π)2 f−(k) δBT,1(k) Beq(k) sin2 (θk) . (90) Performing the angular integration we finally find that δσy = χσyσy Bext , (91) where the magneto-electric field Bext has been introduced above in Eq. (58) and where the in-plane spin suscepti- bility is given by χσyσy = = F,− 1 k(0) F,+ Bext(cid:90) k(0) 2 (cid:90) Λ− ν0 Λ+ dx dk 4π k δBT,1(k) Beq(k) . xu(x) 2¯αx + ¯Σ1(x) (92) ¯Σ1(x) = Σ1 εF,0 = rs√2 F (x) (85) with F (x) =(cid:90) Λ− Λ+ dx(cid:48) x(cid:48) ¯V1(x, x(cid:48)) . (86) C. Interaction corrections to the Drude weight and renormalization of the in-plane spin susceptibility We are now in the position to evaluate the interaction corrections to the Drude weight from the definition in Eq. (36). In the noninteracting rs → 0 limit the vertex correction u tends to unity and the self-energy Σ1 to zero: in this limit Eq. (92) reproduces the well known result for the in-plane spin susceptibility of a 2DEG with Rashba SOC, i.e. χ(0) σyσy = ν0/2. Finally, using Eq. (36) we find that the Drude weight is given by D = πe2 δjx eA/c = πe2(cid:18) n mb − α2χσyσy(cid:19) . (93) This is the most important result of this work. It states that the corrections due to SOC and many-body effects to the universal πe2n/mb Drude weight of a standard parabolic-band 2DEG are completely controlled by the uniform in-plane spin susceptibility χσyσy in the dynam- ical limit. Note that, even though the particular deriva- tion we have given in this Section seems to be related to (and thus dependent on) the TDHFA, Eq. (93) is exact and stems directly from Eq. (22). In the noninteracting limit χσyσy → ν0/2 and thus, in the same limit, D0 = πe2(cid:18) n 2(cid:19) . mb − α2 ν0 (94) In Sect. IV we will present numerical results for the ratio D/D0 as evaluated from the HF expression for the in- plane spin susceptibility in Eq. (92). Normally electron- electron interactions enhance the spin susceptibility: we thus anticipate that the Drude weight of the interacting system is smaller than its value D0 in the absence of interactions. Before concluding this Section we derive a semi- analytical expression for χσyσy up to first order in the coupling constant e2. To this order of perturbation the- ory the solution of Eq. (83) can be found analytically with the result u(x) = 1 + rs 2√2¯α g(x) (95) where g(x) =(cid:90) Λ− Λ+ dx(cid:48) ¯V0(x, x(cid:48)) + ¯V2(x, x(cid:48)) 2 . (96) We notice that the perturbative solution (95) is not of the first order in rs, since ¯α = mbα/kF and Λ± also depend on density via the Fermi wave vector. In the presence of SOC, interaction effects are not solely controlled by rs. Substituting Eq. (95) in Eq. (92) and expanding the ratio in the integrand of this equation in powers of e2 up to first order we finally find that with χσyσy χ(0) σyσy = 1 + rsA 4√2¯α2 , A =(cid:90) Λ− Λ+ dx (cid:20)g(x) − F (x) x (cid:21) . (97) (98) A plot of A as a function of rs for different values of ¯α is reported in Fig. 1: we clearly see that A is positive and that thus the in-plane spin susceptibility is enhanced by electron-electron interactions (at least to first order in e2). In the high-density and/or weak-SOC limit (mbα (cid:28) kF or, equivalently, ¯α (cid:28) 1) we can approximate A in the following manner: 9 FIG. 1: (Color online) The quantity A in Eq. (98) as a func- tion of rs for two values of the Rashba SOC strength α. Notice that A > 0 and thus χσy σy /χ(0) σy σy > 1. In the same limit g(1) − F (1) → 2¯α(cid:20) ¯V0(1, 1) + ¯V2(1, 1) 2 = 4¯α 3π , − ¯V1(1, 1)(cid:21) (100) the last equality being valid only for unscreened Coulomb interactions [see Eq. (121) below]. In this case and for ¯α → 0 we find A → 8¯α2/(3π). Using this result in Eq. (97) we find a rigorous result for the spin suscep- tibility enhancement to linear order in rs: χσyσy σyσy → 1 + χ(0) √2 3π rs . (101) We finally remark that, in the oversimplified case of ultra-short-range interactions, vq = constant = 2πe2
κ , (102) Eqs. (83) and (92) can be solved analytically. In this case indeed all the moments Vm(k, k(cid:48)) of the inter-particle in- teraction but the m = 0 one are zero. The solution of the integral equation (83) is a constant u = [1− (2κaB)−1]−1 and the in-plane spin susceptibility turns out to be equal to u: χσyσy χ(0) σyσy = u = 1 1 − (2κaB)−1 > 1 . (103) D. Interaction corrections to the optical spin Hall conductivity A → (Λ− − Λ+)(cid:20)g(x) − = 2¯α[g(1) − F (1)] . F (x) x (cid:21)x=1 It turns out that the in-plane spin susceptibility χσyσy introduced in the previous Section controls also the "opti- cal spin Hall conductivity" σSH(ω) of the Rashba model. (99) 0.00.20.40.60.81.01.21.4rs0.000.050.100.150.200.250.30Aα=5×10−11eVmα=10−11eVmThis was first shown by Dimitrova27. For the sake of completeness, we briefly summarize here the key steps of the derivation. The spin operator σy q at q = 0, σy tot, satisfies a simple equation of motion: i∂t σy tot = [σy tot, H] = [σy tot, HSOC] = −4imbαjz y , (104) where the q = 0 z-spin current operator in the y direc- tion, jz y , is defined by (see for example Ref. 48) 10 In the high-frequency or clean ωτ → ∞ limit and for noninteracting electrons we have χσyσy → ν0/2 and thus Eq. (109) gives the well-known "universal" value σSH(ωτ → ∞) = e/(8π). As we have seen above, however, electron-electron interactions enhance the high-frequency spin susceptibility, thus yielding an enhancement of the optical spin Hall conductivity. Us- ing Eq. (101) we immediately find that for ωτ → ∞ e 8π(cid:32)1 + √2 3π rs(cid:33) > e 8π . (110) Eq. (110) has to be compared with Eq. (36) in Ref. 27 where an identical result was found modulo the sign of the second term in round brackets. Dimitrova indeed predicts a suppression27 of the spin Hall conductivity due to interactions rather than an enhancement. jz y = 1 2(cid:88)k,i,j ky mb ψ†k,iσz ij ψk,j . (105) σSH = The spin Hall conductivity σSH(ω) describes a z- polarized spin current flowing in the y direction in re- sponse to a homogeneous (q = 0) electric field E = Ex x along the x direction: jz y = σSHEx . (106) For the case of ultra-short-range interactions [see From Eq. (104) it is immediately evident that in the d.c. limit ωτ → 0 (τ is the electron-impurity scattering time) the spin Hall conductivity is zero since in a steady state (cid:104)∂t σy tot(cid:105) = 0. This is the limit that is relevant to d.c. transport. The vanishing of the transport spin Hall con- ductivity in the Rashba model has been widely discussed in the literature (see e.g. Ref. 49). Here we are interested in the high-frequency or clean limit, ωτ → ∞, which can in principle be probed in time- resolved experiments with photo-excited carriers50 or in ballistic transport. In this limit the following analysis is particularly useful. We first notice from Eq. (106) that the optical spin Hall conductivity is related to the spin- current response function by σSH(ω) = ie ω χjz y j(p) x (ω) = ie ω (cid:104)(cid:104)jz y ; j(p) x (cid:105)(cid:105)ω , (107) x = j(p) where j(p) q=0,x is the x component of the q = 0 paramagnetic current operator in Eq. (22). We then sub- stitute Eq. (104) in the response function on the r.h.s. of Eq. (107) and we use Eq. (26). We get 1 1 (cid:104)(cid:104)jz y ; j(p) x (cid:105)(cid:105)ω tot; j(p) x (cid:105)(cid:105)ω tot; j(p) 4mbα(cid:104)(cid:104)∂t σy x (cid:105)(cid:105)ω = − 4mbα(cid:110) − iω(cid:104)(cid:104)(cid:104)σy = − x ](cid:105)(cid:105)(cid:111) 1 ω(cid:104)[σy tot, j(p) − iω 4mb(cid:104)(cid:104)σy tot(cid:105)(cid:105)ω , = − where we have used that (cid:104)(cid:104)σy CM(cid:105)(cid:105)ω = 0 and that [σy tot, P x CM] = 0. The former is a consequence of the fact that total momentum is a conserved quantity (in the ab- sence of impurities) even in the presence of Rashba SOC, while the latter is a trivial commutation rule. Using Eq. (108) in Eq. (107) we finally find tot; σy tot; P x (108) σSH(ω) = e 4mb χσyσy (ω) . (109) Eq. (102)] we find that σSH = e 8π(cid:18)1 − 1 2κaB(cid:19)−1 > e 8π . (111) Before concluding this Section we would like to men- tion that it is possible to derive a relation similar to that in Eq. (109) for the spin Galvanic effect51, i.e. the gener- ation of a charge current in the x direction in response to a homogeneous Zeeman magnetic field B = By y applied along the y direction: jx = σSGBy . (112) Following a procedure analogous to the one that led to Eq. (109), we find σSG(ω) = α gµB 2 χσyσy (ω) , (113) where g is the material Land´e gyromagnetic factor and µB is the Bohr magneton. E. Interaction-induced enhancement of the Rashba SOC From the functional form (71) of the HF bands of the Rashba model it is evident that, for a given density, the real part of the a.c. conductivity (cid:60)e σ(ω) = −e2 lim q→0 ω q2(cid:61)m χρρ(q, ω) , (114) is finite (i.e. absorption occurs) only in a finite interval of frequencies: ∆+ < ω < ∆−, where (cid:40) ∆+ = EHF,+(k(0) ∆− = EHF,+(k(0) F,+) − EHF,−(k(0) F,+) F,−) − EHF,−(k(0) F,−) . (115) in Sect. IV. renormalized by electron-electron interactions13,21. The We can thus view (cid:101)α± as effective Rashba SOC strengths dependencies of (cid:101)α± on rs and α will be illustrated below find that ∆+ = ∆− = 2(cid:101)αkF with (restoring physical (cid:101)α In high-density and/or weak SOC limit (¯α (cid:28) 1) we dimensions for a moment to make contact with earlier work) cos (θ) vqq=2kF sin (θ/2) ,(117) in perfect agreement with the work by Chen and Raikh21. 2π¯h2(cid:90) 2π = 1 + dθ 2π mb α 0 F. The pure Dresselhaus case and the α = ±β case Before turning to the numerical results, we would like to mention that all our results apply equally well to the pure Dresselhaus model, i.e. for α = 0 and finite β. Indeed, replacing the Rashba interaction by the Dressel- haus interaction has the only effect of changing the phase γk of the eigenspinors Ψk,λ(r) in Eq. (8) by π/2, leav- ing the band energies in Eq. (4) unchanged. We have also checked that the statement at the beginning of this Section is true by applying the TDHFA to the pure Dres- selhaus model. The model with α = ±β is much more subtle. For α = β, for example, HSOC reads ψ†k,i(kx + ky)(σx ij − σy ij) ψk,j ψ†k,iσ−ijk+ ψk,j , (118) HSOC = α(cid:88)k,i,j = 2α(cid:88)k,i,j ij−σy In the noninteracting limit these bounds are23,24: ∆(0) 2αk(0) + = F,−. In the interacting case we F,+ and ∆(0) − = 2αk(0) can define ∆+ ≡ 2(cid:101)α+k(0) 2√2¯α(cid:90) Λ− 2√2¯α(cid:90) Λ− rsΛ− = 1 + = 1 + rsΛ+ Λ+ Λ+ α (cid:101)α+ (cid:101)α− α  F,+ and ∆− ≡ 2(cid:101)α−k(0) dx(cid:48) x(cid:48) ¯V1(Λ+, x(cid:48)) F,−, where . (116) dx(cid:48) x(cid:48) ¯V1(Λ−, x(cid:48)) ij)/√2 and k+ ≡ (kx +ky)/√2. From where σ−ij ≡ (σx the second line in Eq. (118) we immediately see that the magneto-electric field generated by the application of a uniform vector potential A(t) is Bext = α (Ax + Ay)(1,−1) , e c (119) i.e. it is parallel to neqα=β = (1,−1)/√2, and does not affect the spin orientation. Thus, in the case α = ±β a uniform vector potential does not reorient spins. The plasmon mass and the Drude weight are thus completely unrenormalized by electron-electron interactions. Of course the plasmon dispersion at finite q will be sensi- tive to interactions. 11 IV. NUMERICAL RESULTS We now turn to a presentation of our main numerical results. As far as the material parameters are concerned, in this article we present results for a 2DEG hosted in a InAs quantum well. In this material the bare elec- tron mass is mb ≈ 0.023 me, where me is the electron mass in vacuum, and the high-frequency dielectric con- stant is
≈ 15. The material Bohr radius turns out to be aB ≈ 348 A. As a consequence, a Wigner-Seitz density parameter rs = 1 corresponds to a rather low electron density, n ≈ 2.6× 1010 cm−2. The SOC strength in InAs varies in the range52 α ≈ (1 − 6) × 10−11 eV m. For the numerical calculations we have used a model interaction potential of the form vq = 2πe2
(q + ξqTF) , (120) where qTF = 2/aB is the 2D Thomas-Fermi screening wave vector in the absence of SOC and ξ ∈ [0, 1] is a di- mensionless control parameter; ξ = 0 implies unscreened Coulomb interactions while ξ = 1 implies Thomas-Fermi screened Coulomb interactions. The TDHFA is well known to overestimate many-body effects when the un- screened Coulomb potential is used. (As we have already mentioned earlier, when the unscreened Coulomb poten- tial is used the TDHFA is exact to first order in e2.) On the other hand, when statically screened Thomas- Fermi interactions are used many-body effects are typi- cally largely underestimated. Thus the spirit of the con- trol parameter ξ is to provide us with upper and lower bounds for the strength of interaction corrections to the various observables we present in this Section. For ξ = 0 the coefficients Vm(k, k(cid:48)) of the angular- momentum expansion in Eq. (62) can be calculated ana- lytically: in dimensionless units these are given by ¯Vm(x, x(cid:48)) = (cid:90) ∞ 0 x(cid:48)m xm+1 = dt Jm(tx)Jm(tx(cid:48)) Γ(m + 1/2) Γ(m + 1)Γ(1/2) × 2F1(m + 1/2, 1/2, m + 1, x(cid:48)2/x2) , (121) for x > x(cid:48). Here Jm(z), Γ(z), and 2F1(a, b, c, z) are the Bessel function, the Euler Gamma function, and the hy- pergeometric function, respectively. For x < x(cid:48) one needs to interchange x ↔ x(cid:48) in Eq. (121). For ξ (cid:54)= 0 the pseu- dopotentials ¯Vm(x, x(cid:48)) have to be calculated numerically. Fig. 2 shows the HF bands EHF,±(k) in Eq. (71) for unscreened Coulomb interactions, while Fig. 3 illustrates the HF self-energies Σ0(k) and Σ1(k), defined in Eqs. (70) and (68), respectively. Note that Σ0(k) is negative while Σ1(k) is positive. In Figs. 4a) and b) we present the minority m(cid:63) + and majority m(cid:63) − effective masses as functions of rs for Thomas-Fermi screened interactions, as calculated from 12 ∼ 1. tions. This result stems from exchange interactions and is the dominant effect at weak coupling43, i.e. for rs < The suppression of the quasiparticle effective mass for α = 0 shown in Figs. 4a)-4b) extends to larger values of rs because of the strong value of the screening parameter (ξ = 1) we have used. We notice that the impact of SOC is opposite in different bands [this is ultimately due to the dependence of the factor k(0) F,λ in the l.h.s. of Eq. (72) on α]: as we can see in Fig. 4a), SOC further suppresses the quasiparticle effective mass in the minority band. On the other hand, as shown in Fig. 4b), SOC enhances the quasiparticle effective mass in the majority band. In the inset to Fig. 4b) we plot the quantities ∆v(cid:63) v(cid:63) λ − v(cid:63) rs (rs = 0.25), where the quasiparticle velocities v(cid:63) defined by λ ≡ λα=0 as functions of ¯α and for a fixed value of λ are v(cid:63) λ = ∂EHF,λ(k) ∂k . (122) (cid:12)(cid:12)(cid:12)(cid:12)k=k(0) F,λ Differently from the effective mass results, we see that ∆v(cid:63) λ is practically the same for both λ = ± bands and that corrections linear in ¯α are absent, in agreement with Refs. 25,39. In Fig. 5 we report the solution δBT,1(k) of the inte- gral equation (79) for both unscreened and screened in- teractions. It is important to note that δBT,1(k) in units of the bare effective magnetic field Bext = eAα/(¯hc) is larger than unity. Kinks are seen in δBT,1(k) at k = k(0) F,λ, which are especially visible at ξ = 0. We also clearly see how the amplitude of δBT,1(k) decreases with increasing ξ. A plot of the in-plane spin susceptibility χσyσy in units of the noninteracting value χ(0) σyσy = ν0/2 is presented in Fig. 6. As expected, the in-plane spin susceptibility is en- hanced by electron-electron interactions. Overscreening their strength by setting ξ = 1 in Eq. (120) substantially reduces the enhancement53. Note also that increasing the SOC strength α the ratio χσyσy /χ(0) σyσy increases. Fig. 7 shows the most important result of this work, i.e. the renormalization of the Drude weight D due to interactions. There we indeed plot the ratio between D and its noninteracting value D0. Since the spin suscepti- bility is enhanced by interactions, D is suppressed. The suppression is quite large within truly first-order pertur- bation theory (ξ = 0) and increases with increasing α. The plasmon mass is thus enhanced by the interactions and thus the plasmon frequency is reduced by the com- bined effect of SOC and interactions with respect to the standard frequency of plasmons in the absence of SOC. The enhancement of SOC due to interactions is il- lustrated in Figs. 8-9 where we have plotted (cid:101)α±/α as calculated from Eq. (116). These two figures refer to two different values of ξ. For the sake of comparison, in Fig. 9 we have also plotted the weak-SOC result by Chen and Raikh21 [see Eq. (117)]. From Fig. 8 we see that the enhancement of SOC is pretty large for unscreened FIG. 2: (Color online) The renormalized Hartree-Fock energy bands EHF,±(k) (in units of εF,0) as functions of k (in units of kF) for rs = 1 and α = 5 × 10−11 eV m for the case of unscreened (ξ = 0) Coulomb interactions. The dashed (blue) line denotes the minority band [EHF,+(k)] while the solid (red) line denotes the majority band [EHF,−(k)]. The thin black lines are the energy bands for the noninteracting case. The vertical lines denote the location of ±k(0) F,±. FIG. 3: (Color online) The Hartree-Fock self energies Σ0(k) (solid line) and Σ1(k) (dashed line) (in units of εF,0) as functions of momentum k (in units of kF) for rs = 1 and α = 5 × 10−11 eV m. The vertical lines denote the location of ±k(0) F,±. Eq. (72). (It is very well know that, to avoid artifacts of the HF theory, it is necessary to screen the Coulomb interaction to get meaningful results for the quasiparti- cle effective mass. The derivative of the HF quasiparticle energy indeed diverges at the Fermi surface and hence the quasiparticle effective mass vanishes.) In the case of no SOC (α = 0) one finds that the quasiparticle ef- fective mass is suppressed by electron-electron interac- −2−1012k/kF−2−10123456−2−1012k/kF−1.5−1.0−0.50.00.51.0Σ1Σ013 FIG. 4: (Color online) The minority [panel a)] and majority [panel b)] quasiparticle effective masses m(cid:63)± (in units of the bare mass mb) as functions of rs for different values of α. These results have been obtained by using the definition (72) and fully-screened Thomas-Fermi interactions (ξ = 1). The solid line represents the classic result by Janak54: our results for α = 0 (filled circles) are in excellent agreement with the analytical expression (16) in Ref. 54. The inset to panel b) illustrates ∆v(cid:63)± ≡ v(cid:63)± − v(cid:63)±α=0 (in units of vF ≡ kF/mb) as functions of ¯α and for rs = 0.25 (dashed and dotted lines). The solid line is the weak-SOC analytical result (74) in Ref. 25. Notice that our numerical results extend up to a large value of the SOC constant since ¯α = 0.7 corresponds to α ∼ 38 × 10−11 eV m. FIG. 5: (Color online) The vertex correction δBT,1(k) [in units of Bext = eAα/(¯hc)] as a function of k (in units of kF) for rs = 1 and different values of α. Panel a) Results for unscreened Coulomb interactions (ξ = 0). Panel b) Results for fully-screened Thomas-Fermi interactions (ξ = 1). Note that when expressing δBT,1(k) in units of Bext a factor α is extracted. Also note that for α = 5 × 10−11 eV m, k(0) F,−/kF (cid:39) 1.3. For α = 10−11 eV m, k(0) F,+/kF (cid:39) 0.6 and k(0) F,+/kF (cid:39) 0.9 and k(0) F,−/kF = 1.1. Coulomb interactions and that it decreases for increasing SOC strength. A. Taking into account the density dependence of the Rashba SOC Until now we have treated the Wigner-Seitz parame- ter rs (or density) and the Rashba SOC constant α as two independent parameters. This is similar in spirit to what has been done for decades in the context of tunnel- coupled double quantum wells where the single-particle symmetric-to-antisymmetric gap ∆SAS and density have been treated as independent parameters (see e.g. Ref. 55 and references therein to earlier work). In reality, when a single gate voltage is applied to the 2DEG to change its density (and thus the rs value) the asymmetry of the quantum well which hosts the 2DEG changes too56. This in turn changes α. In a simple single- band model with infinite barriers the SOC strength α is 0.00.20.40.60.81.01.21.4rs0.30.40.50.60.70.80.91.0m?+/mb(a)α=0,Janakα=0α=10−11eVmα=5×10−11eVm0.00.20.40.60.81.01.21.4rs0.91.01.11.21.31.41.51.61.7m\x{FFFF}−/mb(b)0.00.10.20.30.40.50.60.7¯α−0.30−0.25−0.20−0.15−0.10−0.050.00∆v\x{FFFF}λ/vFrs=0.25Saraga−Lossλ=−λ=+0.00.51.01.52.0k/kF1.21.41.61.82.02.22.4δBT,1(a)ξ=0α=5×10−11eVmα=10−11eVm0.00.51.01.52.0k/kF1.101.121.141.161.18δBT,1(b)ξ=1α=5×10−11eVmα=10−11eVm14 FIG. 6: (Color online) The in-plane spin susceptibility χσy σy (in units of the noninteracting value, χ(0) σy σy = ν0/2) as a function of rs and for different values of α. Panel a) Results for unscreened Coulomb interactions (ξ = 0). Panel b) Results for fully-screened Thomas-Fermi interactions (ξ = 1). FIG. 7: (Color online) The Drude weight D [in units of the noninteracting value, D0 = πe2(n/mb − α2ν0/2)], calculated from Eq. (93), is plotted as a function of rs and for different values of α. Panel a) Results for unscreened Coulomb interactions (ξ = 0). Panel b) Results for fully-screened Thomas-Fermi interactions (ξ = 1). given by52,57,58 α = eαso(cid:104)E(cid:105) ≈ eαson
, (123) with αso = 117 A2 for bulk InAs. Here we have used that the electric field in the well is given by (cid:104)E(cid:105) = nd/
where the density of the donors nd has been approximated by the density of electrons n. In Fig. 10 we present numerical results for the ratio pendence of α according to Eq. (123). From this plot we clearly see that the difference between the effective SOC (cid:101)α±/α calculated by taking into account the density de- constants (cid:101)α+ and (cid:101)α− becomes negligibly small and that the ratio (cid:101)α±/α changes by roughly sixty percent when density is changed over three orders of magnitude (for unscreened Coulomb interactions). The enhancement of SOC due to interactions increases with decreasing den- sity. Overscreening Coulomb interactions washes out this effect yielding a tiny renormalization over the same den- sity range. Finally, in Fig. 11 we show the spin susceptibility en- hancement χσyσy /χ(0) σyσy as a function of rs calculated by taking into account the dependence of the bare α on density via Eq. (123). Note that for unscreened Coulomb interactions the enhancement can be as large as 20− 30% for n ≈ 1010 cm−2 (recall that in InAs rs = 1 corresponds to an electron density ≈ 2.6 × 1010 cm−2). 0.00.20.40.60.81.01.21.4rs1.001.051.101.151.201.251.301.35χσyσyξ=0(a)Eq.(101)α=5×10−11eVmα=10−11eVm0.00.20.40.60.81.01.21.4rs1.001.051.101.151.201.251.301.35χσyσyξ=1(b)α=5×10−11eVmα=10−11eVm0.00.20.40.60.81.01.21.4rs0.800.850.900.951.00Dξ=0(a)RPAα=3×10−11eVmα=5×10−11eVm0.00.20.40.60.81.01.21.4rs0.800.850.900.951.00Dξ=1(b)RPAα=3×10−11eVmα=5×10−11eVm15 (−) bands as functions of rs and for different values of α. These results refer to unscreened Coulomb interactions (ξ = 0). FIG. 8: (Color online) Renormalized SOC coupling strengths(cid:101)α± (in units of the bare value α) for the minority (+) and majority FIG. 9: (Color online) Same as in Fig. 8 but for ξ = 1. The thin solid (black) line in both panels labels the result of Ref. 21 [see Eq. (117)], which is asymptotically exact in the weak SOC limit. V. CONCLUSIONS In summary, we have studied the long-wavelength plasmon dispersion and the Drude weight of a two- dimensional electron gas with Rashba spin-orbit cou- pling. We have shown that these measurable quantities are sensitive to electron-electron interactions due to bro- ken Galileian invariance and we have discussed in de- tail why the random phase approximation is not capa- ble of describing the collective dynamics of these systems even at very long wavelengths. We have then presented approximate microscopic calculations of these quantities based on the so-called time-dependent Hartree-Fock ap- proximation. We have found that interactions enhance the plasmon mass and suppress the Drude weight. These findings can in principle be tested experimen- tally by inelastic light scattering, electron energy loss, In- and far-infrared optical-absorption measurements. elastic light scattering59 has already been extensively used to measure the plasmon dispersion in GaAs quan- tum wells60,61. Notable deviations from the predictions of the random phase approximation have been observed61 at finite momentum transfer q and at low densities. We hope that similar studies can be performed systemati- cally in asymmetric n-doped quantum wells with tunable spin-orbit coupling. Last but not least, we have also computed quantita- tively the renormalization of the Rashba spin-orbit cou- pling constant due to electron-electron interactions and the interaction corrections to the clean-limit spin Hall conductivity. In the future we plan to extend these studies to the complete spin-orbit-coupling model Hamiltonian (3) with both α and β finite. As already stressed in the main body 0.00.20.40.60.81.01.21.4rs1.01.21.41.61.82.0eα+/αξ=0(a)α=10−11eVmα=5×10−11eVm0.00.20.40.60.81.01.21.4rs1.01.21.41.61.82.0eα−/αξ=0(b)α=10−11eVmα=5×10−11eVm0.00.20.40.60.81.01.21.4rs1.001.011.021.031.041.051.061.071.08eα+/α(a)ξ=1α=10−11eVmα=5×10−11eVm0.00.20.40.60.81.01.21.4rs1.001.011.021.031.041.051.061.071.08eα−/α(b)ξ=1α=10−11eVmα=5×10−11eVm16 FIG. 10: (Color online) The renormalized SOC constants(cid:101)α± [in units of the noninteracting value α] are plotted as functions of the logarithm of density (expressed in units of cm−2). In this figure we have taken into account the density dependence of the bare SOC coupling α in the simple approximation given in Eq. (123). Panel a) Results for unscreened Coulomb interactions (ξ = 0). Panel b) Results for fully-screened Thomas-Fermi interactions (ξ = 1). of this article, this complicates things quite a bit since the resulting ground state does not possess rotational invariance. It is worth exploring also other spin-orbit- coupled two-dimensional quantum liquids such as hole gases in quantum wells which have a very rich single- particle band structure41. Acknowledgments FIG. 11: (Color online) The spin susceptibility enhancement χσy σy /χ(0) σy σy as a function of the logarithm of density (ex- pressed in units of cm−2). In this figure we have taken into account the density dependence of the bare SOC coupling α in the simple approximation given in Eq. (123). A.A., T.J., and M.P. gratefully acknowledge fund- ing from the European Community's Seventh Frame- work Programme (FP7/2007-2013) under grant agree- ment n. 215368 (SemiSpinNet). T.J. also acknowledges support from Czech Republic Grants AV0Z10100521, KAN400100652, LC510, and Preamium Academiae. J.S. was supported by NSF grant No. DMR-0547875 and by the Research Corporation Cottrell Scholar Award. G.V. was supported by the National Science Foundation under grant number DMR-0705460. ∗ Electronic address: amit.agarwal@sns.it † Electronic address: m.polini@sns.it; URL: http://qti. sns.it 1 I. Zuti´c, J. Fabian, and S. Das Sarma, Rev. Mod. Phys. 76, 323 (2004); J. Fabian, A. Matos-Abiague, C. Ertler, P. Stano, and I. Zuti´c, Acta Physica Slovaca 57, 565 (2007); also available as arXiv:0711.1461v1. MacDonald, Phys. Today 60(8), 35 (2007); M.I. Katsnel- son, Mater. Today 10, 20 (2007); A.K. Geim and K.S. Novoselov, Nature Mater. 6, 183 (2007). 3 P. Avouris, Z. Chen, and V. Perebeinos, Nature Nanotech. 2, 605 (2007). 4 B. Andrei Bernevig, T.L. Hughes, and S.-C. Zhang, Science 314, 1757 (2006). 2 A.K. Geim, Science 324, 1530 (2009); A.H. Castro Neto, F. Guinea, N.M.R. Peres, K.S. Novoselov, and A.K. Geim, Rev. Mod. Phys. 81, 109 (2009); A.K. Geim and A.H. 5 M. Konig, S. Wiedmann, C. Brune, A. Roth, H. Buhmann, L.W. Molenkamp, X.-L. Qi, and S.-C. Zhang, Science 318, 766 (2007). 11.011.512.012.513.0log10(n)1.01.11.21.31.41.51.6(a)ξ=0eα−/αeα+/α11.011.512.012.513.0log10(n)1.0201.0251.0301.0351.0401.0451.0501.055(b)ξ=1eα−/αeα+/α10.010.511.011.512.012.513.0log10(n)1.001.051.101.151.201.251.301.35χσyσyRPAξ=0ξ=16 M. Konig, H. Buhmann, L.W. Molenkamp, T. Hughes, C.- X. Liu, X.-L. Qi, and S.-C. Zhang, J. Phys. Soc. Jpn. 77, 031007 (2008). 7 C. Brune, A. Roth, E.G. Novik, M. Konig, H. Buh- mann, E.M. Hankiewicz, W. Hanke, J. Sinova, and L.W. Molenkamp, Nature Phys. 6, 448 (2010). 8 J. Moore, Nature Phys. 5, 378 (2009); X.-L. Qi and S.- C. Zhang, Phys. Today 63(1), 33 (2010); J.E. Moore, Nature 464, 194 (2010); M.Z. Hasan and C.L. Kane, arXiv:1002.3895v1. 9 Y. Xia, D. Qian, D. Hsieh, L. Wray, A. Pal, H. Lin, A. Bansil, D. Grauer, Y.S. Hor, R.J. Cava, and M.Z. Hasan, Nature Phys. 5, 398 (2009). 10 H. Zhang, C.-X. Liu, X.-L. Qi, X. Dai, Z. Fang, and S.-C. Zhang, Nature Phys. 5, 438 (2009). 11 T. Zhang, P. Cheng, X. Chen, J.-F. Jia, X. Ma, K. He, L. Wang, H. Zhang, X. Dai, Z. Fang, X. Xie, and Q.-K. Xue, Phys. Rev. Lett. 103, 266803 (2009). 12 Strictly speaking, the low-energy Dirac-Weyl effective Hamiltonian describing electrons close to the Dirac point in graphene2 does not stem from relativistic effects but is purely an effect of the 2D periodic potential of the lattice. The pseudorelativistic character of the low-energy Hamil- tonian in graphene and the consequent lack of Galileian invariance are merely a crystal field effect. 13 A. Shekhter, M. Khodas, and A.M. Finkel'stein, Phys. Rev. B 71, 165329 (2005). 14 A.-K. Farid and E.G. Mishchenko, Phys. Rev. Lett. 97, 096604 (2006). 15 M. Polini, A.H. MacDonald, arXiv:0901.4528v1 (unpublished). and G. Vignale, 16 A. Principi, M. Polini, and G. Vignale, Phys. Rev. B 80, 075418 (2009). 17 W.-K. Tse and A.H. MacDonald, Phys. Rev. B 80, 195418 (2009). 18 S.H. Abedinpour, G. Vignale, A. Principi, M. Polini, W.- K. Tse, and A.H. MacDonald, in preparation. 19 S. Raghu, S.B. Chung, X.-L. Qi, and S.-C. Zhang, Phys. Rev. Lett. 104, 116401 (2010). 20 G.F. Giuliani and G. Vignale, Quantum Theory of the Electron Liquid (Cambridge University Press, Cambridge, 2005). 21 G.-H. Chen and M.E. Raikh, Phys. Rev. B 59, 5090 (1999). 22 G.-H. Chen and M.E. Raikh, Phys. Rev. B 60, 4826 (1999). 23 L.I. Magarill, A.V. Chaplik, and M.V. ´Entin, JEPT 92, 153 (2001). 24 E.G. Mishchenko and B.I. Halperin, Phys. Rev. B 68, 045317 (2003). 25 D.S. Saraga and D. Loss, Phys. Rev. B 72, 195319 (2005). 26 X.F. Wang, Phys. Rev. B 72, 085317 (2005). 27 O.V. Dimitrova, Phys. Rev. B 71, 245327 (2005). 28 M. Pletyukhov and V. Gritsev, Phys. Rev. B 74, 045307 (2006). 29 J. Schliemann, Phys. Rev. B 74, 045214 (2006). 30 A.S. N´unez, R.A. Duine, and A.H. MacDonald, unpub- lished (2006). 31 M. Pletyukhov and S. Konschuh, Eur. Phys. J. B 60, 29 (2007). 32 S. Chesi and G.F. Giuliani, Phys. Rev. B 75, 153306 (2007) and ibid. 75, 155305 (2007); S. Chesi, Ph.D. thesis, Purdue University (2007). 33 C. Li and X.G. Wu, Appl. Phys. Lett. 93, 251501 (2008). 34 S.M. Badalyan, A. Matos-Abiague, G. Vignale, and J. 17 Fabian, Phys. Rev. B 79, 205305 (2009). 35 A. Ambrosetti, F. Pederiva, E. Lipparini, and S. Gandolfi, Phys. Rev. B 80, 125306 (2009). 36 I.A. Nechaev, P.M. Echenique, and E.V. Chulkov, Phys. Rev. B 81, 195112 (2010). 37 S.M. Badalyan, A. Matos-Abiague, G. Vignale, and J. Fabian, Phys. Rev. B 81, 205314 (2010). 38 R.A. Zak, D.L. Maslov, and D. Loss, Phys. Rev. B 82, 115415 (2010) . 39 S. Chesi and G.F. Giuliani, arXiv:1008.2227v1. 40 S. Chesi and G.F. Giuliani, arXiv:1008.3729v1. 41 R. Winkler, Spin-Orbit Coupling Effects in Two- Dimensional Electron and Hole Systems (Springer, Berlin, 2003). 42 M. Trushin, K. V´yborn´y, P. Moraczewski, A.A. Kovalev, J. Schliemann, and T. Jungwirth, Phys. Rev. B 80, 134405 (2009). 43 R. Asgari, B. Davoudi, M. Polini, G.F. Giuliani, M.P. Tosi, and G. Vignale, Phys. Rev. B 71, 045323 (2005). physical (causal) density-density response function and scribes the response to the screened potential and is defined diagramatically as the sum of all the "proper" diagrams 20, i.e. those diagrams that cannot be separated into two parts, each one containing one of the external vertices, by the cutting of a single interaction line. In 2D ε(q → 0, ω) → 44 (cid:101)χρρ(q, ω) = χρρ(q, ω)ε(q, ω), where χρρ(q, ω) is the usual ε(q, ω) is the dielectric function. Physically (cid:101)χρρ(q, ω) de- 1 and thus(cid:101)χρρ(q → 0, ω) = χρρ(q → 0, ω). 45 V.M. Edelstein, Solid State Commun. 73, 233 (1990). 46 We are here dropping the so-called Hartree terms, which have q = 0, and are thus canceled by the presence of the uniform background of neutralizing positive charge. 47 The restriction εF > 0 (i.e. that both chiral Rashba bands are occupied) implies that the following inequality must be satisfied: ¯α < √ 2/2. 48 J. Sinova, D. Culcer, Q. Niu, N.A. Sinitsyn, T. Jungwirth, and A.H. MacDonald, Phys. Rev. Lett. 92, 126603 (2004). 49 R. Raimondi and P. Schwab, Phys. Rev. B 71, 033311 (2005). 50 B.A. Ruzicka, K. Higley, L.K. Werake, and H. Zhao, Phys. Rev. B 78, 045314 (2008). 51 S.D. Ganichev, E.L. Ivchenko, V.V. Bel'kov, S.A. Tarasenko, M. Sollinger, D. Weiss, W. Wegscheider, and W. Prettl, Nature 417, 153 (2002). 52 J. Nitta, T. Akazaki, H. Takayanagi, and T. Enoki, Phys. Rev. Lett. 78, 1335 (1997). 53 S. Yarlagadda and G.F. Giuliani, Phys. Rev. B 40, 5432 (1989). 54 J.F. Janak, Phys. Rev. 178, 1416 (1969). 55 S.H. Abedinpour, M. Polini, A.H. MacDonald, B. Tanatar, M.P. Tosi, and G. Vignale, Phys. Rev. Lett. 99, 206802 (2007). 56 We remark that it is experimentally possible to use a front and a back gate to change rs without changing α (i.e. to change these two parameters independently): see e.g. S.J. Papadakis, E.P. De Poortere, H.C. Manoharan, M. Shayegan, and R. Winkler, Science 283, 2056 (1999). 57 W. Knap, C. Skierbiszewski, A. Zduniak, E. Litwin- Staszewska, D. Bertho, F. Kobbi, J.L. Robert, G.E. Pikus, F.G. Pikus, S.V. Iordanskii, V. Mosser, K. Zekentes, and Yu.B. Lyanda-Geller, Phys. Rev. B 53, 3912 (1996). 58 D. Grundler, Phys. Rev. Lett. 84, 6074 (1999). 59 For a recent review see e.g. V. Pellegrini and A. Pinczuk, Phys. Stat. Sol. (B) 243, 3617 (2006). 60 D. Olego, A. Pinczuk, A.C. Gossard, and W. Wiegmann, Phys. Rev. B 25, 7867 (1982); A. Pinczuk, M.G. Lamont, and A.C. Gossard, Phys. Rev. Lett. 56, 2092 (1986); G. Fasol, N. Mestres, H.P. Hughes, A. Fischer, and K. Ploog, ibid. 56, 2517 (1986). 61 C.F. Hirjibehedin, A. Pinczuk, B.S. Dennis, L.N. Pfeiffer, and K.W. West, Phys. Rev. B 65, 161309 (2002). 18 

1006.0630

1

1006

High-order cumulants in the counting statistics of asymmetric quantum dots

Measurements of single electron tunneling through a quantum dot using a quantum point contact as charge detector have been performed for very long time traces with very large event counts. This large statistical basis is used for a detailed examination of the counting statistics for varying symmetry of the quantum dot system. From the measured statistics we extract high order cumulants describing the distribution. Oscillations of the high order cumulants are observed when varying the symmetry. We compare this behavior to the observed oscillation in time dependence and show that the variation of both system variables lead to the same kind of oscillating response.

cond-mat.mes-hall

cond-mat

High-order cumulants in the counting statistics of asymmetric quantum dots Christian Fricke, Frank Hohls, Nandhavel Sethubalasubramanian, Lukas Fricke, and Rolf J. Haug Institut fur Festkorperphysik, Leibniz Universitat Hannover, 30167 Hannover, Germany. (Dated: November 14, 2018) Measurements of single electron tunneling through a quantum dot using a quantum point contact as charge detector have been performed for very long time traces with very large event counts. This large statistical basis is used for a detailed examination of the counting statistics for varying symmetry of the quantum dot system. From the measured statistics we extract high order cumulants describing the distribution. Oscillations of the high order cumulants are observed when varying the symmetry. We compare this behavior to the observed oscillation in time dependence and show that the variation of both system variables lead to the same kind of oscillating response. PACS numbers: 72.70.+m, 73.23.Hk, 73.63.Kv Current fluctuations in mesoscopic systems allow to obtain information on transport that is not accessible from the average current alone1. Measurements of the shot noise as a first step beyond the mean have been de- ployed successfully to examine correlations in the electron transport through a metallic island2 and through semi- conducting quantum dot systems3,4, but the extraction of higher moments directly from the current fluctuations is a demanding task5 and has not yet been achieved for transport through quantum dots. In contrast higher mo- ments are naturally accessible in the context of full count- ing statistics (FCS)6,7. With the use of a quantum point contact (QPC) as a non-invasive charge detector with suf- ficiently fast time response, FCS became experimentally feasible in quantum dot physics8 -- 11. In a measurement of the 4th and 5th cumulant complex behavior as a func- tion of asymmetry and integration time with local min- ima was found11. A significantly improved experimen- tal technique has made the extraction of very high-order cumulants possible, revealing an oscillating behavior as function of integration time that strongly increases with the order of the moment12,13. Such oscillations are also predicted theoretically in quantum optics15 and particle physics14.The theoretical treatment has shown that these oscillations are a universally expected phenomenon for most physical systems and that in quantum dots they should show up prominently as function of the barrier asymmetry12,16. In this letter we present the experimental verification of the predicted oscillations in the higher cumulants as function of barrier asymmetry in a quantum dot system. To this aim we employ real time single electron count- ing with a high bandwidth detector. This enables us to measure the full counting statistics for electron trans- port through a quantum dot on the µs-timescale, allowing us to gather a unprecedented large amount of counting events. With this large statistical basis we can exam- ine the functional dependence of high order cumulants, which describe the statistical properties of the system, on the tunneling barrier asymmetry and also on the count- ing time. The observed oscillations as function of both parameters show large similarities that strongly support the universal nature of these oscillations. FIG. 1. (a) Average current through the quantum dot ver- sus gate voltage at the in-plane gates G1 and G2. Current is measured counting electron by electron. A bias of 1.3 mV is applied between source and drain contact of the quantum dot. Inset: AFM image of the structure. (b) Measured tunneling rates ΓS (source), ΓD (drain) and the electron transport rate ΓΣ with respect to the asymmetry parameter a. The asym- metry parameter is modified by changing the quantum dot gate configuration along the line shown in (a). Our device is based on a GaAs/AlGaAs heterostruc- ture containing a two-dimensional electron system (2DES) 34 nm below the surface. The electron density is ρ = 4.6·1015 m−2, and the mobility is µ = 64 m2/Vs. We have used an atomic force microscope (AFM) to define the quantum dot (QD) and the quantum point contact (QPC) structure by local anodic oxidation (LAO) of the surface17,18; the 2DES below the oxidized surface is de- pleted and so insulating lines are defined. An AFM image of our device is presented in Fig. 1. The dark lines depict the insulating oxide barriers writ- ten by the AFM. The QPC (left area) is separated from the QD structure (right area) by an insulating line. The QPC can be electrically tuned using the in-plane gate G3. The current through the QPC is amplified by a current amplifier and detected in a time-resolved man- ner with a sampling rate of 500 kHz. To achieve a high 0 1 0 2 n u J 3 ] l l a h - s e m . t a m - d n o c [ 1 v 0 3 6 0 . 6 0 0 1 : v i X r a 400390380370-175-170-165-160-155-1501.00.80.60.4aVG1 (mV)VG2 (mV)rate (Hz)10310430000counts / sΓSΓDΓΣ(a)(a)(b)300 nmSQPCDQPCG3G2G1SDSQPCDQPCG3G2G1SDSDbandwidth charge detection a custom low capacitance wiring is used to reduce technical noise caused by the input voltage noise of the amplifier. The bandwidth of the charge detection is only limited by the current ampli- fier bandwidth and slightly exceeds 100 kHz. The QPC bias is chosen sufficiently small to avoid back-action on the QD19,20. The QD is coupled to source and drain electrodes via two tunneling barriers which can be sepa- rately controlled with gate voltages VG1 and VG2. These gates are also used to tune the number of electrons on the QD. The sample is placed in a He3 refrigerator reaching temperatures down to 400 mK. Setting the device to a situation where electrons can enter and leave the quantum dot results in a fluctuating charge on the quantum dot. The electron number on the dot is sequentially changing from N to N+1 and from N+1 to N electrons. This leads to a corresponding change of the potential at the QPC which has a working point on the edge of the first conductance step. Thus the QPC acts as a sensitive charge detector as each change in the potential due to a changing electron number on the dot leads to a current change at the QPC. The average current versus VG1 and VG2 as measured by single electron counting is shown in Fig. 1a. A bias of 1.3 mV is applied to the quantum dot. The white areas without electron tunneling events correspond to Coulomb blockade with fixed electron number on the quantum dot. In between wide stripes with a significant number of tun- neling events are observed due to the applied QD bias voltage. The charging energy of the quantum dot is 1.6 meV, the single particle levelspacing is 0.27 meV. The nonzero electron temperature leads to an ambiguous tun- neling direction at the edges of these stripes where the Fermi-level of a lead is in resonance with the dot level. To analyze only unidirectional transport, the system is tuned to a situation where the transport state of the dot is sufficiently far from the source and drain Fermi energies, thus allowing no back-tunneling from the dot to the source contact or from drain to the dot (compare inset in Fig. 1b). The symmetry of the tunneling rates are studied using both quantum dot gates to keep the transport in the unidirectional regime and concurrently to change the asymmetry of the tunneling barriers. The line drawn in Fig. 1a indicates the gate voltages used in the following. The resulting tunneling rates are displayed in Fig. 1b with respect to the asymmetry factor a = (ΓS − ΓD)/(ΓS + ΓD) with ΓS and ΓD the tunneling rates between source resp. drain and the quantum dot. For smaller VG1 and more positive VG2 (upper left corner of Fig. 1a) the system is rather asymmetric with ΓS ≈ 20 kHz and ΓD ≈ 0.2 kHz. Following the line ΓS drops down to 2.3 kHz and ΓD increases to 1.2 kHz. Thus the asymmetry factor a is varied from 0.98 to 0.3. The total rate of electron transfer ΓΣ = ΓS · ΓD/(ΓS + ΓD) reaches its highest value ΓΣ = 0.9 kHz at a = 0.3 and drops with increasing asymmetry a. The detection of individual tunneling events allows us to extract the full counting statistics of electron transport 2 FIG. 2. Cumulants of order m = 2 . . . 15 as function of asymmetry factor a. While the low order cumulants show a monotonic behavior the cumulants with order m > 4 reveal distinct oscillations with factorial increase in amplitude (up to 10,000 for the 15th cumulant). The cumulants are determined for a constant dimensionless counting time interval ΓS · t0 = 3.2. through the quantum dot. Before going into this analysis it is useful to introduce a dimensionless time ΓS·t which is equivalent to setting ΓS equal to unity. This is necessary to separate time and asymmetry dependence as accord- ing to Ref.12 time dependent oscillations are universal on the timescale ΓS·t. Then the system is solely governed by the asymmetry parameter a. Thus the statistical prop- erties are also completely determined by this single sys- tem parameter a. Of course we have to note that in our experiment another rate exists, namely the detector bandwidth Γdet. However, if we fix this rate relatively to ΓS we again have a system whose statistical proper- ties are solely varied by the asymmetry. We realize this fixation of the detector bandwidth by a post-selection of detected events with an artificial detector bandwidth of Γdet = 4 · ΓS. We will now examine the counting statistics of the quantum dot. This can be done either by direct anal- ysis of the whole probability distribution of the number of transferred electrons n in a certain time interval or more common by the cumulants cm of this distribution. The first cumulant is the mean of the number of trans- ferred electrons, c1 = (cid:104)n(cid:105), the second, c2, is the variance, and the third is the skewness. Cumulants of higher or- der m are very sensitive to the details of the transport process. Recently it was found experimentally that the high-order cumulants of quantum dot transport oscillate as function of the counting time interval.12 In the same paper it is stated that such oscillations are expected uni- versally for nearly any system when changing a relevant parameter, e.g. the symmetry of the tunneling rates in a quantum dot system12. For our experimental study of this predicted symmetry dependence of high-order cumu- lants we varied the asymmetry parameter a and measured for each a about 1 million tunneling events. This statis- tical sample enables us to determine cumulants up to the -1.0-0.50.00.51.01.00.80.60.4a-150-100-5000501001501.00.80.60.4a1.00.80.60.4am = 23456789101112-10000-5000500010000131415cm / c13 ated for a certain choice of the dimensionless length of the counting interval. We can also determine the dependence on the dimensionless counting interval for each value of the asymmetry a, thereby mapping out the cumulants as function of both parameters. The result obtained for the 15th cumulant c15 is shown in Fig. 3a. The figure nicely reveals the similarity in the dependencies on asymme- try and on dimensionless time -- the general features are nearly symmetric with respect to the diagonal. For a further illustration of these similarities we plot two pro- files in Fig. 3b and 3c, taken at fixed asymmetry resp. fixed normalized time. Both curves, c15 as function of time ΓS · t0 resp. asymmetry a, show oscillations of very similar shape and amplitude. This correspondence be- tween the different parameters nicely demonstrates that the occurrence and the strength of these oscillations are of universal nature12. It is also interesting to examine the behavior at long times or towards the case of symmetric rates, a = 0. For long times the dependence on ΓS · t0 becomes weak and in the long time limit the cumulants will become sole function of the asymmetry. Towards a = 0 the dependence on asymmetry flattens out and the cumulants depend mainly on time. This transitions again reveal the similarities in the dependence on these different parameters and support the underlying univer- sal nature of the oscillatory behavior. In conclusion, we have realized a high-bandwidth de- tection of single electron transport in a quantum dot with a sufficiently large event number to examine the symme- try dependence of the cumulants of the counting proba- bility distribution up to the 15th order. We have verified the theoretically predicted oscillation of the high order cumulants as function of the quantum dot barrier asym- metry. Furthermore we have performed a 2d-parameter study of the cumulants as function of both asymmetry and time. Our experimental results nicely reveal the sim- ilarity of the dependence on these different parameters, thereby demonstrating the underlying universal nature of the cumulant oscillations. We thank C. Flindt for many fruitful discussions. The work was supported by the Federal Ministry of Educa- tion and Research of Germany via nanoQUIT and the German Excellence Initiative via QUEST. FIG. 3. Time and symmetry dependence for the 15th cumu- lant. Oscillations can be observed for varying time base ΓS ·t0 and symmetry a. Two profile lines are shown displaying the substantial similarity of both types of oscillations. 15th order with high precision. In Fig. 2 the normalized cumulants cm/c1 are shown as function of the asymmetry factor a for m = 2 . . . 15. For high asymmetry (a → 1) all nomalized cumulants start from a value of 1 due to the Poissonian limit of a single dominant barrier. The first cumulants then drop monotonically towards the symmetric limit. For m ≥ 4 a more complicated behavior can be observed. Begin- ning with the fourth order cumulant a minimum starts to develop, shifting to higher asymmetry and increasing in amplitude with rising order. For the 6th order an addi- tional maximum appears that also gets more prominent with rising order of the cumulants. This trend continues for even higher cumulants, the number of oscillation in- creases and also the amplitude rises dramatically. For the 12th order values up to 145 are reached and the ampli- tude of the 15th cumulant already exceeds 10000. This factorial increase was shown already for oscillations as function of time12 and appears here again, showing the universal nature of the oscillations. The asymmetry dependence shown in Fig. 2 was evalu- 1 Y. M. Blanter, M. Buttiker, Physics Reports 336 (1-2) 91:196601, 2003. (2000) 1 -- 166. 6 D. A. Bagrets and Yu. V. Nazarov, Phys. Rev. B, 2 H. Birk, M. J. M. de Jong, and C. Schoneberger, Phys. 67(8):085316, 2003. Rev. Lett., 75(8):1610, 1995 3 A. Nauen, I. Hapke-Wurst, F. Hohls, U. Zeitler, R. J. Haug, and K. Pierz, Phys. Rev. B, 66(16):161303, 2002. 4 A. Nauen, F. Hohls, N. Maire, K. Pierz, and R. J. Haug, Phys. Rev. B, 70(3):033305, 2004. 7 L. S. Levitov, H. Lee, and G. B. Lesovik. J. Math. Phys., 37(10):4845, 1996. 8 S. Gustavsson, R. Leturcq, B. Simovic, R. Schleser, T. Ihn, P. Studerus, K. Ensslin, D. C. Driscoll, and A. C. Gossard. Phys. Rev. Lett., 96(7):076605, 2006. 9 T. Fujisawa, T. Hayashi, R. Tomita, and Y. Hirayama, 5 B. Reulet, J. Senzier, and D. E. Prober, Phys. Rev. Lett. Science, 312:1634, 2006. ·10 3-10-50510432101.00.80.60.41.00.80.60.4432101.41.21.00.80.60.40.2010-105-50ΓS·t0t0 (ms)ac15 / c1c15 / c1·10 34 10 C. Fricke, F. Hohls, W. Wegscheider, and R. J. Haug, Phys. Man'ko. Physics Letters A, 193(3):209, 1994. Rev. B, 76(15):155307, 2007. 11 S. Gustavsson, R. Leturcq, T. Ihn, K. Ensslin, M. Rein- wald, and W. Wegscheider, Phys. Rev. B, 75(7):075314, 2007. 12 C. Flindt, C. Fricke, F. Hohls, T. Novotn´y, K. Netocn´y, T. Brandes, and R. J. Haug, Proc. Natl. Acad. Sci. U.S.A., 106(25):10116, 2009. 13 C. Fricke, F. Hohls, C. Flindt, and R. J. Haug, Physica E, 42(4):848, 2010. 14 I. M. Dremin, and R. C. Hwa. Phys. Rev. D, 49(11):5805, 1994. 15 V.V. Dodonov, I.M. Dremin, P.G. Polynkin, and V.I. 16 C. Flindt, T. Novotn´y, A. Braggio, M. Sassetti, and A. Jauho. Phys. Rev. Lett., 100:150601, 2008. 17 R. Held, T. Vancura, T. Heinzel, K. Ensslin, M. Holland, and W. Wegscheider. Appl. Phys. Lett., 73(2):262, 1998. 18 U. F. Keyser, H. W. Schumacher, U. Zeitler, R. J. Haug, and K. Eberl. Appl. Phys. Lett., 76(4):457, 2000. 19 V. S. Khrapai, S. Ludwig, J. P. Kotthaus, H. P. Tranitz, and W. Wegscheider. Phys. Rev. Lett., 97(17):176803, 2006. 20 S. Gustavsson, M. Studer, R. Leturcq, T. Ihn, K. En- sslin, D. C. Driscoll, and A. C. Gossard. Phys. Rev. Lett., 99(20):206804, 2007. 

1709.01682

1

1709

Adaptively time stepping the stochastic Landau-Lifshitz-Gilbert equation at nonzero temperature: implementation and validation in MuMax3

Thermal fluctuations play an increasingly important role in micromagnetic research relevant for various biomedical and other technological applications. Until now, it was deemed necessary to use a time stepping algorithm with a fixed time step in order to perform micromagnetic simulations at nonzero temperatures. However, Berkov and Gorn have shown that the drift term which generally appears when solving stochastic differential equations can only influence the length of the magnetization. This quantity is however fixed in the case of the stochastic Landau-Lifshitz-Gilbert equation. In this paper, we exploit this fact to straightforwardly extend existing high order solvers with an adaptive time stepping algorithm. We implemented the presented methods in the freely available GPU-accelerated micromagnetic software package MuMax3 and used it to extensively validate the presented methods. Next to the advantage of having control over the error tolerance, we report a twenty fold speedup without a loss of accuracy, when using the presented methods as compared to the hereto best practice of using Heun's solver with a small fixed time step.

cond-mat.mes-hall

cond-mat

Adaptively time stepping the stochastic Landau-Lifshitz-Gilbert equation at nonzero temperature: implementation and validation in MuMax3 J. Leliaert,1, a) J. Mulkers,1, 2 J. De Clercq,1 A. Coene,3 M. Dvornik,4 and B. Van Waeyenberge1 1)Department of Solid State Sciences, Ghent University, 9000 Gent, Belgium 2)Department of Physics, Antwerp University, 2020 Antwerp, Belgium 3)Department of Electrical Energy, Metals, Mechanical Constructions and Systems , Ghent University, 9052 Zwijnaarde, Belgium 4)Department of Physics, University of Gothenburg, 412 96, Gothenburg, Sweden (Dated: 7 September 2017) Thermal fluctuations play an increasingly important role in micromagnetic research relevant for various biomedical and other technological applications. Until now, it was deemed necessary to use a time stepping algorithm with a fixed time step in order to perform micromagnetic simulations at nonzero temperatures. However, Berkov and Gorn1 have shown that the drift term which generally appears when solving stochastic differential equations can only influence the length of the magnetization. This quantity is however fixed in the case of the stochastic Landau-Lifshitz-Gilbert equation. In this paper, we exploit this fact to straightforwardly extend existing high order solvers with an adaptive time stepping algorithm. We implemented the presented methods in the freely available GPU-accelerated micromagnetic software package MuMax3 and used it to extensively validate the presented methods. Next to the advantage of having control over the error tolerance, we report a twenty fold speedup without a loss of accuracy, when using the presented methods as compared to the hereto best practice of using Heun's solver with a small fixed time step. I. INTRODUCTION Micromagnetic simulations of systems at nonzero temperatures are an increasingly important tool to numerically investigate magnetic systems relevant for technological applications. Historically, the foundations for a description of thermal fluctuations in micromag- netic systems were laid by Brown when he investigated the thermal switching of single-domain particles2,3. Today, these particles are used in promising biomed- ical applications such as disease detection and tumor treatment4–7. In order for these applications to be successful, a full understanding of the particles' thermal switching is important. For example, many characteriza- tion procedures8–14 require this knowledge to accurately determine the particles' properties. Also diagnostic par- ticle imaging15–23, and therapeutic applications24,25 rely on models of the particles' thermal switching. Currently, these models are often based on approximations that not always take into account that, e.g. the magnetization state in large particles can deviate from a uniform magnetization26, or the particles might interact with each other via the magnetostatic interaction27. In such cases, the analytical models do not accurately reflect the true magnetization dynamics of the particles, and one has to rely on numerical models, the most accurate of which are based on a micromagnetic approach28–34. Next to their relevance in magnetic nanoparticle re- search, thermal fluctuations also play an important role in (exchange-coupled) continuous magnetic systems. One a)Electronic mail: jonathan.leliaert@ugent.be technologically relevant example is domain wall motion through a magnetic nanostrip, proposed as the oper- ating principle for the racetrack memory35–37 and for logic devices38–42. Recently, even smaller magnetiza- tion structures, i.e. skyrmions, have been proposed in both memory43 and logic devices44. As the informa- tion carriers in these devices become smaller, the in- fluence of thermal fluctuations further grows in impor- tance: at such small spatial scales, the thermal stability of the bits themselves starts to become a relevant research question45. At the same time, their thermal depinning becomes an inherently stochastic process46. When nu- merically investigating domain wall motion at low driv- ing forces, the dynamics can only be captured by con- sidering the interplay between thermal fluctuations, the disorder energy landscape of the material and the driv- ing forces. The resulting motion is then called domain wall creep47. Until now, full micromagnetic simulations are still prohibitively expensive in all but the smallest of such systems48. Thermal fluctuations are also critical to the design of magnetic storage elements. They do not only determine the data retention limit of any magnetic storage sys- tem, but can also influence the read and write process of a MRAM cell. So estimation of read and write errors requires stochastic micromagnetic modeling of the spin valve during the application of the spin-torque current. This very challenging because of large time scales that are involved49. There exist different theoretical approaches, each with their respective advantages and disadvantages, to study thermally induced magnetization dynamics32,50. Follow- ing Brown3, it is possible to derive the Fokker-Planck equation describing the time-dependent probability dis- tribution of the magnetization directions of an ensem- 7 1 0 2 p e S 6 ] l l a h - s e m . t a m - d n o c [ 1 v 2 8 6 1 0 . 9 0 7 1 : v i X r a ble of uniformly magnetized magnetic nanoparticles3,32. However, only in the simplest cases, e.g. when the parti- cles' anisotropy axes are aligned with the applied field, an analytical solution can be found. In more complex cases approximations have to be introduced and when consid- ering continuous systems consisting of several exchange- coupled finite difference cells, this approach becomes in- tractable. Alternatively, thermal fluctuations can be included as a stochastic term in the Landau-Lifshitz-Gilbert (LLG) equation, henceforth named stochastic LLG (sLLG) equation by adding a thermal field term to the effective field. This approach was presented by Lyberatos51 and is based on the fact that finite difference cells can be con- sidered as dipoles comparable to single-domain particles. This method has the advantage that it is a straightfor- ward extension of the LLG equation. On the other hand, in contrast to the Fokker-Planck approach, the resulting sLLG equation has to be solved many times in order to gather enough data to draw conclusions about averaged quantities. Integrating stochastic differential equations (SDE) re- quires the use of non-Riemann calculus to be able to deal with the discontinuous thermal field term. A full discus- sion of this topic lies beyond the scope of this article, and for an excellent introduction we refer to Ref.32. In gen- eral, SDE's are not trivial to numerically integrate, and require specialized methods52, which are only suited to integrate SDE's written in either their Ito or Stratonovich form, as otherwise a drift term might appear in the so- lution. When considering variable time stepping algo- rithms, the complexity further increases53,54, sometimes even canceling the relative advantage obtained by using such methods. However, Berkov and Gorn have shown that the drift term in the sLLG equation manifests itself only in the length of the magnetization vector1 which, in contrast to the Landau-Lifshitz-Bloch equation55, is held constant. As shown in Ref.1, this can be seen more clearly when writing the sLLG equation in spherical coordinates. In Cartesian coordinates, numerical noise would build up in this direction if it weren't accounted for by renormal- izing the magnetization after each time step to ensure that the drift term does not influence the magnetization dynamics. Consequently, the Ito and Stratonovich inter- pretation are equivalent for integrating the sLLG equa- tion, enabling the use of higher order solvers to integrate the sLLG eqation56. Despite this result, in literature one often still finds the recommendation to use the second order Heun's solver (here denoted with "RK12", because it is a second order Runge-Kutta type solver with em- bedded first order solution) with a very small fixed time step of the order of femtoseconds to simulate micromag- netic systems at finite temperatures57–59. In this paper, we present the use of higher order solvers with adaptive time stepping for such simulations and show that this method offers significant advantages. 2 II. METHODS The Landau-Lifshitz-Gilbert (LLG) equation60,61 con- tains a precession and a damping term, and describes the magnetization dynamics at the nanometer length scale and the picosecond timescale. m = − γ 1 + α2 (m × Beff + αm × (m × Beff )) (1) In this equation, γ denotes the gyromagnetic ratio, α the dimensionless Gilbert damping parameter and Beff the ef- fective field. There is more than one way to add thermal fluctua- tions to the LLG equation. We choose to add a stochastic thermal field Btherm as a contribution to the effective field term in both the precession and damping term, although it has been shown that the field contribution could also be omitted in the damping term if the size of the thermal field is rescaled adequately58. A second common option32 is to add thermal fluctuation directly as an extra thermal torque term to the LLG equation. The properties of the thermal field Btherm were de- termined by Brown when he investigated the thermal switching of single-domain particles3. Later, it was re- alized that this theory was also applicable to micromag- netic simulations as each finite difference cell can be con- sidered as such a particle51. The thermal field is given by (2) (3) (cid:104)Btherm(cid:105) = 0 (cid:104)Btherm,i(t)Btherm,j(t(cid:48))(cid:105) = qδ(t − t(cid:48))δij 2kBT α MsγV q = (4) Here, the operator (cid:104)·(cid:105) denotes a time average, (cid:104)··(cid:105) a cor- relation, δ the Dirac delta function and the indices i and j run over the x, y and z axes in a Cartesian coordinate system. The thermal field has zero average [Eq. (2)], is uncorrelated in time and space [Eq. (3)] and its size q is given by Eq. (4). In this equation, kB denotes the Boltz- mann constant, T the temperature, Ms the saturation magnetization, and V the volume on which the thermal fluctuations act, i.e. the volume of a single finite differ- ence cell. Equations (2) to (4) are determined such that the effect of the thermal fluctuations is independent of the spatial discretization used: when splitting up a volume into sub- volumes and averaging the thermal fluctuations within those, one will recover the same resulting dynamics as in the undivided volume. The same is also true for the time step ∆t: when averaged out over a larger time, thermal fluctuations decrease in strength and again, this propor- tionality is determined such that the average dynamics do not depend on the time discretization. A. Implementation in MuMax3 MuMax3 is a GPU-accelerated micromagnetic software package which numerically solves the LLG equation using a finite-difference discretization59. The thermal field is included in the effective field as (cid:115) Btherm = η 2αkBT MsγV ∆t (5) where ∆t denotes the time step and η is a random vector drawn from a standard normal distribution whose value is redetermined after every time step57,59. MuMax3 provides several explicit Runge-Kutta meth- ods to time step the LLG equation, the details of which can be found in Ref.59. Here, we will only mention the ones relevant for this work. Previously, simulations at nonzero temperatures in MuMax3 were performed with the widely used Heun's method (RK12), using a very small time step of the order of 5 fs. In Section III, we will validate our results by comparing them to the solution obtained with this solver when there are no analytical solutions available. For dynamical simulations at zero temperature, the default solver is the Dormand-Prince method (RK45). This solver offers 5th order error convergence and con- tains an embedded 4th order method to estimate the error. Generally speaking (i.e. when not investigating very fast dynamics, or in the absence of thermal fluctua- tions), it is not advantageous to implement even higher- order solvers. The reason for this is threefold: 1) For the moderately small torques encountered in typical mi- cromagnetic simulations, the performance of the solver is limited by its stability regime. This means that using even slightly larger time steps will result in much larger errors no matter how small the exerted torques are. 2) Higher order methods typically need more intermediate torque evaluations per time step, thus reducing the ad- vantage obtained by taking a larger time step. 3) Due to the memory required to store the results of the interme- diate torque evaluations, higher-order solvers dispropor- tionately increase the memory consumption compared to the obtained gain in performance. 0 1 6 4 15 2 3 4 1 6 4 75 5 16 75 5 2 6 − 8 25 −4 3 144 −18 5 0 407 11 640 93 640 − 18 5 361 320 5 − 8 1 0 − 11 1 16 25 128 − 11 256 − 11 256 − 11 803 80 160 5 0 0 0 1125 2816 1125 2816 0 160 9 32 9 32 0 31 384 7 1408 − 5 66 3 1 0 0 5 5 66 66 5 5 66 66 55 128 11 256 99 256 125 768 125 768 0 0 5 66 0 0 − 5 66 TABLE I. Butcher tableau62 of the Runge-Kutta-Fehlberg (RKF56) solver with sixth order solution and embedded 5th order solution. The difference between both, used as error estimate, is given in the last row. However, as the stochastic thermal field is not constant in between time steps, the torque continuity requirement is not longer fulfilled, and the first and last torque eval- uations have to be performed separately. Because the RKF56 solver never has the FSAL property, its perfor- mance can only compete with these other methods at nonzero temperatures, where the other methods do not benefit from the FSAL property either. The implementation of the Sixth order Runge-Kutta- Fehlberg solver is subject to the same tests used for the other solvers implemented in MuMax359, and a full re- port is beyond the scope of this article. Nonetheless, Fig. 1 shows that our solution to standard problem 4, proposed by the µMag modeling group63, agrees with the solution obtained with OOMMF64 (not using the RK56 solver). One of the most sensitive checks one can perform to test the implementation of a solver is to investigate the scaling of its error convergence. Figure 2 shows the error
after a single precession without damping of a single spin in a field of 0.1 T as function of the time step ∆t. The solver shows the expected sixth order convergence up to the limit of the single precision implementation59 (
≈ 10−7). B. Sixth order Runge-Kutta-Fehlberg solver Due to the large size of the stochastic thermal field, simulations at nonzero temperatures require much smaller time steps, so that the solver performance is not longer limited by its stability regime, and large enough time steps can be taken to justify the additional interme- diate torque evaluations. Therefore, we also implemented the 6th order Runge-Kutta-Fehlberg (RKF56) method with 5th order embedded solution shown in Table I. Unlike the RKF56 solver, some of the solvers used in MuMax3 like the RK45 method, benefit from the first- same-as-last (FSAL) property. In these solvers, the last torque evaluation of the current time step corresponds to the first evaluation of the next time step, thus effec- tively reducing the number of evaluations per step by 1. C. Time stepping with adaptive time steps When performing simulations at nonzero tempera- √ tures, it is important to note that the size of the thermal field is determined by 1/ ∆t. This implies that, when a large thermal field is generated leading to a bad step (defined as a step where the torque was too large for the used time step), the step will be undone, and the adaptive time step algorithm will decrease the time step, thus fur- ther increasing the size of the field. Luckily, this 1/ ∆t dependency makes the time step smaller at a slower rate than that the error is reduced (∆tN with N the order of the solver65). However, it is important to use higher order solvers like the RK45 or RKF56 method in order to maintain a large time step. √ 4 III. VALIDATION The adaptive time stepping at nonzero temperatures will be tested in several test cases, focusing on different aspects, i.e. static vs. dynamic properties of uncoupled spins or continuous magnets. Each time, the simulation results obtained with adaptive time stepping will be com- pared either to analytical solutions or to the solutions obtained with the RK12 method with fixed time step. A. Spectra of a single spin It will be verified whether the thermal field and the re- sulting magnetization dynamics of a single spin in the ab- sence of an external field shows the theoretically expected behavior. The thermal field should display a white spec- trum SH construction, as its size is given by Gaussian random numbers. Figure 3 proves that this is indeed the case. FIG. 3. The thermal field spectra of a single spin (rescaled with an arbitrary factor for clarity reasons) obtained with the RK45 solver with fixed time steps, and with the RK45 and RKF56 solver with adaptive time steps. All spectra display white noise. The random thermal fields acting on a single isotropic finite-difference cell gives rise to a random walk on the unit sphere11,32. The shape of the spectral density SM(f ) of such a random walk is described by the square root of a Lorentzian66, (cid:115)(cid:18) (cid:19) SM(f ) ∼ f0/2 f 2 0 + (πf )2 , (6) i.e. white noise with a 1/f cutoff at a cutoff frequency f0 given by33,58 f0 = α (1 + α2) γkBT MsV (7) Figure 4 shows the obtained magnetization spectra (gray dots) indeed coincide with the red lines determined by FIG. 1. The solutions to standard problem 4, obtained by MuMax3 (gray points) and OOMMF (red lines). This prob- lem focuses on micromagnetic dynamics and looks at the time evolution of the magnetization during the relaxation of a mag- netic rectangle from an initial s-state. The problem is run for two different applied fields (top and bottom graph) and the space dependent magnetizations when (cid:104)mx(cid:105) crosses zero are shown below, in the left and right plot, respectively. FIG. 2. The error
as function the time step ∆t of the RKF56 solver (gray squares) displays the sixth order convergence (red line). To give the correct solution, the statistical properties of the random numbers η in Eq.(5) should correspond to the ones determined in Eqs. (2) to (4). However, if one would redraw these random numbers after a bad step, the small thermal fields (small η) would be applied during longer time steps and large thermal fields (large η) during shorter time steps, thus virtually changing the distribution of the random numbers, and eventually giv- ing rise to incorrect solutions. In our implementation we avoid this by keeping the previously drawn random numbers and rescaling the thermal field with a factor (cid:112)∆tnew/∆tbadstep in case a bad step is encountered to ensure that the correct statistical properties of the ther- mal field are maintained. -1-0.500.5100.20.40.60.81<m>t (ns)mxmymz-1-0.500.51<m>mxmymz10-810-710-610-510-410-310-210-11e-111e-10ε ()Δt (s) 1 10 100 0.1 1 10 100adaptive RKF56adaptive RK45fixed RK45SH (T/√ Hz )f (GHz)Eqs. (6) and (7). Because all spectra coincide, they were rescaled with an arbitrary factor for clarity. 5 FIG. 4. The magnetization spectra of a single spin (rescaled with an arbitrary factor for clarity reasons) obtained with the RK45 solver with fixed time steps, and with the RK45 and RKF56 solver with adaptive time steps. All spectra display the theoretically expected shape depicted by the red lines. B. Equilibrium magnetization This validation problem checks whether the magneti- zation of an ensemble of uncoupled spins in thermal equi- librium in an externally applied field is described by the Langevin function L(ξ), L(ξ) = coth(ξ) − 1 ξ where the argument ξ stands for ξ = µ0MsV Hext kBT . (8) (9) Figure 5 proves that this is indeed the case for 4 different ξ (realized by 2 different cell sizes and 2 different temper- atures) simulated with the RK45 and RKF56 solver with adaptive time steps over a large range of applied fields. C. Thermal switching After the equilibrium magnetization addressed in the previous problem, we will now concern ourselves with a dynamical problem consisting of the thermal switch- ing rate of a single (macro-)spin particle with uniaxial anisotropy. In the limit of a high energy barrier (com- pared to the thermal energy), the switching rate ν is given by67 (cid:115) ν = γ α 1 + α2 8K 3V 2πM 2 s kB e−KV /kBT . (10) FIG. 5. The average magnetization (cid:104)m(cid:105) of an ensemble of 218 uncoupled spins in thermal equilibrium at different tempera- tures and different cell sizes (reflected in the different values of ξ) for the RK45 (gray circles) and RKF56 (gray crosses) with adaptive time steps. The results agree perfectly with the Langevin function [Eq. (8)], shown by a red line. For an ensemble of uncoupled spins, initialized with all spins pointing in the same direction, this switching gives rise to an exponentially decaying magnetization, with a decay constant 1/2ν. In this test problem, we simulate this decay to determine the numerical switching rate, and compare these values to their theoretical prediction. Figure 6 shows Arrhenius plots for the temperature- dependent switching rate ν of uncoupled finite difference cells with volume V =(10 nm)3 and uniaxial anisotropy constant K=1×104 or 2×104 J/m3. Again, a quantita- tive agreement is seen between the MuMax3 simulations and the theoretically predicted behavior. FIG. 6. Arrhenius plot of the thermal switching rate of a 10 nm large cubic cell, with Ms=1 MA/m, α=0.1,K=10 or 20 kJ/m3. Simulations were performed using the RK12 solver with fixed time steps (∆t=5 fs) or the RK45 or RKF56 solver with adaptive time steps on an ensemble of 218 non- interacting cells for 1 µs or until the ensemble magnetization crossed 0. All results agree with the red solid lines depicting the analytically expected switching rates (Eq. 10). 0.01 0.1 1 10 100 0.1 1 10 100adaptive RKF56adaptive RK45fixed RK45SM (A/m√ Hz )f (GHz) 0 0.2 0.4 0.6 0.8 1 0 0.02 0.04 0.06 0.08 0.1ξ=91ξ=242ξ=725ξ=30<m> ()B (T)LangevinRK45RKF56100101102103104105106107108 0 0.005 0.01 0.015 0.02 0.025 0.03 0.035K=10 kJ/m3K=20 kJ/m3ν (/s)1/T (1/K)RK12RK45RKF56D. Thermally excited magnetization spectrum In this problem we look at the thermally excited mag- netization spectrum of a 10 nm thick disk with a diame- ter of 512 nm, Ms=1 MA/m, exchange constant Aex=10 pJ/m, and α = 1, discretized in cells measuring 4 by 4 by 10 nm3. The equilibrium magnetization structure in such a disk is a vortex structure, as depicted in the inset of Fig. 7. We apply a thermal field corresponding to 300 K, thereby thermally exciting the sample, resulting in the spectra shown in Fig. 7. The spectrum depicted in red was obtained with the RK12 solver with fixed time step (∆t=5 fs), and serves as a reference solution. The gray lines, which agree almost perfectly with the benchmark solution, correspond to the three spectra obtained with the RK45 solver with the time step fixed to ∆t=300 fs and with the RK45 and RKF56s solvers with adaptive time step and
= 10−5. Note that the spectra overlap even at high frequencies, indicating that the adaptive time stepping does not lead to spectral leakages. FIG. 7. The red line corresponds to the spectrum obtained with the Heun solver with ∆t=5 fs. The three gray lines, which overlap almost perfectly with the red one, correspond to the RK45 solver with fixed time step ∆t=300 fs, and with the solution obtained with the RK45 and RKF56 solver with adaptive time stepping with
= 10−5. All spectra were av- eraged out over 25 realizations with a different random seed for the thermal field. The inset shows the equilibrium vortex magnetization structure in the system under considerationa a The minimum in the error is a result between a direct match between the used time step and the gyration period used for the evaluation of the precision59. the material parameters of permalloy: Ms = 860 kA/m, Aex = 13 pJ/m, α=0.01 and simulate the domain wall in the center of a moving window, as shown in Fig. 8. 6 FIG. 8. The transverse domain wall in the center of the nanowire. By compensating the edge charges, we simulate an infinitely long wire. The thermally driven motion can be described by a random walk characterized by a diffusion constant D which scales linearly with temperature. Similarly as in Ref.68 we simulate a transverse domain wall for 100 ns and repeat this simulation with a large number of dif- ferent random seeds. In Tab. II, we compare the results obtained from the full micromagnetic simulations with the diffusion constant D of approximately 310 nm2/ns predicted by the model introduced, and numerically val- idated in Ref.68using the RK12 solver with fixed time step. The results show that the standard errors s are larger than the difference between the obtained diffusion constants and the expected value. This also indicates that, for this problem, an error tolerance
= 10−3 suf- fices for all practical purposes, as the variance between different simulations gives rise to an uncertainty that is larger than the errors due to the use of this relatively large
. As this is problem dependent, the default value in MuMax3 remains 10−5, but we suggest that, depending on the system under consideration, larger values might be suitable in simulations at nonzero temperatures. TABLE II. The diffusion constant D and standard error s, determined from simulations performed using several solvers, with adaptive time stepping with error tolerance
and re- peated for a total number of N realizations at 300 K. The theoretically predicted D approximately equals 310 nm2/ns.
() Solver N () 5 × 10−3 500 RK23 5 × 10−3 500 RK45 1 × 10−3 1000 RK45 RKF56 5 × 10−3 200 RKF56 1 × 10−3 200 RKF56 1 × 10−4 200 D s(D) (nm2/ns) (nm2/ns) 315 309 316 340 335 315 19 19 14 37 32 31 IV. PERFORMANCE E. Thermal diffusion of a domain wall In a last validation problem we investigate the ther- mal driven diffusion of transverse domain walls in a non-disordered permalloy nanowire68. We simulate a nanowire with cross-sectional dimensions of 100×10 nm2 discretized in cells of 3.125 × 3.125 × 10nm3. We use To assess the performance of the presented methods we will consider the problems detailed in III D and III E, i.e the thermal spectrum of a disk and thermally driven domain wall diffusion, respectively, as benchmarks. The benchmark results are shown in Fig. 9 a) and b). For each of these problems, the simulation time was first de- termined when solving the problem with the RK12 solver 0.010.1110100110100S (A/m√Hz )f (GHz)with a fixed time step of 5 fs. As this is a second order solver with embedded first order solver, the difference between both solutions serves as an error estimate, al- lowing us to estimate with which error tolerance
in the adaptive time stepping the results should be compared. This data point is indicated by a black cross in the fig- ure. Next, the problems were solved using adaptive time stepping, once with the RK45, and once with the RKF56 solver, with
ranging from 10−3 to 10−7, shown in gray and red, respectively. The simulation runtime [Figs. 9 a) and b)] show the time it took to simulate 10 ns of magnetization dynamics, while Figs. 9 c) and d) indicate the average time step ∆t used by the solver, for each
. When comparing the results from the adaptive time step- ping methods with the result of the RK12 solver at the same estimated
, one sees that the adaptive time step- ping methods use a considerably larger time step without a loss of accuracy. For these two benchmark problems, this results in a twenty fold speedup of the simulation. We also investigated the performance of the system de- scribed in Section III C, i.e. thermal switching of uncou- pled spins. There, an even higher speedup was achieved, but this was attributed to the fact that such systems do not require the calculation of the demagnetizing field so that other factors, like the generation of the random numbers for the thermal field, become the limiting fac- tor. Because the random numbers have to be generated only once per time step independently of the used solver, a very large performance gain can be achieved by us- ing higher order solvers which allow time steps that are over a 1000 times larger than the ones necessary for the RK12 method with the same accuracy. However, as this highly depends on the used hardware, the performance gained by using the adaptive time stepping methods can lie anywhere between a factor 20 to 10 000, depending on accuracy. These observations indicate that the presented methods are particularly suited for magnetic nanopar- ticle research, where micromagnetic simulations are be- coming increasingly important34. The relative performance of the RK45 and RKF56 solver is comparable and show the expected trends that the RK45 solver is faster at higher
and simulated tem- peratures while the RKF56 solver is faster at low
and temperatures. Generally, only when simulating systems at very high temperatures, or with very small
, it does pay off to use the RKF56 solver. V. CONCLUSIONS In this paper, we have exploited the fact that the drift term in the stochastic Landau-Lifshitz-Gilbert equation is only able to manifests itself in the direction of the mag- netization length, which is fixed. Therefore, we were able to straightforwardly extend existing high order solvers with adaptive time stepping at nonzero temperatures. In an effort to further increase the performance, we have im- plemented the sixth order Runge-Kutta-Fehlberg solver, 7 FIG. 9. Benchmark results for the systems described in Sec- tions III D [panel a) and c)] and III E [b) and d)], respectively. The top row shows the runtime required to simulate 10 ns of magnetization dynamics, while the bottom row shows the average time step used. The black crosses indicate the perfor- mance of the fixed time step RK12 method with
estimated from the difference between the second order and embedded first order solution. All results were obtained an NVIDIA Titan Xp GPU in a system running on a 7th generation i5 CPU. and we extensively validated both the correctness of this newly implemented solver and the adaptive time step- ping method used at nonzero temperature. All presented methods are included in the open-source micromagnetic software package MuMax3 and are thus freely available online. The main advantages of the presented adaptive time stepping methods at nonzero temperatures are that they offer an inherent error control, which is unavailable with fixed time stepping methods, and without a loss of accu- racy one can obtain a twenty fold speedup compared to the commonly best practice of using the RK12 solver with small fixed time step. This enables simulations which previously took too long to be considered feasible and will be useful for micromagnetic research of continuous (exchange coupled) systems like spin valves, or domain wall motion in nanowires, and for uncoupled spins, e.g. in magnetic nanoparticle research. VI. ACKNOWLEDGEMENT This work was supported by the Fonds Wetenschap- pelijk Onderzoek (FWO-Vlaanderen) through Project No. G098917N and a postdoctoral fellowship (A.C.). J. L. is supported by the Ghent University Special Research Fund (BOF postdoctoral fellowship). We gratefully ac- knowledge the support of NVIDIA Corporation with the donation of the Titan Xp GPU used for this research. 1D. Berkov and N. Gorn, Journal of Physics: Condensed Matter 14, L281 (2002). 2W. F. Brown, J. Appl. Phys. 30, S130 (1959). 3W. F. Brown, Phys. Rev. 130, 1677 (1963). 101102103runtime (s)RK12RK45RKF5610-1510-1410-1310-1210-710-610-510-410-3Δt (s)ε()10-710-610-510-410-3ε()a)c)b)d)8 4Q. A. Pankhurst, J. Connolly, S. K. Jones, and J. Dobson, J. Phys. D: Appl. Phys. 36, R167 (2003). 5Q. Pankhurst, N. Thanh, S. Jones, and J. Dobson, J. Phys. D: Appl. Phys. 42, 224001 (2009). 6Y. Gao, Y. Liu, and C. Xu, in Engineering in Translational Medicine (Springer, 2014) pp. 567–583. 7W. Wu, Z. Wu, T. Yu, C. Jiang, and W.-S. Kim, Sci- ence and Technology of Advanced Materials 16, 023501 (2015), http://dx.doi.org/10.1088/1468-6996/16/2/023501. 8S. Bogren, A. Fornara, F. Ludwig, M. del Puerto Morales, U. Steinhoff, M. F. Hansen, O. Kazakova, and C. Johansson, Int. J. Mol. Sci. 16, 20308 (2015). 9F. Ludwig, C. Balceris, T. Viereck, O. Posth, U. Steinhoff, H. Gavilan, R. Costo, L. Zeng, E. Olsson, C. Jonasson, and C. Johansson, Journal of Magnetism and Magnetic Materials 427, 19 (2017). 10F. Ludwig, C. Balceris, C. Jonasson, and C. Johansson, IEEE Transactions on Magnetics (2017). 11J. Leliaert, A. Coene, M. Liebl, D. Eberbeck, U. Steinhoff, F. Wiekhorst, B. Fischer, L. Dupr´e, and B. Van Waeyenberge, Appl. Phys. Lett. 107, 222401 (2015). 12J. Leliaert, D. Eberbeck, M. Liebl, A. Coene, U. Steinhoff, F. Wiekhorst, B. V. Waeyenberge, and L. Dupr´e, Journal of Physics D: Applied Physics 50, 085004 (2017). 13J. Leliaert, D. Schmidt, O. Posth, M. Liebl, D. Eberbeck, A. Co- and ene, U. Steinhoff, F. Wiekhorst, B. V. Waeyenberge, L. Dupr´e, 50 (2017). 14M. Liebl, F. Wiekhorst, D. Eberbeck, P. Radon, D. Gutkelch, D. Baumgarten, U. Steinhoff, and L. Trahms, Biomedical Engi- neering/Biomedizinische Technik 60, 427 (2015). 15F. Wiekhorst, U. Steinhoff, D. Eberbeck, and L. Trahms, Pharm. Res. 29, 1189 (2012). 16D. Baumgarten, F. Braune, E. Supriyanto, and J. Haueisen, Journal of Magnetism and Magnetic Materials 380, 255 (2015). 17A. Coene, G. Crevecoeur, and L. Dupr´e, IEEE Trans. Magn. 48, 2842 (2012). 34D. Cabrera, A. Lak, T. Yoshida, M. Materia, D. Ortega, F. Lud- and F. Teran, wig, P. Guardia, A. Sathya, T. Pellegrino, Nanoscale 9, 5094 (2017). 35S. Parkin, M. Hayashi, and L. Thomas, Science 320, 190 (2008). 36M. Hayashi, L. Thomas, R. Moriya, C. Rettner, and S. S. P. Parkin, Science 320, 209 (2008). 37S. Parkin and S.-H. Yang, Nat. Nanotechnol. 10, 195 (2015). 38D. A. Allwood, G. Xiong, C. C. Faulkner, D. Atkinson, D. Petit, and R. P. Cowburn, Science 309, 16881692 (2005), http://www.sciencemag.org/cgi/reprint/309/5741/1688.pdf. 39D. Atkinson, C. C. Faulkner, D. A. Allwood, and R. Cowburn, "Spin dynamics in confined magnetic structures iii," (Springer, Berlin–Heidelberg, 2006) Chap. Domain-Wall Dynamicsin Mag- netic Logic Devices, pp. 207–223. 40S. E. Barnes, J. Ieda, and S. Maekawa, Appl. Phys. Lett. 89, 122507 (2006). 41J. Vandermeulen, B. Van de Wiele, L. Dupr´e, and B. Van Waeyenberge, J. Phys. D: Appl. Phys. 48, 275003 (2015). 42K. A. Omari and T. J. Hayward, Phys. Rev. Appl. 2, 044001 (2014). 43A. Fert, V. Cros, and J. Sampaio, Nat. Nanotechnol. 8, 152 (2013). 44X. Zhang, M. Ezawa, and Y. Zhou, Sci. Rep. 5 (2015). 45D. Cort´es-Ortuno, W. Wang, M. Beg, R. A. Pepper, M.-A. Bisotti, R. Carey, M. Vousden, T. Kluyver, O. Hovorka, and H. Fangohr, Scientific Reports 7 (2017). 46C. Wuth, P. Lendecke, and G. Meier, J. Phys.: Condens. Matter 24, 024207 (2012). 47P. J. Metaxas, J. P. Jamet, A. Mougin, M. Cormier, J. Ferr´e, V. Baltz, B. Rodmacq, B. Dieny, and R. L. Stamps, Phys. Rev. Lett. 99, 217208 (2007). 48J. Leliaert, B. Van de Wiele, A. Vansteenkiste, L. Laurson, and B. Van Waeyenberge, Sci. Rep. 6, G. Durin, L. Dupr´e, 20472 (2016). 49U. Roy, T. Pramanik, L. F. Register, and S. K. Banerjee, IEEE Transactions on Magnetics 52, 1 (2016). 18A. Coene, J. Leliaert, L. Dupr´e, and G. Crevecoeur, Med. Phys. 50W. T. Coffey and Y. P. Kalmykov, J. Appl. Phys. 112, 121301 42, 6853 (2015). 19A. Coene, J. Leliaert, M. Liebl, N. Loewa, U. Steinhoff, G. Creve- coeur, L. Dupr´e, and F. Wiekhorst, Physics in Medicine and Biology 62, 3139 (2017). 20B. W. Ficko, P. Giacometti, and S. G. Diamond, Biomedical Engineering/Biomedizinische Technik 60, 457 (2015). 21B. W. Ficko, C. NDong, P. Giacometti, K. E. Griswold, and S. G. Diamond, IEEE Transactions on Biomedical Engineering 64, 972 (2017). 22K. Them, Physics in Medicine and Biology (2017). 23B. Gleich and R. Weizenecker, Nature 435, 1214 (2005). 24R. Hergt, S. Dutz, R. Muller, and M. Zeisberger, J. Phys.: Con- dens. Matter 18, S2919 (2006). 25E. A. Prigo, G. Hemery, O. Sandre, D. Ortega, E. Garaio, F. Plazaola, and F. J. Teran, Appl. Phys. Rev. 2, 041302 (2015), http://dx.doi.org/10.1063/1.4935688. 26S.-K. Kim, M.-W. Yoo, J. Lee, H.-Y. Lee, J.-H. Lee, Y. Gaididei, V. P. Kravchuk, and D. D. Sheka, Scientific reports 5 (2015). 27L. C. Branquinho, M. S. Carriao, A. S. Costa, N. Zufelato, M. H. and A. F. Bakuzis, Sci. Rep. 3 Sousa, R. Miotto, R. Ivkov, (2013). 28J. Leliaert, A. Coene, G. Crevecoeur, A. Vansteenkiste, D. Eber- beck, F. Wiekhorst, B. Van Waeyenberge, and L. Dupr, J. Appl. Phys. 116, 163914 (2014). 29O. Laslett, S. Ruta, J. Barker, R. Chantrell, G. Friedman, and O. Hovorka, Appl. Phys. Lett. 106, 012407 (2015). 30J. Leliaert, A. Vansteenkiste, A. Coene, L. Dupr´e, and B. Van Waeyenberge, Med. Biol. Eng. Comput. 53, 309 (2015). 31S. A. Shah, D. B. Reeves, R. M. Ferguson, J. B. Weaver, and (2012). 51A. Lyberatos, D. Berkov, and R. W. Chantrell, J. Phys.: Con- dens. Matter 5, 8911 (1993). 52P. E. Kloeden, E. Platen, and H. Schurz, Numerical solution of SDE through computer experiments (Springer Science & Business Media, 2012). 53S. Mauthner, Journal of computational and applied mathematics 100, 93 (1998). 54P. M. Burrage and K. Burrage, SIAM Journal on Scientific Com- puting 24, 848 (2003). 55U. Atxitia, D. Hinzke, and U. Nowak, Journal of Physics D: Applied Physics 50, 033003 (2016). 56D. V. Berkov and N. L. Gorn, Journal of Magnetism and Mag- netic Materials 290, 442 (2005), proceedings of the Joint Euro- pean Magnetic Symposia (JEMS' 04). 57L. Lopez-Diaz, D. Aurelio, L. Torres, E. Martinez, M. A. Hernandez-Lopez, J. Gomez, O. Alejos, M. Carpentieri, G. Finocchio, and G. Consolo, J. Phys. D: Appl. Phys. 45, 323001 (2012). 58J. L. Garc´ıa-Palacios and F. J. L´azaro, Phys. Rev. B 58, 14937 (1998). 59A. Vansteenkiste, J. Leliaert, M. Dvornik, M. Helsen, F. Garcia- Sanchez, and B. Van Waeyenberge, AIP Adv. 4, 107133 (2014). 60L. Landau and E. Lifshitz, Phys. Z. Sowjetunion 8, 101 (1935). 61T. Gilbert, IEEE Trans. Magn. 40, 3443 (2004). 62J. C. Butcher, Numerical methods for ordinary differential equa- tions (John Wiley & Sons, 2008). 63muMAG Micromagnetic Modeling Activity Group, . 64M. J. Donahue, interagency report NISTIR 6376, NIST, K. M. Krishnan, Phys. Rev. B 92, 094438 (2015). Gaithersburg, MD (1999). 32D. B. Reeves, in Magnetic Characterization Techniques for 65K. Gustafsson, Control of error and convergence in ODE solvers, Nanomaterials (Springer, 2017) pp. 121–156. 33P. Ilg, Physical Review B 95, 214427 (2017). Ph.D. thesis, University of Lund (1992). 66S. Machlup, J. Appl. Phys. 25, 341 (1954). 67L. Breth, D. Suess, C. Vogler, B. Bergmair, M. Fuger, R. Heer, and H. Brueckl, J. Appl. Phys. 112, 023903 (2012). 68J. Leliaert, B. Van de Wiele, J. Vandermeulen, A. Coene, A. Vansteenkiste, L. Laurson, G. Durin, B. Van Waeyenberge, and L. Dupr´e, Appl. Phys. Lett. 106, 202401 (2015). 9 

1103.1867

1

1103

Nonequilibrium transport in quantum impurity models: Exact path integral simulations

We simulate the nonequilibrium dynamics of two generic many-body quantum impurity models by employing the recently developed iterative influence-functional path integral method [Phys. Rev. B {\bf 82}, 205323 (2010)]. This general approach is presented here in the context of quantum transport in molecular electronic junctions. Models of particular interest include the single impurity Anderson model and the related spinless two-state Anderson dot. In both cases we study the time evolution of the dot occupation and the current characteristics at finite temperature. A comparison to mean-field results is presented, when applicable.

cond-mat.mes-hall

cond-mat

Nonequilibrium transport in quantum impurity models: Exact path integral simulations Dvira Segal Chemical Physics Theory Group, Department of Chemistry, University of Toronto, Toronto, Ontario M5S 3H6, Canada Department of Physics, Columbia University, 538 W 120th St., New York, NY 10027. Andrew J. Millis Department of Chemistry, Columbia University, 3000 Broadway, New York, NY 10027 David R. Reichman We simulate the nonequilibrium dynamics of two generic many-body quantum impurity models by employing the recently developed iterative influence-functional path integral method [Phys. Rev. B 82, 205323 (2010)]. This general approach is presented here in the context of quantum transport in molecular electronic junctions. Models of particular interest include the single impurity Anderson model and the related spinless two-state Anderson dot. In both cases we study the time evolution of the dot occupation and the current characteristics at finite temperature. A comparison to mean-field results is presented, when applicable. PACS numbers: 03.65.Yz, 05.60.Gg, 72.10.Fk, 73.63.-b I. INTRODUCTION Understanding charge and energy transport at the nanoscale is essential for the design of stable and reproducible molecu- lar electronic components such as transistors, "refrigerators", and energy conversion devices [1]. While detailed model- ing is necessary for elucidating and optimizing the transport characteristics of such devices, in this paper we embrace an alternative-minimal approach [2]. With the motivation of ex- ploring the fundamentals of quantum transport in correlated electron systems, we focus on the dynamics of "impurity models" [3], consisting a small subsystem (molecule, quan- tum dot) interacting with two electronic reservoirs, driven to a nonequilibrium steady-state by a DC voltage bias. While the impurity object includes only few degrees of freedom, it incorporates many-body interactions, making exact analytical solutions generally inaccessible. Among the standard models considered in this context are the single impurity Anderson model (SIAM), combining a single electronic level with up to two interacting electrons coupled to metallic leads [4], and the spinless two-level Anderson model (2LAM), consisting a spinless dot with two interacting (HOMO and LUMO) levels hybridized with electronic reservoirs [5 -- 7]. Even in the steady-state limit, the analysis of such nonequi- librium systems turns out to be intricate, and analytical so- lutions are lacking, see e.g., [8]. Various numerical simula- tion approaches have been developed, including perturbative treatments [9] and renormalization-group techniques [3, 10]. Even more difficult is the description of the time evolution of the system from some initial preparation towards steady-state under a finite voltage-bias. The transient nonequilibrium dy- namics of the Anderson model, and its variants, has been re- cently simulated using path-integral Monte-Carlo simulations [11 -- 13] and influence-functional methods [14, 15]. Several factors should be considered for fully understanding the dy- namics of such models: (i) the finite external bias, driving the system out-of-equilibrium, (ii) electron-electron interaction, or more generally many-body interactions, (iii) band-structure effects, and (iv) the device temperature. The combined effects of these four ingredients on the time evolution of a nanoscale object have not yet been fully understood [16]. Our objective here is to follow the dynamics of simple nanoscale junctions employing the SIAM and the 2LAM models as prototypes. We explore the role of the temper- ature and interaction strength in determining both the short time evolution of the system and its steady-state proper- ties. For achieving this task we adopt the recently developed numerically-exact influence functional path integral (INFPI) technique [15]. This method relies on the observation that in out-of-equilibrium (and finite temperature) cases bath correla- tions have a finite range, allowing for their truncation beyond a memory time dictated by the voltage-bias and the temperature. Taking advantage of this fact, an iterative-deterministic time- evolution scheme has been developed where convergence with respect to the memory length can in principle be reached. As convergence is facilitated at large bias, the method is well suited for the description of the real-time dynamics of single- molecule devices driven to a steady-state via interaction with biased leads. In this respect the INFPI approach is comple- mentary to methods applicable predominantly close to equi- librium, e.g., numerical renormalization group techniques [3]. The principles of the INFPI approach have been detailed in Ref. [15], where it has been adopted for investigating, at zero temperature, dissipation effects in the nonequilibrium spin-fermion model, and the population dynamics in a cor- related quantum dot, investigating the Anderson model. The focus of the present study are transport characteristics of cor- related nonequilibrium models, thus we introduce the INFPI approach in this context only. We demonstrate that the method can feasibility treat various impurity models. In particular, the population dynamics and the electron current in the SIAM and the 2LAM models are simulated at nonzero temperatures. The paper is organized as follows. In Sec. II we describe the INFPI method in the context of quantum transport junc- tions. The nonequilibrium dynamics of the Anderson dot is studied in Sec. III. The spinless two-level Anderson model is discussed in Sec. IV. Some conclusions follow in Sec. V. present analysis to the special form H1 = U [n1n2 − 1 2 (n1 + n2)]. 2 (2) Here ni are occupation number operators for the subsystem with U as an interaction parameter. The states '1' and '2' may either symbolize the spin orientation, or count the (subsystem) electronic states. This structure allows for the elimination of H1 via the Hubbard-Stratonovich (HS) transformation [17]. In particular, in the Anderson model [see Eq. (17)] H1 ac- counts for the double occupancy energy cost on the dot. Sim- ilarly, in the 2LAM [Eq. (27)] H1 constitutes the repulsion energy between electrons occupying the dot levels. Our objective here is to calculate the dynamics of a quadratic operator A, either given by subsystem or baths de- grees of freedom. This can be done by studying the Heisen- berg equation of motion of an exponential operator eλ A, with λ a variable that is taken to vanish at the end of the calculation, h A(t)i = Tr(ρ A) = lim λ→0 ∂ ∂λ Tr(cid:2)ρ(0)eiHteλ Ae−iHt(cid:3). (3) Here ρ is the total density matrix and the trace is performed over subsystem and reservoirs degrees of freedom. For sim- plicity, we assume that at the initial time (t = 0) the dot and the baths are decoupled, and that the baths are prepared in a nonequilibrium biased state. The time-zero total density ma- trix is therefore given by the product state ρ(0) = ρS(0) ⊗ ρL⊗ρR. We proceed and factorize the time evolution operator using a standard breakup, eiHt = (eiHδt)N , further assuming the Trotter decomposition eiHδt ≈ (cid:0)eiH0δt/2eiH1δteiH0δt/2(cid:1). The many-body term H1 can be eliminated by introducing auxiliary Ising variables s = ± via the Hubbard-Stratonovich transformation [17], e±iH1δt = 1 2 Xs e−sκ±(n2−n1). (4) Here κ± = κ′ ∓ iκ′′, κ′ = sinh−1[sin(δtU/2)]1/2, κ′′ = sin−1[sin(δtU/2)]1/2. The uniqueness of this transformation requires U δt < π. In what follows we use the following short notation, eH±(s) ≡ e−sκ±(n2−n1). (5) Incorporating the Trotter decomposition and the HS transfor- mation into Eq. (3), we find that the time evolution of A is dictated by FIG. 1: Schematic representation of the two models considered in this work: (top panel) the single impurity Anderson model with on- site repulsion terms on the dot; (bottom panel) the spinless two-level Anderson model, with two electronic levels allowing for up to two interacting electrons. II. GENERAL FORMULATION We detail here the INFPI method in the context of quan- tum transport models. A more general presentation, dealing with both transport and dissipation in nonequilibrium open systems, is given in Ref. [15]. The generic setup consid- ered includes a quantum impurity (subsystem) coupled to two metal leads (reservoirs) driven to a non-equilibrium steady- state through the application of a finite DC voltage-bias. The quantum impurity may be realized by a magnetic impurity, double quantum dots, or a multi-state quantum dot. The elec- trodes are modeled by two fermionic continua. System-bath couplings allow for a particle transfer between the impurity and the leads. We assume that the reservoirs' electrons are non-interacting, and include many-body interactions within the subsystem only, accounting for an additional energy cost for double occupancy. For a schematic representation see Fig. 1. Our generic Hamiltonian is given by H = H0 + H1, (1) where H0 includes the exactly solvable noninteracting part combining the two leads, the noninteracting part of the sub- system, and impurity-bath hybridization terms. Many body interactions are incorporated into H1, and we confine our ∂ ∂λ h A(t)i = lim λ→0 Trhρ(0)(cid:16)eiH0δt/2eiH1δteiH0δt/2(cid:17)N 1 ds± × eλ A ×(cid:16)e−iH0δt/2eH−(s− N Trhρ(0)(cid:16)eiH0δt/2eH+(s+ 2 ...ds± 1 )e−iH0δt/2(cid:17) ...(cid:16)e−iH0δt/2eH−(s− eλ A(cid:16)e−iH0δt/2e−iH1δte−iH0δt/2(cid:17)N i N )eiH0δt/2(cid:17) ...(cid:16)eiH0δt/2eH+(s+ N )e−iH0δt/2(cid:17)io. = lim λ→0 ∂ ∂λn 1 22N Z ds± 1 )eiH0δt/2(cid:17) (6) 1 , s± 2 , s± 3 ...s± The above equation is exact in the limit δt → 0. We refer to the integrand as an "Influence Functional" (IF), and denote it by I(s± N ). As discussed in Ref. [15], in standard nonequilibrium situations, even at zero temperature, bath cor- relations die exponentially, thus the IF can be truncated be- yond a memory time τc = Nsδt, corresponding to the time beyond which bath correlations may be controllably ignored. Here Ns is an integer, and the correlation time τc is dictated by the nonequilibrium situation, τc ∼ 1/∆µ. This argument implies the following (non-unique) breakup [15] I(s± 1 , s± ×Is(s± 2 , ...s± N −Ns+1, s± N ) ≃ I(s± 1 , s± 2 , ..., s± Ns N ), )Is(s± 2 , s± 3 , ..., s± Ns+1)... (7) N −Ns+2, ..., s± where each element in the product, besides the first one, is given by the ratio between truncated IF, 3 I(s± I(s± k , s± k , s± k+1, ..., s± k+1, ..., s± k+Ns−1) k+Ns−2) Is(sk, sk+1, ..., sk+Ns−1) = with , (8) (9) I(s± k , ..., s± k+Ns−1) = 1 22Ns Trhρ(0)G+(s+ k+Ns−1)...G+(s+ k )eiH0(k−1)δteλ Ae−iH0(k−1)δtG−(s− k )...G−(s− k+Ns−1)i. Here G+(s+ We now define the following multi-time object, k ) = (cid:16)eiH0δt/2eH+(s+ k )eiH0δt/2(cid:17) and G− = G† +. k+2, ..., s± I(s± 1 , s± k+Ns−1) ≡ 2 , ..., s± Ns )Is(s± 2 , s± 3 , ..., s± Ns+1)... R(s± k+1, s± Xs± 1 ,s± ×Is(s± 2 ,...,s± k , s± k k+1, ..., s± k+Ns−1), (10) and time-evolve it by multiplying it with the subsequent trun- cated IF, then summing over the intermediate variables, R(s± k+2, s± R(s± k+3, ..., s± k+1, s± k+Ns k+2, ..., s± ) = k+Ns−1)Is(s± k+1, s± k+2, ..., s± k+Ns ). Xs± k+1 Summation over the internal variables results in the time local expectation value, e.g., at tk we get (11) heλ A(tk)i = Xs± k+2−Ns ,...,s± k R(s± k+2−Ns , s± k+3−Ns , ..., s± k ).(12) This procedure is repeated for several values of small λ. Tak- ing the numerical derivative with respect to λ, the expectation value of the operator of interest, at a particular time, is re- trieved, h A(tk)i. The truncated influence functional in Eq. (9) is the core of our calculation. Since it includes only quadratic operators [18], it can be exactly calculated utilizing the trace formula for fermions [19], Tr[eM1 eM2...eMp ] = det[1 + em1em2...emp ]. (13) Here mp is a single particle operator corresponding to a c† i (cj) are quadratic operator Mp = Pi,j(mp)i,j c† i cj. fermionic creation (annihilation) operators. At zero temper- ature we can formally write Eq. (9) as I ∝ h0eM1eM2 ...eMp 0i = det[em1em2...emp ]occ, (14) where 0i is the initial (zero temperature) state of the total sys- tem and the determinant is carried over occupied states only. At finite temperatures Eq. (9) can be represented by I ∝ Tr[eM1 eM2...eMp (ρL ⊗ ρR ⊗ ρS(0))], (15) where ρα corresponds to the time-zero density matrix of the α = L, R fermion bath. ρS(0) denotes the subsystem initial density matrix. Assuming that these density operators can be written in an exponential form, eM , with M a quadratic oper- ator [18], application of the trace formula leads to I = Tr(cid:2)eM1eM2 ...eMp (ρL ⊗ ρR ⊗ ρS(0))(cid:3) = detn[IL − fL] ⊗ [IR − fR] ⊗ [IS − fS]. + em1em2...emp [fL ⊗ fR ⊗ fS]o. (16) The matrices Iα and IS are the identity matrices for the α space and for the subsystem, respectively. The functions fL and fR are the bands electrons' energy distribution, fα = [eβα(ǫ−µα) + 1]−1, with the chemical potential µα and tem- perature βα. The subsystem (initial distribution) fS may vary, depending on the particular problem. For example, for the Anderson model (Sec. III) we consider a dot initially empty. In what follows we apply the INFPI method on two quan- tum impurity models, of interest in the context of molecular electronics, the SIAM and the 2LAM, see Fig. 1, with mini- mal modifications to the simulation code. Since both models admit the form (1)-(2), we need only to separately construct the particular (single particle) noninteracting Hamiltonian H0 and the operator of choice A. With this in hand, we can read- ily calculate the truncated IF of Eq. (9) and the ratio in Eq. (8) using the trace formula. We then time evolve the multi-time R structure following Eq. (11). The time evolution of the operator of interest is acquired using Eq. (12). We note that this iterative algorithm can be feasibly adopted for simulat- ing other models, including correlated multi-site chains with quartic interactions. However, the present implementation is limited by efficiency to models with two correlated sites [20]. Before discussing numerical results, we point out the dif- ferent sources of errors in our calculations, and explain how to control and overcome them. There are three sources of systematic error within our approach. (i) Bath discretiza- tion error. The electronic reservoirs are explicitly included in our simulations, and we use bands extending from −D to D with a finite number of states per bath per spin (Ls). This stands in contrast to standard approaches where a wide-band limit is assumed and analytical expressions for the reservoirs Green's functions are adopted [11, 12, 14]. As we show be- low (see Fig. 6), by increasing the number of bath states Ls we can unequivocally reach convergence, typically employ- ing Ls ≥ 100 states. We also note that while it is sometimes advantageous to encompass the leads' effect into self ener- gies terms, complex dispersion relations can be easily handled within our method. (ii) Trotter error. The time discretization error, order of (U δt)2, originates from the approximate fac- torization of the total Hamiltonian into the non-commuting H0 (two-body) and H1 (many-body) terms, see text after Eq. (3). While for U → 0 and for small time-steps δt → 0 the decomposition is exactly satisfied, for large U one should go to a sufficiently small time-step in order to avoid signif- icant error buildup. Extrapolation to the limit δt → 0 is straightforward in principle [15]. (iii) Memory error. Our approach assumes that bath correlations exponentially decay resulting from the nonequilibrium condition ∆µ 6= 0. Based on this crucial element, the influence functional may be trun- cated to include only a finite number of fictitious spins Ns, where τc = Nsδt ∼ 1/∆µ for the population dynamics and τc =∼ 2/∆µ for the particle current (see Figs. 5 and 11). The total IF is retrieved by taking the limit Ns → N , (N = t/δt). However, one should be careful at this point: Increasing the memory length τc by adding more and more Trotter-terms into the truncated IF [Eq. (9)] results in a build-up of the time discretization error, unless the time-step is controlled concur- rently. Thus, one should carefully monitor both the time-step and the memory size for achieving reliable results. This chal- lenge is similar to that encountered in the standard QUAPI method [21, 22]. It should be noted that the convergence with respect to memory error is currently the most challenging aspect of the calculations with the INFPI approach. This limits us to rel- atively small values of the ratio of on-site correlation to hy- bridization strength. Future work will be devoted to algorith- mic optimization of the approach so that significantly larger memory times may be reached. 4 III. ANDERSON DOT A. Model and Observables The single impurity Anderson Model (SIAM) [4] is one of the most important models in condensed matter physics. While it was originally introduced to describe the behavior of magnetic impurities in non-magnetic hosts [23], it has more recently served as a generic model for understanding quantum transport in correlated nanoscale systems [24 -- 26]. In such cases, the impurity is hybridized with two reservoirs main- tained at different chemical potentials, leading to nonequilib- rium particle transport. The model includes a resonant level of energy ǫd, described by the creation operator d† σ (σ =↑, ↓ denotes the spin orientation) coupled to two fermionic leads (α = L, R) of different chemical potentials µα, but equal temperatures β−1. The Hamiltonian H = H0 + H1 [see Eqs. (1)-(2)] includes the following terms H0 = Xσ + Xα,k,σ (U/2 + ǫd)nd,σ + Xα,k,σ Vα,kc† α,k,σdσ + h.c. ǫkc† α,k,σcα,k,σ 1 2 (nd,↑ + nd,↓)(cid:3). (17) H1 = U(cid:2)nd,↑nd,↓ − Here c† α,k,σ (cα,k,σ) denotes the creation (annihilation) of an electron with momentum k and spin σ in the α lead, U stands for the onsite repulsion energy, and Vα,k are the impurity-α lead coupling elements. nd,σ = d† σdσ is the impurity occu- pation number operator. The shifted single-particle energies are denoted by Ed = ǫd + U/2. We also define Γ = Pα Γα, where Γα = πPk Vα,k2δ(ǫ − ǫk) is the hybridization en- ergy of the resonant level with the α metal. In what follows we focus on two observables: the time dependent occupation of the resonant level and the tunneling current through the dot. The population dynamics hnd,σ(t)i can be obtained by substi- tuting A = nd,σ (18) in Eq. (6). The current at the α contact hIα,σi may be resolved in two ways. We may either calculate the population depletion (or gain) in the α lead by defining A as the sum over the α- bath number operators, A = Xk c† α,k,σcα,k,σ. (19) The current itself is given by the time derivative of the A ex- dt h A(t)i. Alternatively, the current pectation value, hIα,σi = d at each end can be directly gathered by adopting the expres- sion A = −2ℑPk Vα,kc† α,k,σdσ, with ℑ as the imaginary part. In practice, we have employed the symmetric definition A = −ℑXk VL,kc† L,k,σdσ + ℑXk VR,kc† R,k,σdσ, (20) since its expectation value directly produces the symmetrized current B. Results 5 hIσi = hIL,σi − hIR,σi 2 . (21) 0.35 0.3 0.25 0.2 0.15 0.1 0.05 〉 σ d , n 〈 0 0 0.5 1 1.5 t Γ 2 2.5 3 FIG. 2: Population of the resonant level in the Anderson model U = 0 (thick full), U = 0.1 (dashed), U = 0.3 (dashed-dotted), U = 0.5 (dotted). The physical parameters of the model are D = 1, ∆µ = 0.4, Ed = 0.3, Γα=0.025, and βΓ = 10. The numerical parameters used are Ls = 240 lead states, τc = 3.2 with Ns = 4 and δt = 0.8. The U = 0 case is compared to the wide flat band limit, Eq. (22) (thin full line). We focus on the following set of parameters: a symmetri- cally distributed voltage bias between two leads with ∆µ = 0.4, flat bands centered at zero (the Fermi energy) with a cut- off at D = ±1, a resonant level energy Ed = 0.3, a hybridiza- tion strength Γα = 0.025 = πVα,k2ρα, with a constant den- sity of states ρα, onsite repulsion U/Γ ∼ 2 − 10, and a zero magnetic field. For these parameters a convergence analysis carried out in Ref. [15] has revealed that supplying Ls ≥ 100 states per spin per bath suffices for mimicking a continuous band structure. We have also found that for ∆µ = 0.4 a mem- ory size τc ∼ 1/∆µ ∼ 3.2 has lead to the convergence of the dot occupation when δt = 0.8 and Ns = 4, provided U Γ . 3 [15, 27]. As we show below, the simulation of the current turns out to be more challenging as a larger memory size is required for reaching converging behavior, τc ∼ 2/∆µ. Before presenting our results we clarify the initial condi- tions adopted here. As explained above, at t = 0 we set the reservoirs and the system in a factorized state: The dot is as- sumed to be empty, and the two reservoirs are decoupled, each maintained in a canonical state characterized by the Fermi- Dirac statistics. This scenario is distinct from the interaction and voltage quenches considered in Ref. [13]. Fig. 2 displays the time evolution of the dot occupancy hnd,σi with increasing on-site interaction for βΓ = 10, es- sentially reproducing the T = 0 data of Ref. [15]. Details about convergence issues, and a comparison to Monte-Carlo data were included in Ref. [15, 27]. In order to examine the effect of the bandwidth on the details of the dynamics the evo- lution of the noninteracting case (U = 0) is further compared to the wide flat band (WFB) behavior [16, 28], 0.25 β Γ=1, 2.5, 10 hnd,σ(t)i = Γ 2π Z ∞ −∞ dǫ[fL(ǫ) + fR(ǫ)] 〉 σ d , n 〈 0.2 0.15 0.1 0.05 0 0 0.5 1 × 1 + e−2Γt − 2e−Γt cos[(ǫ − ǫd)t] Γ2 + (ǫ − ǫd)2 . (22) We find that the D/Γ = 20 case inspected here deviates from the WFB result in both the short time behavior and the long time characteristics. However, general trends are maintained. We have also verified (data not shown) that the INFPI results approach the WFB limit when increasing the bandwidth, for U = 0. The effect of the temperature at different interaction strengths is analyzed in Fig. 3, adopting βΓ = 0.1 − 10. For U = 0.1, a comparison between the long time INFPI limit and the mean-field theory [29, 30], 〉 σ d , n 〈 0.2 0.1 10−1 1/βΓ 100 1.5 t Γ 2 2.5 3 hnd,σ(t → ∞)i = Γ 2π Z ∞ −∞ fL(ǫ) + fR(ǫ) (ǫ − ǫd − U hnd,−σi)2 + Γ2 dǫ, (23) FIG. 3: Population of the resonant level in the Anderson model U = 0 (thick full), U = 0.1 (dashed), U = 0.3 (dashed-dotted) at various temperatures, βΓ = 1, 2.5 and 10; top to bottom. Other parameters are the same as in Fig. 2. The inset compares the long-time U = 0.1 behavior ((cid:3)) to mean-field results (◦) obtained from Eq. (23). reveals a good agreement (inset, Fig 3). For the same set of parameters we calculate next the sym- metric tunneling current hIσ(t)i through the SIAM. Simula- tion results for U = 0 and U = 0.1 are presented in Fig. 4. The current enhancement with U can be reasoned by noting Γ / 〉 ) t ( σ I 〈 0.2 0.15 0.1 0.05 0 0 1 t Γ 2 3 FIG. 4: Current through the Anderson dot, U = 0 (small dots), U = 0.1 (large dots), Ed = 0.3, Γ = 0.05, βΓ = 10. The U = 0 case is compared to the WFB limit obtained from Eq. (24) (thin full line). The numeric parameters are δt = 1.6, Ns = 5 and Ls = 120. Γ / 〉 ) t ( σ I 〈 0.2 0.1 0 Γ / 〉 ) t ( σ I 〈 0.1 0.08 2 2.5 t Γ 3 1 2 t Γ Ns=2 Ns=3 Ns=4 Ns=5 Ns=6 Ns=7 3 6 The convergence of the tunneling current with respect to the number of bath states, time-step, and memory size has been carefully tested. In particular, Fig. 5 demonstrates the behav- ior of the current with increasing memory size τc = Nsδt, showing that convergence is reached when τc ∼ 7 − 8. We note that a significantly shorter memory size (τc ∼ 3 − 4) has been required for converging the dot occupancy [15]. This difference could be reasoned as follows. Since the tunnel- ing current is calculated at a specific contact, the memory size that should be accounted for inside the influence functional (9) should roughly scales with the bias difference at that contact. Thus, τ −1 In contrast, the population dynamics is sensitive to the full bias drop ∆µ, therefore bath correla- tions can be safely truncated beyond τc ∼ 1/∆µ. In Fig. 6 we present the behavior of the current upon increasing the number of bath states. It is interesting to note that the choice Ls = 40 states per spin per bath already reproduces results in a good agreement with the Ls → ∞ limit. Thus, the finite temperature algorithm adopted here [Eq. (16)], is superior to the strictly zero temperature algorithm of Ref. [15], even when applied to relatively low temperatures. c ∼ ∆µ/2. It is also of interest to examine the temperature dependence of the asymptotic electric current. This information is con- veyed in Fig. 7 for zero and finite U using data at tΓ = 5. Results are also compared to the mean-field wide-band ap- proximation [29, 30], hIss,σi = 1 2π Z ∞ −∞ Γ2[fL(ǫ) − fR(ǫ)] (ǫ − ǫd − U hnd,−σi)2 + Γ2 dǫ. (26) FIG. 5: Convergence of the current hIσ(t)i through the Anderson dot with increasing memory size τc = Nsδt. Ed = 0.3, U = 0.1, Γ = 0.05, βΓ = 10. The numerical parameters are Ls = 120 states and δt = 1.6. Ns = 2 (◦), Ns = 3 (⋄), Ns = 4 (+), Ns = 5 (x), Ns = 6 ((cid:3)), Ns = 7 (dotted line). Inset: zooming over the long-time values. Deviations from this result, for U = 0, indicate on the de- parture from the WFB approximation. In the large bias limit examined here (∆µ/Γ = 8) the current saturates at low tem- peratures, βΓ < 2.5, in agreement with the results of Ref. [13]. that the parameter Ed = ǫd + U/2 is fixed, thus the actual dot energy is down-shifted when increasing the interaction U . We again compare the noninteracting behavior with the dynamics in the WFB limit [16], hIL,σ(t)i = hIL,σ(t → ∞)i − Γe−Γt 1 2π Z ∞ −∞ dǫ 1 (ǫ − ǫd)2 + Γ2 0.2 0.15 0.1 0.05 Γ 〉 / ) t ( σ I 〈 Γ 〉 / ) 5 . 2 ( σ I 〈 0.102 0.1 0.098 0 100 200 Ls 0.1 0.095 Γ 〉 / ) t ( σ I 〈 × nΓe−Γt[fL(ǫ) + fR(ǫ)] − Γ cos[(ǫ − ǫd)t][2fR(ǫ) + 1] − (ǫ − ǫd) sin[(ǫ − ǫd)t][2fL(ǫ) − 1]o, (24) with the asymptotic value hIL,σ(t → ∞)i = Γ2 2π Z ∞ −∞ fL(ǫ) − fR(ǫ) (ǫ − ǫd)2 + Γ2 dǫ, (25) and hIR,σ(t)i = −hIL,σ(−∆µ, t)i. Good agreement is ob- served in the long time limit. 0 0 1 2 1.5 3 tΓ tΓ 2 4 2.5 5 6 FIG. 6: Convergence of the current hIσ(t)i through the Anderson dot with increasing number of bath states Ls. Ed=0.3, U = 0.1, βΓ = 10, Γ = 0.05, δt = 1.6, Ns = 5. Ls=40 (heavy full), 80 (dashed), 120 (dotted), 160 (dashed-dotted), and 240 (light full). The data lines for Ls ≥ 80 are almost overlapping, see also the bottom inset. Top inset: Data as a function of Ls at Γt = 2.5. Γ / 〉 σ , s s I 〈 0.15 0.1 0.05 U=0.1 U=0 100 101 β Γ FIG. 7: Steady-state current hIss,σi = hIσ(t → ∞)i through the Anderson dot, U = 0.1 (◦) and U = 0 ((cid:3)), Ed = 0.3, Γ = 0.05. The full lines are the results of a mean-field calculation, Eq. (26). The numerical parameters are Ls = 120, Ns = 5 and δt = 1.6. IV. SPINLESS TWO-LEVEL ANDERSON MODEL A. Model and observables The spinless two-level Anderson model (2LAM) and its extensions have been extensively studied in the context of molecular electronics, for exploring various effects in molec- ular conduction: vibrational effects [31], thermoelectricity in molecular junctions [32, 33], radiation field-induced pro- cesses [34], and Coulomb interaction effects [5]. More re- cently, the mechanism of population inversion [5] has been ex- plored using the asymmetric interacting 2LAM, where the two levels differently couple to the leads. Furthermore, by includ- ing a left-right asymmetry in the dot-leads coupling, the mech- anism of the transmission phase lapses in quantum dots [35] has been resolved within mean-field theories [36, 37], Monte- Carlo techniques [7], and functional and numerical renormal- ization group approaches [38, 39]. The 2LAM model incor- porates an impurity with two electronic levels ǫ1 < ǫ2, de- m, (m = 1, 2), coupled to scribed by the creation operator d† two metal leads (α = L, R) of different chemical potentials. The Hamiltonian H = H0 + H1 includes the following terms H0 = (ǫ1 + U/2)n1 + (ǫ2 + U/2)n2 +Xα,k α,kcα,k ǫkc† + Xα,k,m=1,2 Vα,k,mc† α,kdm + h.c. H1 = U [n1n2 − 1 2 (n1 + n2)]. (27) Here c† k,α denotes the creation (annihilation) of an electron with momentum k in the α lead, nm = d† mdm is the number operator for the impurity levels, and U is the charging energy. We also define the hybridization strength Γm ≡ ΓL,m + ΓR,m with Γα,m = πPk Vα,k,m2δ(ǫ − ǫk) and use flat bands ex- tending symmetrically between ±D. The dot shifted energies are denoted by Em = ǫm +U/2. This model is closely related to the interacting Anderson model analyzed in Sec. III, tak- ing the two states here to emulate different spin orientations. 7 However, here (i) only a single spin specie is considered, al- lowing for interference effects between the two transmission pathways, (ii) the dot levels are nondegenerate, and (iii) the impurity states differently couple to the leads, typically as- suming that the HOMO level, a deep molecular orbital, is cou- pled more weakly to the leads. The population dynamics of each electronic level hnm(t)i and the current through the 2LAM are calculated numerically using the INFPI method, as prescribed in Sec. II. The current plotted will be the total symmetrized current flowing through the system, obtained by defining the operator of interest A as A = −ℑXk,m L,kdm + ℑXk,m R,k,mdm. (28) VR,k,mc† VL,k,mc† 1 0.8 0.6 0.4 0.2 U=0 U=0.1 U=0.2 〈 n (t)〉 1 〈 n (t)〉 2 〉 ) t ( n 〈 m 0 0 1 2 Γ 1 t 3 4 FIG. 8: Population of the 2LAM electronic levels with increasing U term. U = 0 (full), U = 0.1 (dashed), U = 0.2 (dashed-dotted), E1 = −0.1, E2 = 0.3, Γ1,α = 0.025, Γ2,α = 0.05, β = 200. The numerical parameters are δt = 0.8, Ns = 5 and Ls = 120. 〉 ) ∞ → t ( n 〈 1 〉 ) ∞ → t ( n 〈 2 0.8 0.6 0.1 0 0 0.1 U 0.2 FIG. 9: Steady-state population of the 2LAM electronic levels: Comparison between the INFPI asymptotic data, extracted from Fig. 8 (◦), and mean-field results ((cid:3)). B. Results We focus on the symmetric (L − R) case, and use the following set of parameters: ΓL,1=ΓR,1=0.025 and  the level's population satisfy hnm(t → ∞)i = Γm 2π Z ∞ −∞ fL(ǫ) + fR(ǫ) (ǫ − ǫm − U hn ¯mi)2 + Γ2 m 8 dǫ, (29) 〉 ) t ( I 〈 0.025 0.02 0.015 0.01 0.005 U=0, β=200 U=0.1, β=200 U=0, β=20 U=0.1, β=20 0 0 0.5 1 1.5 Γ 1 t 2 2.5 3 FIG. 10: Current dynamics in the 2LAM with increasing U term. U = 0 (full), U = 0.1 (dashed) and β = 200 (heavy) β = 20 (light). E1 = −0.1, E2 = 0.3, Γ1,α = 0.025, Γ2,α = 0.05. The numerical parameters are δt = 0.8, Ns = 7 and Ls = 120. 0.02 0.019 0.018 δ t=0.8 δ t=1.6 〉 s s I 〈 0.017 0.016 0 2 4 6 τ c 8 10 12 FIG. 11: Convergence of the steady state current with increasing memory size τc using different time steps δt = 1.6 (empty sym- bols) and δt=0.8 (full symbols) for β = 200 (circle) and β = 20 (square). Parameters are the same as in Fig. 10. ΓL,2=ΓR,2=0.05. The bias (∆µ = 0.4) will be symmetri- cally distributed between the leads, assuming flat bands cen- tered around zero with a cutoff at D = ±1. The interaction strength will be limited to U/Γ1 . 4 and the temperature will be varied between βΓ1 ∼ 1 − 10. Fig. 8 displays the levels' occupation as a function of time, for several interac- tion values, U =0, 0.1 and 0.2. We find that the HOMO pop- ulation hn1(t)i is increasing with U . In conjunction, due to the increased importance of repulsion effects on the dot, the LUMO population hn2(t)i depletes with U . The convergence of the data with respect to the number of bath states Ls, time step δt, and memory size τc = Nsδt has been verified. A comparison to the WFB limit for the noninteracting case re- veals dynamical properties similar to those identified in Fig. 2. Steady-state mean-field results are obtained by using ex- pressions analogous to Eqs. (23) and (26) [5]. For example, where ¯m = 2, 1 if m = 1, 2. In Fig. 9 we plot the asymptotic population dynamics, using the data from Fig. 8, and com- pare those values to mean-field results. As expected, the dis- crepancy between these two calculations increases for larger U . Deviations at U = 0 probably stem from the fact that the INFPI method assumes finite bands of D = ±1, while mean- field results are calculated for WFB leads. We examine the temporal behavior of the current in Fig. 10, varying the temperature and the many-body interaction strength. For the present set of parameters we conclude that the current decreases for large U , and that the temporal os- cillations are washed out with increasing temperature. Fi- nally, we use this data as highlighted in Fig. 11 to expose a subtle convergence issue: the counteracting effect of differ- ent sources of errors, the time-step and the memory-size, and the challenge to overcome them both together. Employing the same set of parameters as in Fig. 10, we extract the steady- state value for the current, and display it as a function of τc, at two different temperatures, using two different time-steps. We find that for 4.5 < τc < 8 the steady state results are almost fixed, fluctuating by only 1%. However, for τc > 8 a departure from the apparent steady state occurs, becoming larger for larger τc. This behavior is caused by buildup of the Trotter factorization error within the truncated IF, Eq. (9). As expected, the error increases at larger U . To control this error, at large τc a shorted time-step should be selected. Future work will be dedicated to the strong coupling limit, Γm > ǫ2 − ǫ1, for analyzing the charge oscillation effect [5]. The asymmetric L − R setup is also of great importance, for studying the phase lapses mechanism beyond the mean-field approximation, at strong driving [36]. V. SUMMARY We have employed here the INFPI method [15] for study- ing the population dynamics and the current behavior of two eminent molecular junction models: the single impurity An- derson model, and the 2-level Anderson dot. Considering voltage-biased junctions, the effect of the intra-dot electron- electron repulsion energy and the temperature were jointly analyzed. We have compared our results to mean-field calcu- lations, showing an increased discrepancy when many-body interactions are enhanced. A careful convergence analysis has been performed, demonstrating how to adequately converge the INFPI simulations. The INFPI method has been described here in connection with molecular transport junctions. We expect this flexi- ble tool to become useful for studying other-related impurity models, and for exploring nonlinear thermoelectric effects in molecular junctions [33]. In particular, future work will be fo- cused on simulating the dynamics of extended junctions, e.g., a multi-site chain, and on extending the method to include vi- brational effects [31] in a non perturbative manner. like to acknowledge the NSF for financial support. 9 Acknowledgments DS acknowledges support from NSERC. AJM was sup- ported by NSF under Grant No. DMR-1006282. DRR would [1] N. J. Tao, Nature Nanotech. 1, 173 (2006). [2] A. W. Ghosh, P. S. Damle, S. Datta, and A. Nitzan, MRS Bul- letin, 29, 391 (2004). N. Makri and D. E. Makarov, J. Chem. Phys. 102, 4611 (1995); N. Makri, J. Math. Phys. 36, 2430 (1995). [22] J. Eckel, S. Weiss, and M. Thorwart, Eur. Phys. J. B 53, 91 [3] R. Bulla, T. A. Costi, and T. Pruschke, Rev. Mod. Phys. 80, 395 (2006). (2008). [4] P. W. Anderson, Phys. Rev. 124, 41 (1961). [5] M. Sindel, A. Silva, Y. Oreg, and J. von Delft, Phys. Rev. B 72, 125316 (2005). [6] V. Kashcheyevs, A. Schiller, A. Aharony, and O. Entin- Wohlman, Phys. Rev. B 75, 115313 (2007). [7] X. Wang and A. J. Millis, Phys. Rev. B 81, 045106 (2010). [8] R. M. Konik, H. Saleur, and A. Ludwig, Phys. Rev. B 66, 125304 (2002). [9] Y. Meir, N. S. Wingreen, and P. A. Lee, Phys. Rev. Lett. 70, 2601 (1993). [10] A. Rosch, J. Kroha, and P. Wolfle, Phys. Rev. Lett. 87, 156802 (2001); C.-H. Chung, K. Le Hur, M. Vojta, and P. Wolfle, Phys. Rev. Lett. 102, 216803 (2009); H. Schoeller and F. Reining- haus, Phys. Rev. B 80, 045117 (2009). [11] P. Werner, T. Oka, and A. J. Millis, Phys. Rev. B 79, 035320 (2009); M. Schiro and M. Fabrizio, Phys. Rev. B 79, 153302 (2009). [12] P. Werner, A. Comanac, L. de Medici, M. Troyer, and A. J. Millis, Phys. Rev. Lett. 97, 076405 (2006); E. Gull, P. Werner, A. Millis, and M. Troyer, Phys. Rev. B 76, 235123 (2007); P. Werner, T. Oka, and A. J. Millis, Phys. Rev. B 79, 035320 (2009). [13] P. Werner, T. Oka, M. Eckstein, and A. J. Millis, Phys. Rev. B 81, 035108 (2010). [14] S. Weiss, J. Eckel, M. Thorwart, and R. Egger, Phys. Rev. B 77, 195316 (2008); J. Eckel, F. Heidrich-Meisner, S. G. Jakobs, M. Thorwart, M. Pletyukhov, and R. Egger, New J. Phys. 12, 043042 (2010). [15] D. Segal, A. J. Millis, and D. R. Reichman, Phys. Rev. B 82, 205323 (2010). [23] A. C. Hewson, The Kondo Problem to Heavy Fermions, (Cam- bridge University Press, Cambridge, England, 1993). [24] I. L. Aleiner, P. W. Brouwer, and L. I. Glazman, Phys. Rep. 385, 309 (2002). [25] D. Natelson, Nature Nanotech. 4, 406 (2009). [26] J. Paaske, A. Rosch, and P. Wolfle, Phys. Rev. B 69, 155330 (2004); J. Paaske, A. Rosch, J. Kroha, and P. Wolfle, Phys. Rev. B 70, 155301 (2004). [27] A recent analysis (G. Cohen and E. Rabani, to be published) shows that for cases such as U/Γ > 3 presented here, memory times can be significantly longer than the truncation times used in this work. Thus while INFPI is in principle numerically ex- act, one must cope with the numerical expense of long memory times at large U. While future work will be devoted to algorith- mic improvements that potentially will allow for the attainment of such memory times within INFPI, we are currently limited in this regard. Thus, results presented in Figs. 2, 3, 8 and 9 have a small systematic error when U/Γ > 3. For crude reference, the maximum errors in the population values for U/Γ = 6 and ∆µ = 0.4 found in Ref. [15] with memory time truncations similar to those used here are of the order of 5%. [28] For simplicity, the analytical expressions (22)-(26) and (29) are all written assuming a symmetric junction, ΓL = ΓR. [29] A. Komnik and A. O. Gogolin, Phys. Rev. B 69, 153102 (2004). [30] B. Horvath, B. Lazarovits, O. Sauret, and G. Zarand, Phys. Rev. B 77, 113108 2008. [31] M. Galperin, M. A. Ratner, and A. Nitzan, J. Phys.: Condens. Matter 19, 103201 (2007). [32] M. Paulsson and S. Datta, Phys. Rev. B 67, 241403R (2003). [33] P. Reddy, S.-Y. Jang, R. A. Segalman, and A. Mujamdar, Sci- ence 315, 1568 (2007). [16] T. L. Schmidt, P. Werner, L. Muhlbacher and A. Komnik, Phys. [34] M. Galperin and A. Nitzan, Phys. Rev. Lett. 95, 206802 (2005); Rev. B 78, 235110 (2008). [17] J. E. Hirsch, Phys. Rev. B 28, 4059 (1983). [18] The time-zero density matrix of the subsystem is assumed to have a diagonal form, and it is represented by a canonical-type distribution. The reservoirs are assumed to be prepared in a thermal-canonical state. [19] I. Klich, in "Quantum Noise in Mesoscopic Systems", edited by Yu. V. Nazarov and Ya. M. Blanter (Kluwer, 2003). [20] Technically, one could feasibly generalize the method to de- scribe transport through a chain of Anderson dots. Practi- cally, reaching convergence becomes very demanding in such a model. [21] N. Makri and D. E. Makarov, J. Chem. Phys. 102, 4600 (1995); J. Chem. Phys. 124, 234709 (2006). [35] A. Yacoby, M. Heiblum, D. Mahalu, and H. Shtrikman, Phys. Rev. Lett. 74, 4047 (1995); R. Schuster, E. Buks, M. Heiblum, D. Mahalu, V. Umansky, and H. Shtrikman, Nature 385, 417 (1997); M. Avinun-Kalish, M. Heiblum, O. Zarchin, D. Ma- halu, and V. Umansky, Nature 436, 529 (2005). [36] D. I. Golosov and Y. Gefen, Phys. Rev. B 74, 205316 (2006). [37] M. Goldstein and R. Berkovits, New J. of Phys. 9, 118 (2007). [38] V. Meden and F. Marquardt, Phys. Rev. Lett. 96, 146801 (2006). [39] C. Karrasch, T. Hecht, A. Weichselbaum, J. von Delft, Y. Oreg, and V. Meden, New J. of Phys. 9, 123 (2007).